Reverse-Osmosis Desalination of Seawater Powered by Photovoltaics Without Batteries

by

A. Murray Thomson

A Doctoral Thesis

Submitted in partial fulfilment of the requirements

for the award of

Doctor of Philosophy of Loughborough University

30th June 2003

 by A. Murray Thomson 2003

i
Certificate of originality

ii
 Abstract

The design, construction and testing of a photovoltaic-powered reverse-osmosis (PV-RO)

desalination system is presented. The system operates from seawater and requires no
batteries, since the rate of production of freshwater varies throughout the day according
to the available solar power. Initial testing of the system, with the modest solar resource
available in the UK, provided freshwater at approximately 1.5 m3/day. Nearer to the
equator and with a PV array of only 2.4 kWP, a software model of the system predicts
production of over 3 m3/day throughout the year. The system employs a Clark pump
brine-stream energy recovery mechanism and this, coupled with variable water recovery
ratio, achieves a specific energy consumption of less than 4 kWh/m3 over a broad range
of operation. Standard industrial inverters, motors and pumps are employed and provide
good energy and cost efficiency. Maximum power point tracking (MPPT) for the
photovoltaic array is provided by a novel control algorithm, developed by the author.
Instrumentation and data acquisition of the hardware test rig using LabVIEW is
described. Testing and modelling of the system components in MATLAB-Simulink is
presented, together with a discussion of the full system modelling and design procedure,
in which the aim was to minimise the cost of water. This led to a capital cost estimate of
£23,055 including the PV array, and an overall cost of water, including full maintenance,
of £2.00 per m3.

iii
 Acknowledgments

I wish to thank Tony Marmont for initiating and supporting CREST, the renewable
energy research group at Loughborough University.

Also, I thank David Infield, Director of CREST, for giving me the opportunity to work in
this area and for his guidance and encouragement throughout.

I thank Marcos Miranda for working with me during the earlier stages of this project.
Marcos is now completing his thesis on wind-powered reverse osmosis. The reverse-
osmosis test-rig at CREST is shared, and we both contributed to its development, as duly
acknowledged in the chapters of this thesis. I have not seen Marcos’s thesis but I imagine
it will form a useful complement to this one for anyone seeking to further develop these
systems.

I gratefully acknowledge financial support from the DTI through ETSU grant
S/P2/00305.

Murray Thomson

June 2003

iv
 Contents

Certificate of originality......................................................................................................ii
Abstract ..............................................................................................................................iii
Acknowledgments..............................................................................................................iv
Contents ..............................................................................................................................v
List of figures .....................................................................................................................xi
List of tables.....................................................................................................................xiv
The new PV-RO test rig....................................................................................................xv
Chapter 1 Overview.........................................................................................................1
1.1 Desalination and Renewable Energy ..................................................................1
1.2 Reverse Osmosis (RO)........................................................................................3
1.3 Photovoltaics (PV) and Batteries ........................................................................4
1.4 Proposed PV-RO system.....................................................................................5
1.5 Modelling............................................................................................................7
1.6 System hardware testing .....................................................................................7
1.7 Performance figures ............................................................................................7
1.7.1 Predictions from software model ................................................................7
1.7.2 Actual measurements ..................................................................................8
1.8 Global relevance .................................................................................................8
Chapter 2 Technology and literature review .................................................................10
2.1 Desalination ......................................................................................................10
2.1.1 A growth industry .....................................................................................10
2.1.2 Seawater versus brackish water ................................................................11
2.1.3 Product water specification.......................................................................11
2.2 Desalination techniques ....................................................................................13
2.2.1 Distillation.................................................................................................13
2.2.2 Multi-Effect Distillation (MED) ...............................................................13
2.2.3 Multi-Stage Flash (MSF) ..........................................................................14
2.2.4 Vapour Compression ................................................................................14
2.2.5 Reverse Osmosis (RO)..............................................................................15
2.2.6 Electrodialysis...........................................................................................15
2.3 Energy for desalination .....................................................................................15
2.3.1 Theoretical ................................................................................................15
2.3.2 In practice..................................................................................................17
2.3.3 Energy sources ..........................................................................................18
2.4 Renewable energy for desalination ...................................................................19
2.4.1 Intermittency .............................................................................................19
2.4.2 Compatibility ............................................................................................20
2.4.3 Simple solar stills ......................................................................................20
2.4.4 Indirect solar thermal desalination............................................................21
2.4.5 Membrane Distillation ..............................................................................21
2.4.6 Freezing.....................................................................................................22
2.4.7 Wind-powered Reverse Osmosis ..............................................................22
2.4.8 Other studies .............................................................................................23
2.5 Photovoltaics (PV) ............................................................................................23

v
 2.5.1 Cost ...........................................................................................................24
2.5.2 Materials ...................................................................................................25
2.5.3 Watts-peak ................................................................................................25
2.5.4 Maximum Power Point Tracking (MPPT)................................................25
2.6 Batteries ............................................................................................................27
2.7 Reverse Osmosis (RO)......................................................................................29
2.7.1 Membrane materials..................................................................................29
2.7.2 Hollow-fibre permeators ...........................................................................29
2.7.3 Spiral-wound membrane elements............................................................30
2.7.4 Fouling, scaling and membrane life..........................................................31
2.7.5 Temperature effect ....................................................................................31
2.7.6 High-rejection and high-flow membranes ................................................31
2.8 Brine-stream energy recovery...........................................................................32
2.8.1 Pelton wheel..............................................................................................32
2.8.2 Hydraulic Turbo Booster ..........................................................................33
2.8.3 DWEER Work Exchanger ........................................................................33
2.8.4 ERI’s Pressure Exchanger.........................................................................34
2.8.5 Relative efficiencies..................................................................................35
2.8.6 Vari-RO.....................................................................................................35
2.8.7 Energy recovery in small-scale RO ..........................................................36
2.8.8 Energy recovery pumps ............................................................................36
2.8.9 Hydraulic motor ........................................................................................37
2.8.10 Clark pump................................................................................................37
2.9 Photovoltaic-powered Reverse Osmosis (PV-RO) ...........................................39
2.9.1 Brackish-water PV-RO systems ...............................................................39
2.9.2 Seawater PV-RO demonstration systems .................................................41
2.9.3 Batteries or not ..........................................................................................42
2.9.4 Applications ..............................................................................................43
Chapter 3 Initial test rig hardware and results ...............................................................44
3.1 Operational results ............................................................................................45
3.2 Sankey diagram.................................................................................................46
3.3 Danfoss hydraulic motor...................................................................................47
Chapter 4 Instrumentation and data-acquisition system................................................49
4.1 LabVIEW ..........................................................................................................49
4.2 Data acquisition hardware.................................................................................50
4.3 Acquiring data into LabVIEW..........................................................................50
4.4 Signal processing ..............................................................................................51
4.5 Pulse inputs .......................................................................................................51
4.6 Data logging......................................................................................................52
4.7 Control ..............................................................................................................52
4.8 User interface ....................................................................................................52
4.9 Flow measurement ............................................................................................53
4.9.1 Turbine flow meters ..................................................................................53
4.9.2 Oval-gear flow meters...............................................................................53
4.10 Pressure sensors ................................................................................................53
4.10.1 Medium-pressure sensor ...........................................................................53
4.10.2 High-pressure sensors ...............................................................................54
4.11 Concentration measurement..............................................................................54
4.12 Temperature measurement................................................................................55
4.13 Temperature control..........................................................................................55

vi
 4.14 Torque ...............................................................................................................56
4.15 Speed.................................................................................................................56
4.16 Voltage, current and electrical power ...............................................................56
4.17 Irradiance ..........................................................................................................56
Chapter 5 Component options, testing and modelling...................................................57
5.1 Modelling strategy – MATLAB-Simulink .......................................................58
5.2 Clark pump........................................................................................................60
5.2.1 Theory .......................................................................................................61
5.2.2 Testing.......................................................................................................67
5.2.3 Modelling..................................................................................................70
5.2.4 Efficiencies and conclusions.....................................................................71
5.3 Plunger pump ....................................................................................................72
5.3.1 Modelling..................................................................................................73
5.3.2 Modelled efficiency ..................................................................................74
5.4 Moineau pump ..................................................................................................75
5.4.1 Modelling data ..........................................................................................76
5.4.2 Simulink model .........................................................................................77
5.4.3 Modelled efficiency ..................................................................................78
5.5 Motors and inverters .........................................................................................79
5.6 Induction motors ...............................................................................................80
5.6.1 Modelling data ..........................................................................................80
5.6.2 Simulink model .........................................................................................82
5.6.3 Induction motor efficiencies .....................................................................83
5.6.4 Value of motor efficiency .........................................................................83
5.6.5 Motor selection .........................................................................................84
5.6.6 Test rig motors ..........................................................................................85
5.7 Standard industrial variable-speed drive inverters............................................85
5.7.1 Modelling..................................................................................................86
5.7.2 Parameter adjustments ..............................................................................88
5.8 Solar irradiance .................................................................................................89
5.8.1 Solar-trajectory tracking ...........................................................................90
5.8.2 Hourly irradiance ......................................................................................91
5.8.3 Average daily irradiance ...........................................................................93
5.8.4 Average monthly irradiance......................................................................95
5.9 Ambient temperature ........................................................................................96
5.9.1 Hourly ambient temperatures....................................................................96
5.9.2 Average monthly ambient temperatures ...................................................97
5.10 PV array ............................................................................................................98
5.10.1 Two-diode model ......................................................................................98
5.10.2 Simulink model .........................................................................................99
5.10.3 Electrical power available from PV ........................................................101
5.11 Seawater ..........................................................................................................106
5.11.1 Concentration: ppm versus mg/L............................................................106
5.11.2 Salinity versus TDS ................................................................................106
5.11.3 Typical concentrations of seawater.........................................................107
5.11.4 Composition of seawater.........................................................................107
5.11.5 Osmotic pressure.....................................................................................108
5.11.6 Straight NaCl solution.............................................................................109
5.11.7 Summary .................................................................................................109
5.12 Feed-water temperature...................................................................................110

vii
 5.13 Reverse osmosis membrane elements.............................................................111
5.13.1 Model structure .......................................................................................111
5.13.2 Calculations.............................................................................................113
5.14 Product tank ....................................................................................................114
Chapter 6 System modelling, optimisation and performance predictions...................116
6.1 System design strategy....................................................................................116
6.1.1 System model structure...........................................................................116
6.1.2 Modifying the system model ..................................................................118
6.1.3 Capital cost modelling ............................................................................118
6.1.4 Annual performance figures ...................................................................120
6.2 Review of configuration ideas ........................................................................123
6.2.1 Single feed pump ....................................................................................123
6.2.2 Delayed injection – single motor ............................................................124
6.2.3 Two motor – variable recovery ratio.......................................................126
6.2.4 Solar-thermal feed-water pre-heating .....................................................127
6.3 RO and PV array sizing ..................................................................................129
6.3.1 RO array sizing .......................................................................................129
6.3.2 PV array sizing........................................................................................130
6.3.3 Life-cycle costing....................................................................................133
6.4 Completed design............................................................................................134
6.5 Performance predictions .................................................................................135
6.5.1 Product water flow ..................................................................................136
6.5.2 Product water quality ..............................................................................141
6.5.3 Predicted energy flows – Sankey diagram..............................................142
6.5.4 Power usage – average day .....................................................................144
6.5.5 Specific energy........................................................................................145
6.6 Manufacturer’s cost analysis...........................................................................146
Chapter 7 New test rig hardware .................................................................................147
Chapter 8 Recovery-ratio control algorithm................................................................149
8.1 System characterisation ..................................................................................149
8.2 Control strategy...............................................................................................150
8.3 Measured performance....................................................................................152
Chapter 9 Maximum power point tracking (MPPT) algorithm ...................................153
9.1 Common MPPT algorithms ............................................................................154
9.1.1 Constant voltage......................................................................................154
9.1.2 Constant voltage with temperature compensation ..................................155
9.1.3 Percentage of open-circuit voltage..........................................................155
9.1.4 Perturb and observe (hill climbing) ........................................................156
9.1.5 Incremental conductance.........................................................................156
9.2 Standard industrial variable-speed drive inverters..........................................156
9.3 New MPPT algorithm .....................................................................................157
9.4 Basic implementation......................................................................................158
9.5 Enhanced implementation...............................................................................160
9.6 Two inverters ..................................................................................................162
9.7 In practice........................................................................................................163
9.8 Discussion .......................................................................................................165
9.9 Pros and cons ..................................................................................................167
Chapter 10 Batteryless PV-RO demonstration system measured performance ........168
10.1 Irradiance and product flow ............................................................................168
10.2 Product flow versus irradiance........................................................................169

viii
 10.3 Product flow versus DC power .......................................................................170
10.4 Specific energy consumption (kWh/m3) .........................................................171
10.5 Water recovery ratio .......................................................................................172
10.6 Product concentration .....................................................................................173
10.7 Pressures .........................................................................................................176
10.8 Pump efficiencies............................................................................................177
10.9 Inverter and motor efficiencies .......................................................................178
Chapter 11 Practical considerations...........................................................................180
11.1 Clark pump......................................................................................................180
11.2 Membrane fouling...........................................................................................180
11.3 Feed water intake arrangement – beach wells ................................................181
11.4 Membrane flushing .........................................................................................181
11.5 Product tank chlorination ................................................................................182
11.6 Osmotic suck-back..........................................................................................182
Chapter 12 Conclusions.............................................................................................183
12.1 Renewable-energy powered desalination .......................................................183
12.2 Batteryless PV-RO..........................................................................................183
12.3 Variable-flow ..................................................................................................184
12.4 Product concentration .....................................................................................184
12.5 Energy recovery – The Clark pump................................................................185
12.6 Variable recovery ratio ...................................................................................185
12.7 Maximum power point tracking and industrial inverters................................185
12.8 Batteryless PV-RO hardware demonstration ..................................................186
12.9 System complexity and reliability ..................................................................186
12.10 Instrumentation ...............................................................................................186
12.11 Software modelling.........................................................................................187
12.12 Market identification.......................................................................................187
Published Papers .............................................................................................................188
Desalination ................................................................................................................188
Others ........................................................................................................................188
.

References.......................................................................................................................189
Appendix A Details of testing carried out using the initial test rig .............................196
A.1 Instrumentation and data-acquisition ..............................................................196
A.2 Procedure and results ......................................................................................197
A.2.1 Results data summary table ....................................................................208
A.3 Power flow and efficiency calculations for Sankey diagram..........................208
Appendix B Sensor details and calibration .................................................................213
B.1 Turbine flow meters ........................................................................................213
B.1.1 Product-flow turbine flow meter.............................................................213
B.1.2 Feed flow ................................................................................................214
B.2 Oval-gear flow meter calibration sheets .........................................................216
B.3 Medium-pressure sensor specification and calibration sheet..........................219
B.4 Concentration from conductivity ....................................................................220
B.5 Torque load-cell and calibration .....................................................................224
B.5.1 Load-cell datasheet .................................................................................224
B.5.2 Torque measurement calibration.............................................................225
B.6 Irradiance pyranometer calibration certificate ................................................227
Appendix C Clark pump details, test results and analysis ..........................................228
C.1 Specification ...................................................................................................228
C.2 Test procedure.................................................................................................228

ix
 C.3 Measured data .................................................................................................230
C.4 Analysis...........................................................................................................231
C.5 Graphs and discussion.....................................................................................234
C.6 Modelling........................................................................................................236
C.7 Testing of the Simulink model........................................................................238
Appendix D Netzsch 021 Moineau pump datasheet ...................................................240
Appendix E Excerpt from (Thomson et al. 2001).......................................................241
E.1 Cost of water over the lifetime of the equipment............................................241

x
 List of figures

Figure 1-1 – Atmospheric carbon dioxide over the last 1000 years, illustrating the
dramatic rise since the industrial revolution (IPCC 2001)..........................................2
Figure 1-2 – Schematic of a simple reverse osmosis (RO) system.....................................3
Figure 1-3 – System overview ............................................................................................6
Figure 2-1 – Growth of worldwide desalination (Wangnick 2002; via Wiseman 2002)..10
Figure 2-2 – Theoretical minimum energy required to desalinate seawater at 25ºC
(Johnson et al. 1966 page 357) .................................................................................16
Figure 2-3 – Cost of PV (€/Wp) (Gottschalg 2001 page 254)...........................................24
Figure 2-4 – Typical I-V and P-V curves for a polycrystalline-silicon PV array..............26
Figure 2-5 – Spiral-wound RO element............................................................................30
Figure 2-6 – Pelton-wheel energy recovery......................................................................32
Figure 2-7 – Hydraulic Turbo Booster energy recovery...................................................33
Figure 2-8 – DWEER Work Exchanger energy recovery.................................................34
Figure 2-9 – ERI’s Pressure Exchanger............................................................................34
Figure 2-10 – Basic mechanics of a Clark pump ..............................................................38
Figure 2-11 – Simple configuration of a Clark pump in an RO system ...........................39
Figure 3-1 – Configuration of the initial test rig using a Danfoss hydraulic motor..........44
Figure 3-2 – Sankey diagram showing power flows in the initial test rig, using very old
membranes ................................................................................................................46
Figure 5-1 – Simulink model structure .............................................................................58
Figure 5-2 – Simulink structure possibilities ....................................................................60
Figure 5-3 – Ideal Clark pump..........................................................................................61
Figure 5-4 – Leakages.......................................................................................................62
Figure 5-5 – Pressure losses..............................................................................................65
Figure 5-6 – Configuration used for Clark pump testing..................................................67
Figure 5-7 – Simulink model of Clark pump....................................................................70
Figure 5-8 – Simulink model of CAT 237 triplex plunger pump .....................................73
Figure 5-9 – Modelled efficiency of CAT 237 plunger pump..........................................74
Figure 5-10 – Simulink model of Netzsch 021 Moineau pump........................................77
Figure 5-11 – Efficiency of Netzsch 021 Moineau pump modelled from manufacturer’s
performance curves ...................................................................................................78
Figure 5-12 – Standard industrial drive circuit .................................................................80
Figure 5-13 – Active and reactive motor currents versus torque......................................81
Figure 5-14 – Structure of Simulink motor model............................................................82
Figure 5-15 – Structure of Simulink inverter model.........................................................86
Figure 5-16 – Inverter losses versus motor current...........................................................87
Figure 5-17 – Annual average global irradiance versus time of day for a fixed PV array
and for single-axis and dual-axis solar-trajectory tracking arrays ............................90
Figure 5-18 – Hourly global irradiance in the plane of a single-axis tracking PV array for
the first week of the year-long data set .....................................................................92
Figure 5-19 –Average daily global irradiance in the plane of a single-axis tracking PV
array for the whole of the year-long data set ............................................................93
Figure 5-20 – Average monthly global irradiance in the plane of a single-axis tracking
PV array ....................................................................................................................95

xi
Figure 5-21 – Hourly ambient temperature for the first week of the year-long data set ..96
Figure 5-22 – Average monthly ambient temperature ......................................................97
Figure 5-23 – Two-diode model of a PV cell (Gottschalg 2001 page 60)........................98
Figure 5-24 – Two-diode model of a PV cell in Simulink ...............................................99
Figure 5-25 – Simulink model of a PV module ..............................................................100
Figure 5-26 – Simulink programme to locate maximum power point (MPP) ................101
Figure 5-27 – Electrical power available – hourly for the first week of the year-long data
set ............................................................................................................................102
Figure 5-28 – Electrical power available versus irradiance............................................103
Figure 5-29 – Electrical power available versus ambient temperature...........................104
Figure 5-30 – Average monthly electrical power available from PV array....................105
Figure 5-31 – Red Sea water temperature profile provided by Dulas Limited...............110
Figure 5-32 – Red Sea water temperature: data derived from Figure 5-31 and fitted sine
curve........................................................................................................................111
Figure 5-33 – Structure of the Simulink model of the array of RO elements.................112
Figure 5-34 – Structure of Simulink model of product tank...........................................114
Figure 6-1 – Top layer of the hierarchical Simulink model of the complete system......116
Figure 6-2 – Simulink model of the RO rig ....................................................................117
Figure 6-3 – Pie chart of annual energy consumption in the completed system design .122
Figure 6-4 – Delayed injection – single motor ...............................................................124
Figure 6-5 – Two motor – variable recovery ratio..........................................................126
Figure 6-6 – Optimum recovery ratio versus DC power available from PV ..................127
Figure 6-7 – A solar-thermal pre-heating configuration that was considered ................128
Figure 6-8 – Annual water production versus PV array size ..........................................131
Figure 6-9 – Capital cost per daily water production versus PV array size....................132
Figure 6-10 – Completed design.....................................................................................134
Figure 6-11 – Predicted product water flow versus irradiance .......................................136
Figure 6-12 – Predicted product water flow versus DC power available from the PV ..137
Figure 6-13 – Predicted product water flow – Monthly averages ..................................138
Figure 6-14 – Predicted product water flow – Daily totals.............................................139
Figure 6-15 – Predicted volume in product water tank...................................................140
Figure 6-16 – Predicted salt concentration in product water tank ..................................141
Figure 6-17 – Predicted energy flows – Sankey diagram ...............................................142
Figure 6-18 – Predicted analysis of power usage over an average day ..........................144
Figure 6-19 – Predicted specific energy vs. electrical input power ................................145
Figure 7-1 – Test rig configuration.................................................................................147
Figure 8-1 – Measured product flow (L/h) against pump speed setpoints .....................150
Figure 8-2 – Measured specific energy consumptions (kWh/m3) against pump speed
setpoints ..................................................................................................................151
Figure 8-3 – Measured specific energy consumption against DC power .......................152
Figure 9-1 – Indicative power curves for a PV array at 25°C ........................................154
Figure 9-2 – Indicative power curves for a PV array at 50°C ........................................155
Figure 9-3 – I-V and P-V curves .....................................................................................157
Figure 9-4 – MPPT hardware/software arrangement......................................................158
Figure 9-5 – MPPT basic implementation ......................................................................159
Figure 9-6 – MPPT enhanced implementation ...............................................................160
Figure 9-7 – Inverter power consumption vs. speed setpoint .........................................161
Figure 9-8 – Complete two-inverter MPPT control system structure.............................162
Figure 9-9 – Trajectory of MPPT operation, power vs. voltage .....................................163
Figure 9-10 –MPPT operation, power vs. time...............................................................164

xii
Figure 9-11 – MPPT operation over a 15-minute period................................................165
Figure 10-1 – Measured irradiance and product flow June 9th and 10th 2003 ................168
Figure 10-2 – Product flow versus irradiance.................................................................169
Figure 10-3 – Product flow versus DC power ................................................................170
Figure 10-4 – Specific energy (kWh/m3)........................................................................171
Figure 10-5 – Water recovery ratio.................................................................................172
Figure 10-6 – Product concentration...............................................................................173
Figure 10-7 – Product tank volume and concentration, basic case.................................174
Figure 10-8 – Product tank volume and concentration, diverting > 6000 mg/L.............175
Figure 10-9 – Membrane feed pressure ..........................................................................176
Figure 10-10 – Moineau pump efficiency ......................................................................177
Figure 10-11 – Combined efficiency of inverter and motor driving the Moineau pump178
Figure 10-12 –Combined efficiency of inverter and motor driving the plunger pump ..179
Figure A-1 – Speed setpoint profile................................................................................197
Figure A-2 – Measured speed .........................................................................................198
Figure A-3 – Measured inverter input power consumption............................................199
Figure A-4 – Measured feed flow versus pump speed ...................................................200
Figure A-5 – Measured water temperature .....................................................................201
Figure A-6 – Measured feed and concentrate pressures .................................................202
Figure A-7 – Measured product flow..............................................................................203
Figure A-8 – Measured product flow versus feed pressure ............................................204
Figure A-9 – Measured product flow versus inverter input power consumption ...........205
Figure A-10 – Specific energy versus inverter input power ...........................................206
Figure A-11 – Measured product concentration versus inverter input power ................207
Figure B-1 – Product-flow turbine flow meter calibration .............................................213
Figure B-2 – Feed-flow turbine flow meter calibration..................................................215
Figure B-3 – Clark-pump-inlet oval-gear flow meter calibration...................................216
Figure B-4 – Plunger-pump-inlet oval-gear flow meter calibration ...............................217
Figure B-5 – Product-flow oval-gear flow meter calibration .........................................218
Figure B-6 – Medium-pressure sensor specification and calibration .............................219
Figure B-7 – Concentration versus conductivity – linear scales.....................................221
Figure B-8 – Concentration versus conductivity – logarithmic scales ...........................222
Figure B-9 – Concentration trendline errors ...................................................................223
Figure B-10 – Torque load-cell manufacturer’s test datasheet .......................................224
Figure B-11 – Torque measurement calibration .............................................................225
Figure B-12 – Kipp and Zonen CM11 pyranometer calibration certificate....................227
Figure C-1 – Clark pump leakages and pressure losses..................................................234
Figure C-2 – Clark pump efficiencies.............................................................................235

xiii
 List of tables

Table 2-1– Typical practical energy consumptions for seawater desalination (Simplified
from Wangnick 2002, Table 4/2)..............................................................................17
Table 2-2 – A selection of existing brackish-water PV-RO systems................................40
Table 2-3 – A selection of seawater PV-RO demonstration systems ...............................41
Table 3-1– Measurements from the initial test, using very old membranes .....................45
Table 5-1 – Clark pump measured efficiency...................................................................71
Table 5-2 – CAT plunger pump geometries .....................................................................72
Table 5-3 – Netzsch 021 Moineau pump data extracted from manufacturer’s performance
curves (Appendix D).................................................................................................76
Table 5-4 – Toshiba motor data ........................................................................................81
Table 5-5 – Toshiba motor – derived data ........................................................................81
Table 5-6 – Alternative specifications for the same motor (Siemens1LA9 106-6KA) ....83
Table 5-7 – Modelled results for motor efficiency changes .............................................84
Table 5-8 – Costs and efficiencies of various 6-pole 1.5 kW induction motors...............84
Table 5-9 – Data for Figure 5-24 ......................................................................................91
Table 5-10 – Data for Figure 5-19 ....................................................................................94
Table 5-11 – Data for Figure 5-20 ....................................................................................95
Table 5-12 – Data for Figure 5-22 ....................................................................................97
Table 5-13 – Data for Figure 5-30 ..................................................................................105
Table 5-14 – The major constituents of seawater (Lyman and Fleming via Tonner 1999)
.................................................................................................................................108
Table 6-1 – Capital cost modelling estimates summary table ........................................119
Table 6-2 – Annual performance figures summary table ...............................................121
Table 6-3 – Delayed injection modelling results ............................................................125
Table 6-4 – RO array size modelling results ..................................................................129
Table 6-5 – PV array size modelling results ...................................................................130
Table 6-6 – Major parts list for completed design ..........................................................134
Table 6-7 – Predicted annual product volume for a fixed PV array and for single-axis and
dual-axis solar-trajectory tracking arrays................................................................138
Table 6-8 – Capital cost Dulas estimates summary table ...............................................146
Table A-1 – Caluclated powers and efficeincies ............................................................212
Table B-1 – Conductivity (µS/cm) and concentration (ppm) data for sodium chloride
solution (DOW 1995 section 10.2) .........................................................................220
Table B-2 – Data for Figure B-11...................................................................................225
Table C-1 – Clark pump specification ............................................................................228
Table C-2 – Clark pump test results data........................................................................230
Table C-3 – Clark pump analysis equations ...................................................................231
Table C-4 – Clark pump analysis results data (ordered to match Table C-2).................233
Table C-5 – Alternative formula considered for modelling leakages QL and pressure
losses PL ..................................................................................................................237
Table C-6 – Simulink model output data and errors (ordered to match Table C-2) .......239
Table E-1 – Outline maintenance schedule.....................................................................241
Table E-2 – Calculation of cost per cubic metre.............................................................242

xiv
The new PV-RO test rig

xv
 Chapter 1 Overview

1.1 Desalination and Renewable Energy

The desalination of seawater and brackish groundwater to provide fresh drinking water is
an established and thriving industry. The most commonly used technologies are thermal
distillation and reverse-osmosis (RO) filtration. Many towns and cities, particularly in the
Middle East and the US, already rely heavily on large-scale desalination plants for their
municipal water supplies. Small-scale desalination is also well-established, for example
on ships.

Meanwhile, the world is facing a serious water crisis: 1.1 billion people (one-sixth of the
world’s population) have no access to improved drinking water, and “All the signs
suggest that it is getting worse and will continue to do so, unless corrective action is
taken” (UNESCO 2003). Desalination will undoubtedly play an increasing role in
meeting worldwide water needs, but is limited by its cost, which is largely dominated by
energy costs. Desalination intrinsically consumes a lot of energy, and this is quantified in
the specific energy, which is the energy consumed in desalinating a unit volume of water,
and is usually given in kWh/m3. The theoretical minimum is around 0.7 kWh/m3, when
operating from seawater (Johnson et al. 1966 p. 357). In practice today, energy
consumption many times greater than this is typical, which explains why the desalination
industry exists mainly in countries with large fossil-fuel reserves.

The energy consumption of desalination also has an environmental impact, in particular

the release of carbon dioxide (CO2) into the atmosphere through the burning of fossil
fuels. Prior to the industrial revolution in the 1760’s, the concentration of CO2 in the
earth’s atmosphere was around or below 280 parts per million (ppm), and had been for
hundreds of thousands of years. Since the industrial revolution, mankind has raised this
dramatically, as shown in Figure 1-1.

1
 360

340
CO2 (ppm)

320

300

280

260

1000 1200 1400 1600 1800 2000

Year

Figure 1-1 – Atmospheric carbon dioxide over the last 1000 years, illustrating the
dramatic rise since the industrial revolution (IPCC 2001)

Graphs similar to Figure 1-1 exist for other greenhouse gases. The full consequences of
making such significant changes to the earth’s atmosphere are as yet unknown, but to
continue to increase concentration levels seems foolhardy to say the least. Desalination is
already a significant contributor to worldwide CO2 emissions (see section 2.3.3), and this
of course will increase as the industry grows.

CO2 emissions can be greatly reduced through the application of renewable energy
technologies, which are already cost competitive with fossil fuels in many situations.
Good examples include large-scale grid-connected wind turbines, solar water heating and
off-grid solar photovoltaics (PV). The use of renewable energy for desalination is,
therefore, a very attractive proposition.

Renewable energy can come from a variety of sources and can be captured by a variety of
technologies; likewise, desalination can be achieved by various methods. The numerous
permutations combining these have been reviewed several times in recent years.
Repeatedly, the combination of photovoltaics with reverse osmosis (PV-RO) has been
considered one of the more promising, particularly for small-scale systems where other

2
technologies are less competitive. Indeed, PV-RO systems for operation from brackish
water are now commercially available as discussed in section 2.9.1. Operation from
seawater is more challenging from an energy perspective, and is the subject of this thesis.

1.2 Reverse Osmosis (RO)

Reverse osmosis is a form of filtration, in which the filter is a semi-permeable membrane
that allows water to pass through, but not salt. When a membrane of this type has
saltwater on one side and freshwater on the other, and in the absence of applied
mechanical pressure, water will flow through the membrane towards the saltwater side,
evening out the concentrations and reducing the quantity of freshwater. This is the
natural process of osmosis, and is widely employed in the cells of all living species. In
desalination, of course, the aim is to increase the quantity of freshwater and so a pump is
employed to make the flow reverse, hence the name: reverse osmosis.

Osmosis is a surprisingly powerful phenomenon; the osmotic pressure of typical

seawater is around 26 bar, and this is the pressure that the pump must overcome in order
to reverse the flow. (26 bar also equates to the theoretical minimum energy consumption
of 0.7 kWh/m3, mentioned earlier.) In practice, a significantly higher pressure is used,
typically 50-70 bar, in order to achieve a generous flow of freshwater, which is the
product, also known as the permeate.

Of course, as freshwater passes through the membrane, the remaining saltwater becomes
more concentrated and, for the process to continue, this concentrate, also known as the
brine, must be continuously replaced by new feed water. To achieve this, the feed water
is pumped across the membrane as well as through it; hence, RO is a cross-flow filtration
process as depicted in Figure 1-2.

Reverse-osmosis membrane

Pump
Freshwater
Seawater High- Product
Feed pressure Permeate
Concentrate
Brine

Figure 1-2 – Schematic of a simple reverse osmosis (RO) system

3
The ratio of product flow to that of the feed is known as the recovery ratio. With
seawater RO, a recovery ratio of 30% is typical, meaning that the remaining 70% appears
as concentrate, which is returned to the sea. However, this concentrate comes out of the
reverse osmosis module at a pressure only slightly below that of the feed, meaning that it
contains roughly two thirds of the total hydraulic power originally supplied by the pump.
In large RO systems, this energy is usually recovered by way of a Pelton turbine and
returned to the shaft of the main pump, allowing the motor size to be roughly halved and
dramatically improving the overall system efficiency. This is known as brine-stream
energy recovery. In small RO systems, brine-stream energy recovery is often omitted,
which reduces capital costs but adds considerably to running costs (energy).

Reverse osmosis is a widely scaleable technology: the membranes found in small

systems used on pleasure yachts to produce a few litres per day are virtually identical to
those used in large municipal plants producing thousands of cubic metres per day. The
balance of plant (pumps etc.) is, however, widely different and this usually leads to lower
energy efficiencies in smaller systems, particularly when brine-stream energy recovery is
neglected.

1.3 Photovoltaics (PV) and Batteries

Photovoltaic panels convert sunlight directly into electricity, and are already widely used
in critical applications such as vaccine refrigeration, water pumping and battery charging
for lighting and communications. PV is highly reliable and is often chosen because it
offers the lowest life-cycle cost, especially for applications requiring less than 10 kW,
where grid electricity is not available and where internal-combustion engines are
expensive to maintain. PV is a rapidly developing technology with costs falling year on
year, and this will soon lead to its broad application in systems requiring larger powers.
Today however, it is clear that PV-RO will initially be most cost-competitive at the
small-scale, perhaps for supplying remote villages or small hotels. The challenge then
becomes the energy efficiencies in the balance of plant: pumps, motors and the critical
brine-stream energy recovery mechanism.

Batteries are widely used in PV systems, storing the energy during the day and making it
available through the night. Unfortunately, batteries are notoriously problematic in
practice, especially in PV systems in hot countries. Experienced PV system designers

4
avoid batteries whenever possible. PV water pumping, for example, is usually designed
without batteries, in which case the pump runs only during the day and the water is stored
in a tank, if necessary. Likewise, PV refrigeration is possible without batteries, by virtue
of storing the cold in the form of ice or other phase-change materials. From the outset,
the project described in this thesis aimed to design a PV-RO system that would operate
without batteries: desalinating water during the day and storing the product water in a
tank. This seems straightforward, but is contrary to the normal 24-hour-a-day operation
of mainstream RO systems.

Furthermore, in the absence of batteries, the power available from a PV array varies with
the intensity of the sunlight, the irradiance, and, in order to make best use of this
available power, a connected RO system must also operate at variable power. Again, this
is contrary to the normal operation of mainstream RO systems. Also, the efficiency of the
RO system must be maintained over a broad range of operating power; this is particularly
challenging in the balance of plant.

1.4 Proposed PV-RO system

The aim of the project leading to this thesis was the design of a cost-effective PV-RO
system that can desalinate seawater and does not require batteries. Such a system is
presented in Figure 1-3. The design was developed by the author, with assistance from
Spectra Watermakers Inc, Dulas Limited and Marcos Miranda, as specifically
acknowledged in later chapters.

The chosen brine-stream energy recovery mechanism is the Clark pump from Spectra
Watermakers Inc. It is a positive-displacement reciprocating pressure intensifier, and an
animation of its operation can be viewed online at:
[Link] (accessed: March 04) .
Testing at CREST showed that the Clark pump is very efficient, typically above 90%, which
is impressive indeed for any kind of pump on this scale.

5
 Photovoltaic
Array

Inverter Controller Inverter Product

Tank
Solar
Tracker
Induction Induction
Motor Motor
Pulsation
Dampener

Pressure
Plunger Relief
Valve
Coarse Fine Pump
Filter Filter
Moineau Reverse-Osmosis
Pump Modules

Pulsation Clark
Dampener Pump

Beach-well
Seawater Concentrate
Intake Discharge

Figure 1-3 – System overview

Referring to Figure 1-3, the Moineau (progressing cavity) pump sucks seawater from a
beach well and raises it to a medium-pressure (6 – 11 bar). A submersible version of the
same pump is available, at extra cost, and could readily be substituted. The Clark pump
raises its medium-pressure feed water to high-pressure (40 – 70 bar) by virtue of the
energy it recovers from the concentrate. The flow ratio of the Clark pump is fixed by
design, which normally gives a 10 % water recovery ratio at the membranes. However,
the plunger pump injects an additional high-pressure feed, which increases the water
recovery ratio to any desired value.

The controller Figure 1-3 in provides speed-set-point signals for the inverters and motors
that drive the two rotary pumps. Firstly, this provides maximum power point tracking
(MPPT), which ensures that the total power drawn from the PV array tracks the
maximum available as the irradiance varies throughout the day. Secondly, the relative
speeds of the two pumps are controlled in order to maximise the flow of product water.
The controller algorithms are discussed in chapters 8 and 9.

6
1.5 Modelling
There are a vast number of possible configurations of motors, pumps, energy-recovery
mechanisms and so on that can be used in a reverse osmosis system. The configuration
outlined in Figure 1-3 was developed largely through computer simulation of numerous
possibilities and an evaluation of their cost, performance and practicality. The simulation
model was developed in Matlab-Simulink and includes everything from the solar
irradiance striking the photovoltaic panels through to the water in the product tank. Each
of the hardware components is modelled in detail and the critical sections are based on
accurate measurements made in CREST’s laboratory. The model fully represents the
variations of flows and pressures throughout the system with respect to variations in
irradiance and feed-water temperature. It can, for example, perform an hour-by-hour
simulation of a whole year of operation.

1.6 System hardware testing

A test rig was constructed, with assistance from Marcos Miranda, and operated from a
PV array at CREST to demonstrate and verify the design.

Of course, many other researchers worldwide have designed and built similar PV-RO
systems, and these will be discussed in section 2.9. Many can work only from brackish
water and most employ batteries or very large PV arrays or both. It is acknowledged
however that many of these systems are operating in the field, whereas the system
described here has, to date, only been tested in a laboratory.

1.7 Performance figures

To supply a remote village or small hotel, a freshwater production of 3 m3/day was
suggested by Dulas Limited, and was adopted as a design target.

1.7.1 Predictions from software model

Using solar irradiance data for a site in Eritrea, as an example, and taking seawater feed
at 38,000 mg/L total dissolved solids (TDS) varying annually between 17 and 33 °C:

Annual freshwater production was predicted at 1424 m3, which is an average of just over
3.9 m3/day, and had a minimum monthly average of 3.3 m3/day.

7
Specific energy consumption (photovoltaic-electricity) was typically between 3.2 and
3.7 kWh/m3 depending on the solar irradiance and feed water temperature

Capital costs, including the PV array, were estimated (with assistance of Dulas Limited)
at £23,055, with an overall cost of water of £2.00 per m3, including full maintenance. The
system has no fuel costs and no batteries.

1.7.2 Actual measurements

The completed test rig at CREST, achieved a specific energy consumption (DC-
electricity supply, rectified mains) of between 3.5 and 3.8 kWh/m3, with freshwater
production adjustable between 181 and 488 L/h. Production was adjustable down to
17 L/h at lower efficiency.

PV-powered operation was demonstrated and, though limited by ambient UK irradiance

levels coupled to an undersized PV-array (64% of design size), achieved freshwater
production of 1.5 m3/day.

Testing used NaCl solution at 32,800 mg/L, which is isosmotic with ASTM (1998)
standard substitute ocean water, at 25 °C.

1.8 Global relevance

It may appear that a PV-RO system to produce 3 m3/day is but a drop in the ocean
compared to what is required to tackle the world’s energy and water problems. Indeed it
is, but the concept of matching variable loads, particularly RO desalination to naturally-
varying renewable-energy sources, is valuable at a much larger scale. Renewable-energy
sources, including solar, wind and wave power, are naturally varying, and often do not
coincide with man’s perceived energy requirements. Because of this, it is often assumed
that very large-scale energy storage will be required in order to operate large electrical
power systems from these sources. Certainly, energy storage is very valuable in the
operation of power systems, but it is also very expensive. Batteries are expensive and
problematic even at the small scale, pumped hydro is expensive both in civil engineering
and environmentally, and other means of storing electrical energy are mostly at the
development stage. The alternative to electrical energy storage is demand-side
management. This is sometimes referred to as curtailment or use of deferrable loads.
These expressions suggest that the customer will have to endure interruptions in their

8
supply, but this need not be the case. It is better to liken demand-side management to
making hay when the sun shines, which, in the case of RO desalination, translates to
making water when electricity form renewable energy is available. Demand-side
management can also be readily applied to air conditioning, another major consumer of
energy, through the use of ice storage. Systems doing exactly this are commercially
available, but rarely applied. The low uptake of demand-side management, even in
countries where it would be very advantageous, is largely due to split responsibilities:
electricity generation, RO desalination and air conditioning are all operated by separate
agencies. There are, however, some technical factors. In particular, RO desalination
performs best when run at a constant flow 24-hours-a-day. Operation of large-scale RO
desalination plants at variable flow to enable greater use of renewable energy sources
will require significant changes to existing practices. It makes good sense to build up
practical experience on small systems, such as that described in this thesis.

Another problem facing our planet is the general migration of people towards cities,
driven partly by water and electricity shortages in rural areas. Small-scale renewable-
energy powered systems providing these resources locally could perhaps reduce this
trend.

9
 Chapter 2 Technology and literature review

2.1 Desalination

2.1.1 A growth industry

25

20
Capacity (Mm3/d)

15

10

0
1965 1970 1975 1980 1985 1990 1995 2000
Year

Figure 2-1 – Growth of worldwide desalination (Wangnick 2002; via Wiseman

2002)

The worldwide desalination industry now provides over 24 million cubic metres of
freshwater per day. The growth of the industry over the last 37 years is shown in
Figure 2-1 and can be expected to continue. The capacity shown is dominated by large-
scale plants supplying large centres of population. In particular, the data includes only
plants with capacities of 100 m3/d and above; there are over 15,000 such plants
worldwide. Additionally, there are many more thousands of smaller desalination units on
ships, on islands and in remote areas.

10
2.1.2 Seawater versus brackish water
The desalination industry makes a distinction between seawater and brackish water
(DOW 1995). Seawater typically has a salt concentration in the order of 36,000 mg/L
total dissolved solids (TDS), while brackish water, usually from underground, might be
between 3,000 and 10,000 mg/L TDS. Seawater intrusion of over-exploited groundwater
supplies can blur the distinction, but still it is a useful starting point.

With desalination by reverse osmosis, the energy required to desalinate brackish water is
considerably less than that for seawater. The system described in this thesis is
specifically intended for seawater desalination; it could be used for brackish water but
would be expensive. Also, brackish water tends to be more variable in terms of the
proportions of salts it contains and this can complicate fouling and scaling
considerations.

2.1.3 Product water specification

An important consideration for any desalination system is the quality required of the
product water. Some systems provide very pure water for particular industrial processes,
but the majority are designed to provide drinking water, and that of course is the aim of
the PV-RO system in hand.

RO membranes are not perfect; they do allow some salt through. Seawater-RO
membranes are generally designed to give product concentrations of 200 to 300 mg/L,
when operating at full flow and pressure on the feed-water side. But, operation at
significantly lower flow and pressure causes the product concentration to rise
considerably. This is an important consideration when designing RO systems for best
energy efficiency, since best energy efficiency usually occurs at flow and pressure well
below maximum. The consideration has further importance in the case of a batteryless
PV-RO system, because, when the available solar power is low, the system will operate
at further reduced flow and pressure. The batteryless PV-RO system described in this
thesis does at times operate at very low flow and pressure and, as expected, the
concentration of the product is very high. Of course, this is concurrent with very low
product flow and, once mixed in a tank with better quality water made earlier, may well
be acceptable. Nonetheless, it is clearly important to establish an appropriate upper limit
for the concentration of the water delivered to the consumer.

11
In desalination literature, it is often stated that the World Health Organisation (WHO)
recommends a limit of 500 mg/L TDS, but the exact origin of this figure is unclear. The
WHO is currently preparing the Third Edition of its Guidelines for Drinking-Water
Quality and a draft (February 2003) is available on line (WHO 2003). It is a very
comprehensive document, and includes notes describing the historical development of
the guidelines with reference to guidelines given in previous editions. There is no
evidence of a 500-mg/L limit within the WHO Guidelines.

The US Environmental Protection Agency (EPA), however, does set a guideline of

500 mg/L maximum TDS in their National Secondary Drinking Water Regulations (EPA
2004), and it is possible that this is source of the figure often attributed to the WHO in
desalination literature. The following discussion is based on the WHO Guidelines.

The salt in seawater is primarily sodium chloride, NaCl, and unsurprisingly it is primarily
sodium chloride that makes its way into the product.

Regarding sodium, the WHO Guidelines (Third Edition, 2003, chapter 8, page 248) say:

…no health-based guideline value is proposed. However, concentrations in excess

of 200 mg/litre may give rise to unacceptable taste (see chapter 10).

Likewise for chloride (chapter 8, page 236):

No health-based guideline value is proposed for chloride in drinking-water.

However, chloride concentrations in excess of about 250 mg/litre can give rise to
detectable taste in water (see chapter 10).

And for total dissolved solids (TDS) (chapter 8, page 250):

…no health-based guideline value is proposed. However, the presence of high

levels of TDS in drinking-water may be objectionable to consumers (see
chapter 10).

Apparently, the first edition of the guidelines (1984) did suggest a limit of 1000 mg/L for
TDS, based on taste. This limit was removed in the second edition. There is no mention
of any limit at 500 mg/L.

Chapter 10 of the Guidelines indicates that the taste threshold of sodium chloride is
around 200 mg/L (pages 215 and 217). And regarding total dissolved solids (page 218):

12
 The palatability of water with a TDS level of less than 600 mg/litre is generally
considered to be good; drinking-water becomes significantly unpalatable at TDS
levels greater than 1200 mg/litre.

In summary, the acceptable concentration of salt in RO-desalinated seawater is primarily

a matter of taste, not health – assuming that the water is not so unpalatable as to
encourage people to drink from less-safe sources (chapter 10, page 213). The acceptable
taste will depend greatly on who the consumers are, what they are accustomed to and
what their alternatives are. A product concentration of 500 mg/L remains a useful target,
but in areas with no alternative supply, 1000 mg/L may well be perfectly acceptable,
especially if this can be produced in greater quantity or with improved reliability.

The above discussion relies entirely on the WHO Guidelines. These are widely used as a
basis for national and regional standards, and there is little variation from country to
country. In the main, the WHO Guidelines assume a surface water or groundwater
source; there are additional considerations for desalination systems (section 6.5 of the
Guidelines), which are not specific to PV-powered systems and are not discussed further
in this thesis.

2.2 Desalination techniques

The separation of freshwater from saltwater can be achieved in many ways and these are
discussed at length in standard texts on desalination (Spiegler et al. 1994; El-Dessouky et
al. 2002). A excellent overview is available on the internet (Buros 2000). The following
section provides a brief introduction, intended only to set reverse osmosis in context.

2.2.1 Distillation
Heating water in a tank causes it to vaporise, leaving behind any salt. Cooling the vapour
causes it to condense as freshwater, which may be collected in a separate tank. A simple
still of this type is easy to construct, but very inefficient in energy terms. The heat energy
required is the latent heat of evaporation, which is around 627 kWh/m3, plus losses. A
comparison of energy consumptions is given later in section 2.3.2.

2.2.2 Multi-Effect Distillation (MED)

Most of the heat energy consumed in a simple still ends up in the coolant of the
condenser. Recycling this heat energy can improve efficiency several fold. Of course, the

13
temperature of the condenser is not high enough to heat the saltwater in the original tank,
but it can be used to heat a second tank held at a lower pressure. Practical distillation
systems often have many tanks, known as effects, hence the term Multi-Effect Distillation
(MED).

MED was developed for desalination purposes during the first half of the twentieth
century, but had a major practical problem with the build up of scale on the outside of the
heating pipes, rather like the scaling of the heating element in an electric kettle (Arrindell
et al. 2002).

2.2.3 Multi-Stage Flash (MSF)

In flash distillation, the water is heated under pressure, which prevents it from vaporising
while being heated. It then passes into a separate chamber held at lower pressure, which
allows it to vaporise, but well away from the heating pipes, thus preventing them from
becoming scaled. Like MED, practical flash-distillation systems are divided into sections,
but this time known as stages, hence the term Multi-Stage Flash (MSF).

When first introduced in the 1960’s, MSF offered slightly lower energy efficiency than
MED, but this was outweighed by scaling considerations and MSF became the industry
standard.

2.2.4 Vapour Compression

Compressing water vapour raises its temperature, which allows it to be used at a heat
source for the same tank of water that produced it. This allows heat recycling in a single-
effect distillation process.

In Thermal Vapour Compression, the compressor is driven by steam, and such systems
are popular for medium-scale desalination because they are simple, in comparison to
MSF.

In Mechanical Vapour Compression, the compressor is driven by a diesel engine or

electric motor.

14
2.2.5 Reverse Osmosis (RO)
Reverse Osmosis (RO) is a membrane filtration process and, in contrast to the distillation
processes just described, does not involve vaporising the water. This generally leads to it
being much more energy efficient. RO is the technology chosen for the system described
in this thesis, and is described in sections 1.2 and 2.7.

2.2.6 Electrodialysis
Electrodialysis also uses membranes, but unlike RO, the salt ions are deliberately carried
through the membranes, leaving behind the freshwater. Two types of membranes are
required: one that lets anions through but not cations, and the other that does the
opposite. These membranes are stacked alternately and held apart by spacers. The
saltwater is fed into the spacer layers on one side of the stack, and a DC voltage is
applied to the stack as a whole. The salt ions are attracted through one membrane or the
other depending on their polarity, and by the time the water comes out of the other side
of the stack, it is alternately freshwater and concentrate in the spacer layers. Reversing
the polarity of the applied voltage reverses the freshwater and concentrate layers, and this
can be done periodically (several times per hour) in order to reduce fouling, and is
termed Electrodialysis Reversal.

Electrodialysis was commercialised during the 1960’s and is widely used today for
desalinating brackish water. The energy consumption depends very much on the
concentration of the feed water and so electrodialysis is rarely used for seawater
desalination.

2.3 Energy for desalination

2.3.1 Theoretical
It has already been stated that the theoretical minimum energy required to desalinate
seawater is around 0.7 kWh/m3. This is true, but is also a little misleading, because it
assumes that the volume of seawater is infinite. Of course, the oceans are virtually
infinite, in the scale of mankind’s freshwater requirements, but practical desalination
systems have to work with a finite flow of seawater, since it must usually be pumped
from the sea.

15
The proportion of freshwater extracted from the seawater is known as the recovery ratio.
Obviously, to minimise feed-water use and pumping requirements, it is best to use a high
recovery ratio. On the other hand, in order to limit scaling and desalination energy, it is
best to use a low recovery ratio.

The theoretical minimum energy required for desalination increases with recovery ratio
as shown in Figure 2-2. The increase is readily explained in context of reverse osmosis,
because, as freshwater is extracted, the remaining brine becomes more concentrated,
which increases its osmotic pressure and the pump has to work harder to extract further
freshwater.

3.5

2.5
Minimum Energy (kWh/m 3)

1.5

0.5

0
0 10 20 30 40 50 60 70 80 90 100
Water Recovery Ratio (%)

Figure 2-2 – Theoretical minimum energy required to desalinate seawater at 25ºC

(Johnson et al. 1966 page 357)

The separation of freshwater from saltwater is a reversible process, in the sense that
mixing them back together will release the same theoretical minimum energy that went
into separating them. Thus, the process must obey the Second Law of Thermodynamics,
meaning that the minimum energy required is the same regardless of the method used to
achieve the separation. Therefore, the theoretical minimum energy shown in Figure 2-2

16
applies to all desalination techniques, including distillation and electrodialysis (Spiegler
et al. 1994).

An appreciation of the theoretical minimum energy is useful in the design of reverse

osmosis systems, but as we will see in the next section, practical desalination technology
is, as yet, far from ideal.

2.3.2 In practice
It is common practice to couple large thermal desalination plants (MSF and MED) to
electricity generating stations, since this provides better overall efficiency. The energy
consumption for the desalination may then be quantified in terms of the electrical energy
that could otherwise have been generated. This approach is taken in the IDA Worldwide
Desalination Plants Inventory (Wangnick 2002) and leads to the data presented in
Table 2-1.

Electrical-equivalent
energy consumption
(kWh/m3)
MSF 15.5
MED 6.5
Mechanical Vapour Compression 8 – 14
Reverse Osmosis (RO) 4–7
Table 2-1– Typical practical energy consumptions for seawater desalination
(Simplified from Wangnick 2002, Table 4/2)

Comparing the figures in Table 2-1 against the latent heat of evaporation: 627 kWh/m3,
illustrates the considerable benefits brought to the thermal processes by heat recycling
and by coupling MSF and MED to electricity generating stations. Despite these, it also
illustrates that RO is currently the most energy efficient technology for seawater
desalination. There are other considerations of course, particularly the capital and
maintenance costs. Also, RO is a relative newcomer and is considered less reliable,
particularly with regard to fouling caused by difficult feed waters (Wiseman 2002).
Nonetheless, the attraction of RO for renewable-energy powered desalination is already
apparent.

The data presented in Table 2-1 is typical for installed plants with capacities of 100 m3/d
and above. State-of-the-art RO plant can now achieve 2 kWh/m3 (MacHarg 2001). But,

17
older and smaller RO systems, particularly those without brine-stream energy recovery,
can easily consume up to 15 kWh/m3.

2.3.3 Energy sources

Virtually all desalination today is powered by fossil fuels and nuclear energy; countries
with large hydroelectric capacity do not generally require water through desalination.

It is tempting to estimate the total worldwide energy consumption for desalination, but
this is almost impossible because it is so intertwined with electricity production. As
already noted, many desalination plants are coupled to electricity generation stations, but
also, the provision of water allows cities to grow thereby increasing electricity demand.
Nonetheless, this calculation was attempted in 1996 (Rodriguez-Girones et al. 1996) and
suggested that the total worldwide energy consumption for desalination at that time was
436 TWh/yr (37.5 Mtoe/yr), which was comparable to the gross energy demand of
Sweden. Considering the growth of desalination presented earlier in Figure 2-1, it is safe
to conclude that desalination is already a significant contributor to worldwide CO2
emissions, and that this is set to increase considerably.

In addition to the consideration of CO2 emissions, is the fact that many countries with
water shortages simply do not have significant fossil-fuel reserves, and the import of
such fuels places a tremendous burden on their economies.

18
2.4 Renewable energy for desalination
CO2 emissions can be greatly reduced through the use of renewable energy, which is
often abundant in countries with water shortages. This has led to great interest in using
renewable energy for desalination. Indeed, many pilot demonstration plants have already
been built and operated (See databases in: Rodriguez-Girones et al. 1996; CRES 1998;
Wangnick 2002), and an even greater number of theoretical investigations completed.

Wind energy and solar energy have the biggest immediate potential for powering
desalination. Wave power also has great potential, but a reliable means of capturing it has
yet to be demonstrated. Tidal and geothermal energies may well be suitable, but are only
available in a few locations. Hydropower and biomass are not widely available in arid
regions.

2.4.1 Intermittency
Unfortunately, both wind and solar power are intermittent, which leads to three options:

a) Use fossil fuel to make up the gaps, so that the desalination plant can run
continuously.

b) Store the renewable energy, so that the desalination plant can run continuously.

c) Run the desalination plant intermittently.

Option a) includes grid-connected systems, in which power from a large electricity grid
is used to run the desalination plant whenever wind or solar power is not available. This
greatly simplifies design and operation. Of course, the wind or solar power plant can be
made large enough so that, on average, it supplies at least as much energy as is
consumed by the desalination plant. This is commendable, since it offsets the burning of
fossil fuel elsewhere, but it should be understood that the desalination plant is still reliant
on the grid connection and therefore on the burning of fossil fuel. Considering an
electricity grid as a whole, it is often found that the penetration of intermittent sources
such as wind and solar is limited to around 20% of the total energy consumption (Jensen
2000 Table 1 page viii). In order to increase the allowable penetration requires either
energy storage or demand-side management, which correspond to options b) and c)
respectively, and which were discussed in section 1.8. In summary, use of a grid

19
connection, where available, greatly simplifies the operation of the wind or solar
powered desalination plant today, but for the future, we must look to options b) or c).

Systems without a grid-connection are generally described as stand-alone, and can

operate under any of the options a), b) or c). A system employing a diesel engine to cover
periods when wind or solar power is not available comes under option a), and, in the
absence of energy storage, the contribution from wind or solar is likely to be limited to
about 20%, as before. In order to reduce the reliance on diesel requires a move towards
options b) or c).

Dealing with intermittency, either through energy storage or intermittent operation of the
desalination plant, is central to the development of the use of renewable energy for
desalination.

2.4.2 Compatibility
The next few sections discuss the viability of some of the more promising combinations
of renewable-energy and desalination technologies.

2.4.3 Simple solar stills

Simple solar stills are probably the oldest form of man-made desalination equipment, and
the sheer simplicity of the concept gives it eternal appeal. A simple solar still comprises a
glass-covered basin of saltwater. The greenhouse effect heats the water causing it to
evaporate. It then condenses on the glass as freshwater and dribbles down into a
container separate from the original saltwater. Unfortunately, simple solar stills yield
very little water: 5 L/d/m2 is a typical best, meaning that 3 m3/d would require an area of
600 m2. (The PV-RO system described in this thesis should achieve the same output with
only 20 m2 of PV.) The low productivity arises partly from the fact that a simple solar
still uses single-effect distillation, meaning that it uses the latent heat of evaporation
(627 kWh/m3) and none of this heat is recycled. Also, having the solar collector and the
evaporators as one unit leads to significant heat losses (Spiegler et al. 1994). Even if land
is freely available, the maintenance of large solar stills (keeping the glass clean and
vapour-tight) has repeatedly proved impractical (Delyannis et al. 2001).

20
2.4.4 Indirect solar thermal desalination
Innumerable attempts have been made to improve the performance of solar stills, and the
more successful of these generally involve separating the solar thermal collector from the
evaporator, which is known as indirect solar thermal desalination. Storage of the thermal
energy in hot-water tanks is then possible, which allows production to continue through
the night, and can improve efficiency by virtue of the condensing surface being cooler.
Improved solar collectors may be used, such as parabolic-trough concentrators (mirrors)
and evacuated-tube collectors. For further efficiency, it is usual to employ some
recycling of the latent heat of evaporation, which moves the design towards multi-effect
distillation (MED) or multi-stage flash (MSF).

A thorough discussion of the various solar-thermal desalination options is given by

Kalogirou (1997) and a review of numerous demonstration plants is given by Garcia-
Rodriguez (2002). A more detailed discussion of the economic feasibility of small solar
MED seawater desalination plants for remote arid areas is presented by El-Nashar
(2001).

Unfortunately, as shown earlier in Table 2-1 the heat energy demand of even the most
efficient MED and MSF systems is considerable. Furthermore, to achieve these
efficiencies, MED and MSF systems are complex to operate and therefore tend to be
more successful at large-scale. In conclusion, indirect solar thermal desalination remains
interesting at medium and large scale, but will face increasing competition from
photovoltaic-powered reverse osmosis. However, there may be an alternative, at the
small scale, in the form of Membrane Distillation.

2.4.5 Membrane Distillation

The membranes used in Membrane Distillation are unlike Reverse Osmosis or
Electrodialysis membranes in that they are hydrophobic, meaning that they resist getting
wet and therefore block the passage of both water and dissolved salts. They are, however,
permeable to water vapour, rather like Gortex™. Membrane Distillation is primarily a
heat-driven process and can achieve the recycling of the latent heat of evaporation
without the complexity of multiple effects or a vapour compressor. This makes
Membrane Distillation particularly interesting for small-scale application. Furthermore,
since the membrane does not have to be selective between water and salt ions, the

21
effective pore size can be 1000 times larger than for RO, and this, coupled with the fact
that it does not get wet, greatly reduces concerns regarding fouling. Membrane
Distillation is still at the research stage but holds great promise for small-scale solar-
powered application (Koschikowski et al. 2003; Cabassud et al. 2003).

2.4.6 Freezing
When saltwater is frozen the ice that forms is nearly pure and the salt is left in the
remaining liquid. Freeze desalination was vigorously researched during the 1950’s and
60’s because it promised to be more energy efficient and have less scaling problems than
thermal distillation processes (Snyder 1966). But, difficulties in the mechanical handling
and separation of the ice from the water were never resolved and research was abandoned
when reverse osmosis became available.

Meanwhile freezing has been commercially developed for the purpose of energy storage
in association with air conditioning. Tanks containing eutectic aqueous solutions are
frozen during the night using off-peak electricity and melted during the day to provide
cooling ([Link] accessed: March 04).

Observing that desalination is often employed in areas that also have a large air
conditioning demand, one wonders if these two could not profitably be combined. This
may be particularly interesting for wind-powered systems where the energy source is
intermittent.

2.4.7 Wind-powered Reverse Osmosis

Wind power is now an established technology, in many situations providing electricity at
a cost well below that of electricity derived from fossil fuels. Thus, the combination of
wind-power and reverse osmosis appears very attractive, even for sites with a modest
wind resource. The challenge lies in the variability of the wind.

Standing in a field, one might have the impression that a steady wind is blowing;
measuring that wind with an anemometer and displaying it on an oscilloscope usually
reveals that it is in fact highly variable. Dealing with this variability is one of the main
challenges that faces wind-turbine designers. And, connecting an RO system directly to a
wind turbine, without some form of energy storage buffer, brings that variability to the
RO membranes. Most RO designers shy away from this, possibly with good reason, and

22
wind-powered RO systems often rely to some extent on fossil fuels: systems with an on-
site diesel engine are often called hybrid systems; others rely on a grid-connection as
discussed in section 2.4.1. Reverse osmosis systems relying solely on wind power are
few and far between. The SDAWES project on Gran Canaria employed flywheel energy
storage and switching on and off of eight sets of RO membranes (Rahal 2001). Enercon
have a number of demonstration systems running, but reveal little about them (Lührs
2003). Vergnet are more open, but perhaps less successful (Fabre 2003). A full review of
wind-powered reverse osmosis is doubtless contained in Marcos Miranda’s thesis (2003).

A rather different form of wind-powered reverse osmosis, but worthy of mention if only
for its sheer elegance, is the Waterlog ([Link] accessed:
March 04). This is a device that can be dragged through the sea behind a sailing boat. It
has a water turbine that operates a pump that feeds an RO membrane. The towline is
hollow and carries the freshwater back to the boat.

2.4.8 Other studies

The foregoing sections discuss the viability of some of the more promising combinations
of renewable-energy and desalination technologies. Several rather more comprehensive
studies have been completed in recent years (Rodriguez-Girones et al. 1996; CRES 1998;
Oldach 2001). A shorter discussion of the same subject is available online
(Assimacopoulos et al. 2001) and, in broad agreement with the comprehensive studies,
states: “PV-RO (Photovoltaic powered-Reverse Osmosis) is clearly the favoured
desalination combination for small stand-alone systems”. PV-RO is of course the
combination chosen for work described in this thesis, and will be discussed in
section 2.9, after discussing PV and RO separately.

2.5 Photovoltaics (PV)

Photovoltaic cells convert sunlight directly into electricity. Green (1995) discusses their
physics in detail. Roberts (1991), Wenham (1994) and Markvart (1999) include more of
their application.

A typical cell is 10 by 10 cm and produces a few watts of DC electricity at around

0.5 volts. To achieve greater powers and voltages, cells are connected in series and
assembled into modules, also called panels. A typical module comprises 36 series-

23
connected cells. This provides a voltage that is convenient for charging 12-volt lead-acid
batteries, but such modules are also widely used in systems not employing batteries,
including grid-connected systems. Modules are connected together to form arrays, the
size of which is limited usually by cost.

2.5.1 Cost
25

20

15
Euro/Wp

10

0
1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000
Year

Figure 2-3 – Cost of PV (€/Wp) (Gottschalg 2001 page 254)

The cost of PV has fallen steadily, as shown in Figure 2-3, and this is expected to
continue for several years yet. The cost today is still considerable, but there are many
situations where PV already offers a cost-competitive solution, although, to make this
apparent, it is usually necessary to consider the full lifetime costs of the alternatives.
Life-cycle costing is a standard accounting technique and can readily be applied to PV
systems (Markvart 1999 page 148).

Often the choice is between PV and diesel-fed internal combustion engines. The latter are
typically very cheap to buy but expensive to operate, both in fuel and maintenance. The
operating costs of diesel engines per kWh increase dramatically as the average demand
(kWh/day) is reduced. In contrast, PV costs per kWh are almost constant, and thus, PV
becomes competitive below a certain average demand (kWh/day). Parish (1999) showed
that the crossover point can be between 2.5 and 5 kWh/day, depending on PV costs

24
($/Wp) and diesel costs. A far more detailed study, using data specific to India, showed a
much higher crossover point: between 15 and 68 kWh/day (Kolhe et al. 2002). Both of
these studies include substantial costs for batteries and battery replacement in the PV
systems. For applications not requiring batteries, the crossover point is higher still.
Nonetheless, there is still a crossover point, above which a diesel engine is more
economic, excluding environmental considerations. In summary, PV is most competitive
for systems with modest average demand, and thus PV-RO will be most competitive,
initially at least, in small-scale.

Even using life-cycle costing, the PV often dominates the costs of a proposed system.
This focuses attention on maximising the efficiency of the load to a far greater extent
than might be pursued if operating from grid electricity or diesel. Indeed, much of the
design effort described in this thesis relates to improving the efficiency of the load, in
this case a small RO system.

2.5.2 Materials
Most PV is made of silicon, which itself is very cheap. The cost arises out of the need to
purify it, the manufacture of the cells and their assembly into modules. Three forms of
silicon are widely used: mono-crystalline silicon PV has the highest efficiency and the
highest cost; poly-crystalline silicon PV is in the middle and amorphous silicon PV has
lowest efficiency and least cost. Other materials, such as cadmium telluride, are also
under development.

2.5.3 Watts-peak
The available power from a PV module or array is generally quoted in watts-peak (Wp),
or kilowatts-peak (kWp), and this is the maximum power available under standard test
conditions (STC): 1000 W/m2 irradiance, AM 1.5 spectrum and a cell temperature of
25°C. Under real operating conditions, the maximum power available can be higher than
the quoted watts-peak, but is usually lower, and, of course, varies through the day.

2.5.4 Maximum Power Point Tracking (MPPT)

The current-voltage (I-V) curve and corresponding power-voltage (P-V) curve for a
polycrystalline-silicon PV array operating at standard test conditions are shown in
Figure 2-4.

25
 9

6
Current (A)

0
0 20 40 60 80 100 120 140 160 180
Voltage (V)

1200

1000

800
Power (W)

600

400

200

0
0 20 40 60 80 100 120 140 160 180
Voltage (V)

Figure 2-4 – Typical I-V and P-V curves for a polycrystalline-silicon PV array

26
In order to obtain maximum power from a PV array it must be operated at the peak of the
power curve. In general, the connection of an arbitrary load to a PV array will not draw
the maximum power available. This may be acceptable in small low-cost systems, but for
large arrays, operating close to the maximum power point (MPP) is an economic
necessity.

Furthermore, the natural variation of irradiance and temperature throughout the day and
with passing clouds will cause the I-V curve to vary, and the P-V curve likewise, and
thus, the MPP is not stationary. The control of voltage or current, in order to maintain
operation close to the MPP is known as maximum power point tracking (MPPT) and is
critical to the efficient operation of a PV-RO system, as discussed in Chapter 9.

2.6 Batteries
Rechargeable batteries are widely used in PV systems, mostly for storing the energy
during the day and making it available through the night, but also sometimes for
smoothing out variations due to passing clouds. Most batteries used in PV systems are
lead-acid, because their size and weight can usually be accommodated, and because the
alternative battery technologies (nickel-cadmium and so on) are usually considerably
more expensive. Unfortunately, batteries are notoriously problematic in practice,
especially in PV systems in hot countries:

When there is a problem it is usually to do with the lead-acid batteries (Roberts 1991
page 37).

The disadvantages of using batteries, however, are quite formidable (Green 1995 page
219).

The battery is the weakest part of a stand-alone PV system today (IEA 1999 page 6).

Numerous problems may arise in stand-alone PV systems. Premature battery failure is

among the most common, and is a major concern throughout the PV industry and user
groups. Although they are a relatively low fraction of a system’s initial costs, batteries
can be the most expensive component in the overall life cycle cost of stand-alone PV
systems (Dunlop et al. 2001 page 1).

27
The electricity and telecommunications industries have been seeking reliable and long-
life rechargeable batteries for well over a century (Kapp 1888; via Norris 1999). More
recently, the PV industry and research community have also made considerable effort to
improve the design and operating strategies of batteries for PV applications (Sauer et al.
1997; Dunlop et al. 2001; IEA 2002)

The need to use deep-cycle lead-acid batteries is well known throughout the PV industry
and most professional PV system installers will specify batteries of this type. The
common alternative: car and truck batteries, also known as SLI (starting, lighting and
ignition) batteries, are designed to supply a large current for a short duration, and are
totally unsuited to most PV applications. Despite this, SLI batteries are frequently used in
PV systems in practice, because they are more readily available (Green 2003). SLI
batteries are certainly not suitable for a PV-RO system, and the following discussion
applies to deep-cycle lead-acid batteries.

Battery lifetime in PV systems in central Europe is typically 3 to 8 years (Sauer et al.

1997), but in hot countries, this reduces to typically 2 to 6 years (Gwillim 2001), since
high ambient temperature dramatically increases the rate of internal corrosion. Even in
temperate countries, batteries used for critical telecommunications are often kept in air-
conditioned rooms, simply to extend the battery life. Such luxury is not economic for
batteries in PV systems. Similar degradation can occur in transit and storage, and
batteries may have lost a significant part of their potential lifetime before even being
installed (IEA 1999 page 23). Also, batteries require regular maintenance and will
degrade very rapidly if the electrolyte is not topped up. In practice, these factors combine
and battery replacement costs often exceed original expectations.

Battery efficiency also has a significant cost, since the PV array must be oversized to
cover the energy loss in the batteries. Typical battery efficiency is around 85% (Green
1995), but in hot countries, this reduces to below 75 % (Gwillim 2001).

In many PV applications, the use of batteries is almost unavoidable; hence, the research
into improved batteries specifically for PV. In other applications, however, batteries are
not required, PV-pumping being the obvious example. The requirement of batteries or
not in PV-RO will be discussed in section 2.9.3.

28
2.7 Reverse Osmosis (RO)
Reverse osmosis was outlined fleetingly in the Section 1.2. Here the technology and its
development will be discussed in a little more detail.

The early development of reverse osmosis is described by Merten et al. (1966) and the
mainstream industrial application of the technology is discussed by Byrne (1995). Also,
several of the membrane manufacturers provide comprehensive handbooks (DOW 1995;
Koch 2000; Wagner 2001).

An RO membrane may be thought of, initially at least, as an extremely fine sieve that
allows water to pass through, but not salt. However, microscopic study of an RO
membrane reveals that it is not really a sieve, but rather that the water diffuses through
the membrane while salt cannot. Such membranes are described as semi-permeable and
were developed for the purpose of desalination during the 1950’s and 60’s. RO
desalination was introduced commercially during the 1970’s and now represents over
43% of installed desalination capacity worldwide (Wangnick 2002; via Wiseman 2002).

2.7.1 Membrane materials

The early commercial membranes were cellulose acetate, but since the 1980’s, the
desalination market has been dominated by thin-film composite membranes, comprising
a thin semi-permeable polyamide (nylon) layer supported on a thicker porous
polysulphone backing. The physical strength of the membrane and its support structure is
critical for it to withstand the high pressures required in desalination.

2.7.2 Hollow-fibre permeators

The first commercial desalination membranes were made in the form of hollow fibres.
The saltwater is pumped in at one end of the fibre, freshwater emerges through the
membrane wall of the fibre, and the concentrate comes out of the other end. A large
number of fibres are assembled into a bundle to form a permeator. Hollow-fibre RO
permeators are still manufactured, but the market is now dominated by the spiral-wound
arrangement.

29
2.7.3 Spiral-wound membrane elements
In the spiral-wound arrangement, the membrane is in sheet-form and has the feel of stiff
coated paper. Two sheets are assembled to form a sandwich with a mesh spacer layer in
the middle. The pressurised feed water will be on the outside and the freshwater will pass
through the membrane into the spacer layer. The spacer layer is sealed on three sides and
the fourth is joined to a freshwater collection tube.

In order to save space and cost in pressure vessels, the membrane sandwich is rolled on
to the tube as illustrated in Figure 2-5 to form a membrane element. A second mesh
spacer layer is required to carry the feed water, and, in practice, two or four sandwiches
are wound onto each collection tube, which reduces the distance that the freshwater has
to travel round the spiral into the tube. The feed water is pumped in at one end of the
spiral element and the concentrate appears at the other.

Freshwater
Product
Seawater
Permeate
Feed

Concentrate
Brine

Figure 2-5 – Spiral-wound RO element

RO elements are usually 20, 40 or 60 inches long and 2½, 4 or 8 inch diameter (508,
1016 or 1524 mm long and 64, 102, or 203 mm diameter), the use of inches underlining
the fact that most RO elements are manufactured, or at least designed, in the US. For use,
the membrane elements are loaded into tubular pressure vessels. Typical pressure vessels
hold between one and seven 40-inch elements and can be connected in series and parallel
to achieve the desired plant capacity. Large-scale RO plants have hundreds of pressure
vessels.

30
2.7.4 Fouling, scaling and membrane life
A major consideration in the design and operation of any RO system is the avoidance, or
at least management, of fouling and scaling of the membranes, since this determines the
frequency of required membrane cleaning and replacement. The rate of membrane
fouling and scaling is very dependent upon feed-water quality and pre-treatment, and the
membrane manufacturers’ handbooks all devote substantial sections to these topics.

Scaling, in the context of RO, refers to the precipitation of sparingly soluble salts on the
membrane surface when they become too concentrated. Scaling is commonly a limiting
factor in systems using brackish feed water with high recovery ratios. Whereas with
seawater, the osmotic pressure tends to limit the recovery ratio and scaling is rarely a
major problem. With seawater, the main consideration is biological fouling, caused by
bacteria.

Pre-treatment often includes the addition of chemicals to the feed water that control
fouling and scaling of the membranes. However, there is a trend within the industry to
reduce the use of chemical additives and to pay more attention to the design of the feed
water intake, so as to obtain cleaner water in the first place (Koch 2000).

2.7.5 Temperature effect

The rate of diffusion of water through an RO membrane, and hence the product flow, is
largely dependent on the driving pressure, but temperature is also a very significant
factor. An increase in feed temperature of 4 °C will cause the product flow to increase by
about 10 % (DOW 1995 section 6.7), assuming that other factors are kept constant. But,
in systems employing efficient brine-stream energy recovery, the overall effect of
temperature is greatly reduced.

2.7.6 High-rejection and high-flow membranes

RO membrane manufacturers typically offer seawater RO membrane elements in two
grades: High-rejection elements provide low-concentration product, while high-flow
elements provide a greater volume flow of product, but at slightly increased
concentration. Even with high-flow elements, product concentration is usually perfectly
acceptable in normal seawater desalination applications. The costs for the two grades are
similar.

31
2.8 Brine-stream energy recovery
As noted in the introductory section 1.2, the energy efficiency of seawater RO is heavily
dependent on recovering the energy from the pressurised concentrate (brine). This was
recognised and investigated several years before RO became commercially viable (Bray
1966). With brackish water, much higher water recovery ratios are possible, meaning that
there is much less energy in the concentrate, which makes brine-stream energy recovery
less critical. The following discussion applies mostly to seawater RO.

2.8.1 Pelton wheel

High-
pressure
pump RO membranes
Feed

Motor

Concentrate
Pelton
wheel

Figure 2-6 – Pelton-wheel energy recovery

In large systems, Pelton-wheel turbines are commonly employed. They are simple,
reliable and very well proven in the field, but they are far from perfect. Their efficiency
is usually significantly below what might be expected in a hydropower plant, first,
because they are coupled to the shaft of the main high-pressure pump, which is really too
fast, and second, because the buckets tend to be rough, due to cost-cutting in manufacture
and corrosion in service (Doujak et al. 2003). Furthermore, seasonal variations of flow
and pressure in a RO plant can be significant, due to variations in water demand, feed
water temperature and the condition of the membranes. The efficiency of a Pelton wheel
can be significantly reduced when operating away from the design flow and pressure.
Lastly, the energy that is recovered then has to go back through the main pump, suffering
a further loss before it can usefully be applied to the RO membranes.

In the light of the modest efficiency offered by Pelton wheels in RO systems, several
manufacturers have developed alternative brine-stream energy recovery mechanisms.

32
Healthy competition in this market has led to a steady improvement in the energy
efficiency of seawater RO. In no particular order, the leading technologies are the
Hydraulic Turbo Booster, the Dual Work-Exchanger Energy-Recovery (DWEER) and
ERI’s Pressure Exchanger.

2.8.2 Hydraulic Turbo Booster

Hydraulic Turbo Booster

Pump Medium- High- RO membranes

Feed pressure pressure

Pump Product
Shaft
Motor
Turbine
Concentrate
Discharge

Figure 2-7 – Hydraulic Turbo Booster energy recovery

The Hydraulic Turbo Booster is manufactured by Fluid Equipment Development Co.

([Link] accessed: March 04). It comprises a single-stage radial
inflow turbine and a single-stage centrifugal pump on the same shaft, but totally
independent of the motor-driven pump. This independence allows the rotational speed of
the Turbo Booster to be chosen to optimise its efficiency rather than being fixed by the
electric motor. It also facilitates throttling arrangements: control valves that are used to
fine-tune membrane pressure. Furthermore, the Hydraulic Turbo Booster can discharge
the concentrate at above ambient pressure.

2.8.3 DWEER Work Exchanger

The Dual Work-Exchanger Energy-Recovery (DWEER) is manufactured by
DesalCo Ltd. ([Link] accessed: March 04). It comprises two cylinders, each
with a free piston. During one stroke, high-pressure concentrate water pushes the piston
in one of the cylinders, which pressurises the feed water on the other side of the piston.
At the same time, low-pressure feed water pushes the piston in the other cylinder, which
pushes the old concentrate out of the system. On the other stroke, the pistons reverse
roles. Automatic valves are used to control and coordinate the flows into the two
cylinders, and their design is central to the DWEER technology.

33
 High-
pressure
pump RO membranes

Motor Boost
pump
Feed

Concentrate

Figure 2-8 – DWEER Work Exchanger energy recovery

Notice, in Figure 2-8, that an additional (motor-driven) pump is required to make up for
the small pressure loss that occurs to the concentrate in the membranes and the work
exchanger.

An excellent review of large-scale work-exchanger energy recovery mechanisms, leading

up to and including the DWEER, is presented by Andrews and Laker (2001).

2.8.4 ERI’s Pressure Exchanger

Low-pressure
concentrate
High-pressure
concentrate
Rotation

Low-pressure
feed
High-pressure
feed

Figure 2-9 – ERI’s Pressure Exchanger

The Pressure Exchanger (MacHarg 2001) is manufactured by Energy Recovery Inc.

([Link] accessed: March 04). It is a form of work exchanger, but has
twelve cylinders with no pistons. The cylinders are co-axial around the circumference of

34
a ceramic rotor, rather like the holes in the magazine of a revolver. As with other work
exchangers, high-pressure concentrate water pushes pressurised feed water through one
cylinder, while low-pressure feed water pushes the old concentrate out of another. The
absence of a piston allows the water to mix a little, which increases the concentration of
the feed slightly but is acceptable in moderation. The automatic valve gear required in
other work exchangers is replaced by the rotation of the cylinders past stationary inlet
and outlet ports. The rotation is powered by the flow of water through the device, and the
speed of rotation is critical to minimising the mixing.

2.8.5 Relative efficiencies

The relative efficiencies of the four energy recovery mechanisms just described are the
subject of much debate. It is meaningless to compare the efficiencies of the mechanisms
themselves since their outputs have different forms: The Pelton-wheel gives mechanical
torque. The Hydraulic Turbo Booster raises the pressure of an arbitrary flow of water.
The DWEER raises the pressure of a flow of water equal to that of the concentrate, and
requires a further pump. ERI’s Pressure Exchanger is similar, but involves some mixing
of the flows.

In order to compare the efficiencies at all, the mechanisms must be considered within an
RO system. The difficulty then is that parameters and efficiencies of all the other system
components come in to play and, depending on these, any of the energy recovery
mechanisms can appear in a favourable light (Andrews, Pergande et al. 2001; Oklejas
2002; MacHarg 2002). Losses in control mechanisms such as throttling valves and
variable speed drives also need to be accounted for, as do seasonal variations in operating
points (Manth et al. 2003).

2.8.6 Vari-RO
The Vari-RO combines water hydraulics with oil hydraulics. Three pistons are used to
achieve the high-pressure pumping of the feed. These are driven by both the high-
pressure concentrate and the oil hydraulics, which is powered by a motor. Computer
controlled valves direct the high-pressure concentrate between three different cylinders.
Importantly, they open and close at zero flow in order to minimise transients. A
prototype was demonstrated (Childs et al. 1998; Childs and Dabiri 1999) and further
developed for solar application (Childs, Dabiri et al. 1999; Childs et al. 2000).

35
2.8.7 Energy recovery in small-scale RO
Small reverse-osmosis systems are often built without any energy recovery mechanism.
They have a manually-operated needle valve or pressure-operated relief valve to control
the back-pressure in the concentrate. This keeps the capital cost down but is very
wasteful of energy. Typically, 70 % of the input power is wasted in the valve and,
consequently, such systems often consume more than 10 kWh/m3, making them very
expensive to run.

Turbines tend to have poor efficiency at small sizes. Gwillim (1996) looked into the
possibility of using a Pelton wheel for energy recovery in a 3-m3/d seawater RO system.
It needed a jet size of less than 1 mm acting on a wheel of diameter ~300 mm. High
windage losses were expected, together with high manufacturing costs, and in general,
the approach was considered impractical.

For energy recovery in small-scale RO systems, positive-displacement devices generally

offer much higher efficiencies. The DWEER Work Exchanger and ERI’s Pressure
Exchanger are positive-displacement devices but are not currently manufactured at small
sizes. ERI did produce some small-scale Pressure Exchangers but found that they were
prone to fouling (see also Kunczynski 2003); presumably this is less of problem in larger
Pressure Exchangers with larger clearances. Other than ERI’s Pressure Exchanger, most
positive-displacement energy recovery mechanisms involve high-pressure valves that
need to operate with precise timing to ensure a smooth flow of water. Combining this
precision with corrosion-tolerance of concentrated seawater is very challenging, and
many design concepts have failed in practice (Andrews and Laker 2001).

2.8.8 Energy recovery pumps

Integrated designs, combining brine-stream energy recovery with positive-displacement
pumping, were developed by Bowie Keefer during the 1980’s. He patented a hand-
operated pump with energy recovery for application to seawater RO (Keefer 1980). The
patent also suggested that the device could be operated by the reciprocating action of a
traditional water-pumping wind turbine.

Later, he patented a shaft driven energy-recovery pump (Keefer 1984). This was based
on a standard plunger pump, but with positive-displacement energy-recovery added
between the crank assembly and the plungers. The plungers also served as spool valves

36
for the energy-recovery. Prototypes were built and demonstrated very good energy
efficiency, including some that were operated from PV (Doman et al. 1982; Keefer et al.
1985). This work showed great promise, but unfortunately was not continued; perhaps
the cost of manufacture was high.

2.8.9 Hydraulic motor

Oil-driven hydraulic motors are very widely used in many industrial applications; they
are very efficient and very well-proven. Water-driven hydraulic motors are relatively
new, but are used for example in the food industry, where risk of hydraulic oil leakage is
not acceptable. Danfoss manufacture a range of small water-driven hydraulic motors for
such applications. The Danfoss motors have axial pistons acting on a swash plate to
rotate the shaft. They are lubricated by the driving water.

Dulas Limited (Gwillim 1996) demonstrated the use of a Danfoss hydraulic motor for
energy recovery in a small seawater RO system. This reduced the specific energy
consumption from 13 kWh/m3, for a system using a needle valve, to around 5.6 kWh/m3
– a vast improvement in the context of PV-RO. The work described in this thesis was
founded on the success at Dulas and, in the early stages, employed a similar Danfoss
hydraulic motor. However, testing at CREST indicated slightly lower efficiencies and
corrosion problems, as discussed in Chapter 3, and the hydraulic motor was dropped
from the design, in favour of the Clark pump. In contrast, Kunczynski (2003), who is
also developing PV-RO, was using Clark pumps but has now switched to a hydraulic
motor.

2.8.10 Clark pump

The Clark pump is a little like the dual work exchanger (section 2.8.3), except that the
two cylinders are in-line and the two pistons are connected by a rod. The rod creates a
difference in the effective areas on the two sides of each piston, which allows the relative
flows to be adjusted by design. Furthermore, having the two pistons connected allows
energy from the feed to be added to that of the concentrate, yielding an output pressure
higher than that of the concentrate. Hence, the Clark pump is sometimes described as a
pressure intensifier. A further description of its operation is presented below, but it is
perhaps best understood from the animation at:
[Link] (accessed: March 04).

37
 a) Cylindrical
housing Connecting rod

Pistons

b) Discharge
Concentrate

Medium High
pressure
pressure
feed

c) Concentrate Discharge
Medium
High pressure
pressure
feed

Figure 2-10 – Basic mechanics of a Clark pump

The basic mechanics of a Clark pump are shown in Figure 2-10a. The two pistons are
solidly connected by the rod, and this assembly reciprocates inside the cylindrical
housing. In Figure 2-10b the medium pressure and the concentrate pressure both act to
push the piston assembly to the right, thus driving the high pressure, as shown. At the
end of stroke, an internal mechanism reverses the ports, as shown in Figure 2-10c, and
the piston assembly travels back to the left, until it again reverses.

The general arrangement of the two pistons and the rod, and its application to RO
systems, was presented in patents many year ago (Pinkerton 1979; Wilson 1983). But, it
was not until a practical valve-operating mechanism was developed and patented by
Clark Permar (hence the name Clark pump) that the arrangement became commercially
viable (Permar 1995). Permar licensed the design to Spectra Watermakers Inc. who

38
developed a commercial product during 1997. It was aimed at the yachting market and
was sized to suit a single 2.5 by 40-inch spiral-wound seawater RO element.

Clark pump
Pump Medium- High- RO membranes
Feed pressure pressure
Product

Motor

Concentrate
Discharge

Figure 2-11 – Simple configuration of a Clark pump in an RO system

The simplest configuration of a Clark pump in an RO system is shown in Figure 2-11. It

requires only one motorised pump and no pressure regulating mechanism: the water
recovery ratio (product flow to feed flow) is fixed by the ratio of the cross-sectional area
of the rod to that of the piston. This simple configuration is marketed by Spectra and
achieves a specific energy consumption as low as 3.2 kWh/m3 for seawater at
35,000 ppm, 25 °C (Smith 2000), which is excellent for a small system.

CREST obtained a Clark pump in August 2000 and tested its performance thoroughly. Its
energy efficiency is excellent (see section 5.2) and, importantly, this is maintained over a
very wide range of flow and pressure. The Clark pump is central to the PV-RO system
presented in this thesis.

2.9 Photovoltaic-powered Reverse Osmosis (PV-RO)

As discussed in sections 2.2 to 2.4, photovoltaic-powered reverse osmosis (PV-RO) is
considered one of the most promising forms of renewable-energy-powered desalination.
And, as discussed in section 2.5.1, PV is currently most competitive in small-scale
systems. Thus, small-scale PV-RO has received much attention in recent years and
numerous demonstration systems have been built.

2.9.1 Brackish-water PV-RO systems

Brackish water has a much lower osmotic pressure than seawater and therefore its
desalination requires much less energy, and therefore a much smaller PV array in the

39
case of PV-RO. Also, the lower pressures found in brackish-water RO systems permit
use of low-cost plastic components. Thus, the total cost of water from brackish-water
PV-RO is considerably less than that from seawater, and systems are starting to be
offered commercially.

Feed
Capacity PV Batteries
Location References Water
(m3/d) (kWpeak) (kWh)
(ppm)
Sadous,
(Alawaji et al. 1995)
Riyadh, 5800 15 10 264
(Hasnain et al. 1998)
Saudi Arabia
3.5 plus
Haifa, Israel (Weiner et al. 2001) 5000 3 36
0.6 wind
Elhamrawien, [Link]/[Link]
3500 53 18 200
Egypt (accessed: March 04)
Heelat ar
Rakah, (Al Suleimani et al. 2000) 1000 5 3.25 9.6
Oman
White Cliffs,
(Richards et al. 2003) 3500 0.5 0.34 NONE
Australia
(Mathew et al. 2001)
Solarflow, (Maslin et al. 2003)
5000 0.4 0.12 NONE
Australia [Link]
(accessed: March 04)
Table 2-2 – A selection of existing brackish-water PV-RO systems

Many of the early PV-RO demonstration systems were essentially a standard RO system,
which might have been designed for diesel or mains power, but powered from batteries
that were charged by PV. This approach tends to require a rather large PV array for a
given flow of product, due to poor efficiencies both in the standard RO systems and in
the batteries. Large PV arrays and regular replacement of batteries would tend to make
the cost of water from such systems rather high.

The Solarflow system was developed at Murdoch University, Australia (Mathew et al.
2001). It has no batteries and the pump has been designed specifically for the application.
The water recovery ratio is fixed at 16% or 25%, which is very low compared to most
brackish-water RO systems. This greatly reduces scaling potential, which is a very
important consideration with brackish water, less so with seawater. However, the low
recovery ratio also gives a much higher flow of concentrate, and therefore energy in the
concentrate. Thus, brine-stream energy recovery becomes worthwhile, which is unusual
in brackish-water RO. The Solarflow system has energy recovery integrated into the

40
main motorised pump, along the lines of Keefer’s designs mentioned in section 2.8.8.
Furthermore, the Solarflow system requires no batteries, see section 2.9.3.

2.9.2 Seawater PV-RO demonstration systems

The osmotic pressure of seawater is much higher than that of brackish water, and
therefore its desalination requires much more energy, and, unavoidably, a rather bigger
PV array. Also, the higher pressures found in seawater RO systems require use of
mechanically stronger components. Thus, the total cost of water from seawater PV-RO is
likely to remain higher than that from brackish-water, and systems have not yet passed
the demonstration stage.

Feed
Capacity PV Batteries
Location References Water
(m3/d) (kWpeak) (kWh)
(ppm)
Lampedusa, (Sardi 1996; via
Seawater 40 100 880
Italy CRES 1998)
CRES,
4 plus 0.9
Laviro, (Tzen et al. 2003) 36,000 <1 44
wind
Greece
(Herold et al. 1998)
ITC, Canaries
(Herold et al. 2001) Seawater 3 4.8 19
(DESSOL)
(Espino et al. 2003)
La Paz,
(Kunczynski 2003) 32,600 4 to 19 5 to 25 Yes
Mexico
Table 2-3 – A selection of seawater PV-RO demonstration systems

The Lampedusa seawater PV-RO demonstration was a flagship project commissioned in

1990, with a very large PV array and battery bank. It includes an energy-recovery turbine
of modest efficiency, and achieved a specific energy consumption of around 5.5 kWh/m3.
Since 1995, the PV has been supplemented by mains electricity.

The CRES and ITC projects are current. Neither have energy-recovery.

The project at La Paz is particularly interesting (Kunczynski 2003). It has included trial
of three energy-recovery mechanisms: a small ERI’s Pressure Exchanger, several Clark
pumps and, most recently, the Danfoss hydraulic motor (Danfoss 2002). The test site has
a large PV array and battery bank and the PV array size shown in Table 2-3 is based on
the measured electrical power consumption of the RO system.

All known seawater PV-RO demonstrations employ batteries.

41
2.9.3 Batteries or not
As discussed in section 2.6 batteries are a necessary evil in many stand-alone PV
systems. They are almost unavoidable in some applications, but anyone familiar with
battery performance in real PV systems appreciates the great benefit of eliminating them
wherever possible.

On the other hand, anyone familiar with RO desalination equipment appreciates the
benefit of operating at constant flow 24-hours-a-day. Obviously, it maximises production
and makes best use of the invested capital, but also, it maintains the quality of the
product water and helps in the management of membrane fouling.

The reduction of product water quality (increase in salt concentration) may well be
perfectly tolerable in some situations, see section 2.1.3, and, otherwise could be
addressed through use of high-rejection (lower flow) membranes, as discussed in
section 2.7.6, or diversion of the lower quality product, as discussed in section 10.6.

Thus, the main concern is the possible increase in membrane fouling caused by
intermittent and variable-flow operation. As noted in section 2.7.4, the critical factor in
membrane fouling is the quality of the feed water, which is very site specific and
dependent on the intake arrangement. The vast majority of recorded experience of RO
membrane fouling is with 24-hour-a-day operation. Systems that are operated
intermittently, such as those on boats and ships, tend to be small and unrecorded. Hence,
there is little data available to quantify any increase in membrane fouling.

Likewise, data quantifying battery failure in remote areas is very limited. And thus, the
choice of using batteries or not within a PV-RO system depends on the relative lifetimes
of batteries and membranes, neither of which is well quantified.

Kunczynski (2003) tried intermittent operation, but now favours using batteries. Most
other PV-RO systems rely entirely on batteries. The notable exceptions are the two
Australian brackish water systems shown in Table 2-2.

Abufayed (2003) reported on a 10,000 m3/d seawater RO plant in Tajoura, Libya that
was operated intermittently for over a year: Cyclic operation with variable length
operate/shut-down periods had no discernable quantitative adverse effects on plant
efficiency.

42
The University of Las Palmas de Gran Canaria recently set up a small RO rig for testing
under variable conditions and have experienced no deterioration after 7000 hours of
operation (Gotor et al. 2003). In more limited testing (Gocht et al. 1998), nothing was
observed ruling out a transient operation of the RO.

These points perhaps support the decision made by CREST and Dulas Limited in 1998 to
pursue the development of a batteryless PV-RO system.

2.9.4 Applications
The existing PV-RO demonstrations typically produce between 1 and 100 m3/d, and, as
noted at the beginning of this section 2.9, future systems are likely to be in the same
range, initially at least.

At this scale, PV-RO is competing mainly with diesel-powered RO and water transported
by tanker. The cost of diesel versus PV was discussed in section 2.5.1, although the
comparison may be altered by the fact that a RO unit in a diesel-powered system will
tend to be designed to minimise capital cost and therefore be less efficient than one
designed to operate from PV. The cost of transported water is very site-specific, but a
couple of examples are $5-6/m3 for supply by road-tanker to remote parts of Brazil
(Laborde et al. 2001) and $3.75/m3 for supply by boat to small Greek islands (Avlonitis
et al. 2003).

PV is already used to good effect in many developing countries, particularly for vaccine
refrigeration, water pumping, lighting and communications. PV-RO will be added to this
list, once its cost has reduced sufficiently, which may be fairly soon in the case of
brackish-water PV-RO (Maslin et al. 2003). Seawater RO, the subject of this thesis,
requires significantly more energy and the cost of PV may have to reduce a little further
before seawater PV-RO can be widely applied in developing countries. The early
adopters of PV-RO may well be in the luxury market: hotels or prestige dwellings in
remote areas or on islands, particularly those keen to protect their environment.

43
 Chapter 3 Initial test rig hardware and results

The work described in this thesis was founded on a previous project conducted by Dulas
Limited (Gwillim 1996). In that project, Dulas had demonstrated the use of a Danfoss
hydraulic motor for energy recovery in a small seawater RO system. This had reduced
the specific energy consumption from 13 kWh/m3, for a system using a needle valve, to
around 5.6 kWh/m3 – a vast improvement in the context of PV-RO. The Dulas test rig
was re-commissioned and transported to CREST during 1999.

Mains Inverter
Pressure 2½-inch by 40-inch
Relief
Valve Reverse-Osmosis Modules
Induction
Motor
Pulsation
Dampener
Cartridge Captive
Filter Acceleration
Tube
Plunger
Pump

Danfoss
Hydraulic
Concentrate Motor
Mains Discharge

Induction Freshwater product

Motor

Centrifugal “Seawater”
Pump Intake Tank

Figure 3-1 – Configuration of the initial test rig using a Danfoss hydraulic motor

Starting from the tank at the bottom of Figure 3-1, a constant-speed centrifugal pump
raises the pressure to around 2 bar, at the inlet of the cartridge filter. The plunger pump
operates at variable-speed, by virtue of the inverter, and raises the pressure to between
~40 and 69 bar, which feeds the RO modules. The high-pressure concentrate from the
RO modules passes through the hydraulic motor and the energy is returned to the shaft of
the plunger pump through the pulleys and a toothed rubber belt. Since both the plunger

44
pump and the hydraulic motor are positive-displacement devices, the water recovery ratio
is fixed by the ratio of the pulley sizes. Thus, the recovery ratio could readily be adjusted
on-site during commissioning to suit a particular feed water, or perhaps on a seasonal
basis to allow for changes in the water temperature. Not all positive-displacement brine-
stream energy recovery mechanisms have this flexibility.

The captive acceleration tube and the pulsation dampener, shown either side of the
plunger pump in Figure 3-1, serve to smooth out the ripple in the flow created by the
pump itself. The pressure relief valve is a safety mechanism.

3.1 Operational results

Surprisingly, the original membranes, which had been left unused and neglected for 3
years, were still functional and capable of producing water at less than 300 mg/L TDS,
albeit at very reduced flow. It was decided that they should be used for the initial testing,
particularly that of the other components and instrumentation.

The pulleys were configured to give a recovery ratio of around 10 % and the inverter was
used to adjust the speed, between zero and maximum in a total of 38 steps. The feed
flow, feed pressure and product flow all varied in keeping with the speed and no
problems were experienced with the variable operation of the RO membranes. Details of
the tests and results are provided in Appendix A and a summary is shown in Table 3-1.

Electrical input power (W) 502 997 1515

Speed (rpm) 494 864 1130
Electric motor torque (Nm) 6.9 8.7 10.3
Feed pressure (bar) 38.7 58.9 68.9
Concentrate pressure (bar) 37.7 56.4 65.1
Feed concentration (mg/L) 29000 29000 29000
Product concentration (mg/L) 550 283 262
Feed flow (L/s) 0.134 0.233 0.300
Product flow (L/s) 0.0075 0.022 0.0291
Specific energy (kWh/m3) 18.6 12.6 14.4

Table 3-1– Measurements from the initial test, using very old membranes

The high specific energy consumptions shown in Table 3-1 are primarily due to the use
of very old membranes. Despite this, it was very instructive to consider exactly where the
energy was being consumed.

45
3.2 Sankey diagram
Details of power flow and efficiency calculations are shown in section A.3 of Appendix
A and are summarised here in the form of a Sankey diagram.

Electrical Inverter and

input motor Motor Pump Main
997W combined shaft shaft pump Feed RO membranes and pipework
79% 787W 1614W 85% 1372W
Desalination
51W

210W Throughflow
242W Crossflow and loss 78W
high-pressure
pipe losses
53W
Concentrate
Recovered 1190W
827W

Toothed Hydraulic
belt motor
95% 871W 73%

44W
319W

Figure 3-2 – Sankey diagram showing power flows in the initial test rig, using very
old membranes

The line widths shown in Figure 3-2 represent the power flows corresponding to the
middle column in Table 3-1. These powers can readily be converted to specific energy
consumption (kWh/m3) simply by dividing by 79, which is the product flow in L/h.

Starting on the left, 997 W is drawn from the mains supply. The combined losses in the
inverter and the motor are 210 W, meaning that their combined efficiency is around 79%,
which is as expected for a motor of this size.

The plunger pump is 85% efficient, which gives a loss of 242 W. Two losses are shown
in relation to the RO membranes themselves. First, the 53 W represents the slight
pressure drop found between the feed and the concentrate flows, and includes pressure
drops in the interconnecting pipe work. Second, the 78 W represents the viscous losses
associated with driving the product water through the membrane. The 51 W, on the far
right of the diagram, represents the actual desalination power. (51/79 = 0.65 kWh/m3,

46
which is a little lower than the theoretical minimum energy figures presented in section
2.3.1, because the concentration of the feed is a little lower.) The massive 1190 W, also
emanating from the membranes, represents the energy contained in the pressurised
concentrate water. The hydraulic motor recovers part of this energy and delivers it,
through the toothed belt, back to the shaft of the pump.

The Sankey diagram, Figure 3-2, shows that the hydraulic motor is supplying more than
half of the shaft power to the main pump. Without the hydraulic motor, the electric motor
would have to supply the whole of the shaft power, and the electrical consumption would
be more than double: ~2045 W, assuming the same motor efficiency. Rephrasing this, the
hydraulic motor is reducing the energy consumption by 51%. The previous testing at
Dulas (Gwillim 1996) had shown a reduction from 13 kWh/m3 to around 5.6 kWh/m3: a
reduction of 57%.

3.3 Danfoss hydraulic motor

At this stage, early 2000, the Danfoss hydraulic motor appeared to be a good choice as an
energy recovery device for a seawater PV-RO system. However, it was soon to be
dropped from the design in favour of the Clark pump. Given that Kunczynski (2003) has
recently switched in the other direction, our experience with Danfoss hydraulic motors is
reviewed here.

As noted in section 2.8.9, water-driven hydraulic motors are relatively new, and were
developed for use in the food industry for example, in which case the driving water is
freshwater. For use in RO, as a brine-stream energy recovery mechanism however, the
driving water is concentrated seawater, which is very corrosive.

The first Danfoss hydraulic motor, used at Dulas 1996, was a MAH-10. It did not survive
the three-year standstill and was found to be totally unserviceable in 1999. The main
high-pressure pump, a CAT 317, had endured similar abuse without ill effect. CAT
pumps are widely used in small-scale RO systems, and are fully seawater compatible.

A new hydraulic motor, a MAH-5, was purchased from Danfoss in 1999, and at the time,
Danfoss claimed that the design had been improved to give a greater resilience to
corrosion. However, in mid 2000, after standing for approximately six weeks, some
distinctly rust-coloured water emerged from the unit. Several days later, during routine

47
tests, the rig emitted sounds suggestive of an imminent component seizure and was
immediately stopped. Turning the shaft of the hydraulic motor by hand strongly
suggested that it was the cause – the torque was uneven and there was a noticeable
grating in places. Nonetheless, the rig was restarted and, surprisingly, the problem
disappeared. In conclusion, the long-term compatibility of the Danfoss hydraulic motor
with concentrated seawater was in doubt.

Perhaps related to this, the hydraulic motor was occasionally reluctant to start. Because
the hydraulic motor is coupled to the main pump via the toothed belt, and because
initially there is no water pressure, the electric motor has to start both the pump and
hydraulic motor. Any stiction in the system could well be a problem in a batteryless PV-
RO system, particularly for early morning starting when the power available from the sun
is low. A ratchet coupling could be employed but this would add mechanical complexity.

The Sankey diagram, Figure 3-2, shows the hydraulic motor operating with an efficiency
of 73%. The toothed belt shows 95% and the main high-pressure pump 85%. Thus, the
water-to-water efficiency of the complete arrangement is only 59%. Furthermore, when
operating at reduced speed and pressure, as in the first columns of Table 3-1 and
Table A-1, the efficiency of hydraulic motor reduced from 73% to 56%, giving an overall
water-to-water efficiency of only 45%. In-house testing of the Clark pump, see section
5.2, showed water-to-water efficiency above 90% across a very wide range flow and
pressure.

Danfoss now offer a new range of axial piston pumps, the APP range (Danfoss 2002),
which are designed for use in seawater RO and can serve as hydraulic motors. These are
the units now in use by Kunczynski (2003).

48
Chapter 4 Instrumentation and data-acquisition system

The early testing described in the previous chapter and in Appendix A also indicated the
importance of accurate and reliable instrumentation and data-acquisition. In order to
assess the energy efficiencies of the various system components, with sufficient accuracy
to compare different system configurations, accurate measurements of flows, pressures,
speeds, torques and so on are essential. Furthermore, in order to assess efficiencies over a
broad range of operation, a computer-based data-acquisition system is required. The
quality of the component models discussed in the next chapter is totally reliant on the
quality of data from which they are constructed.

Instrumentation and data-acquisition is a major branch of science and engineering in its

own right and this chapter is not intended to introduce the subject. Rather, it is intended
to pass along particular experience related to the instrumentation of a small RO rig and to
indicate the basis of data used in the modelling described later.

4.1 LabVIEW
LabVIEW software from National Instruments is perhaps the industry standard data-
acquisition package. Several alternatives were considered at the start of the project in
1999, but overall it seemed that LabVIEW would be the most suitable and it was
selected. Four years on, and with much experience in using LabVIEW, the choice still
seems good.

LabVIEW is a graphical programming language, very different from the text-based

languages that most programmers are familiar with. The data appears to flow
simultaneously through different parts of the application program, and this, coupled with
the fact that data-acquisition is inherently time-critical, can be off-putting, even to well-
experienced text-based programmers. Data types (integers, reals, booleans, arrays and so
on) are represented by line colours and thicknesses, and can seem awkward to
programmers used to less strongly-typed languages such as Matlab, Basic or Perl. As
with all programming languages, there are innumerable ways of tackling a particular
problem. Used elegantly, LabVIEW can deliver impressive functionality with minimal

49
programming. Conversely, using LabVIEW as if it were text-based language leads to
very clumsy and limited programs. National Instruments provide excellent
documentation both for LabVIEW itself and for data acquisition in general. Careful study
of this documentation has proved a very good investment of time throughout the course
of this project.

4.2 Data acquisition hardware

Data acquisition cards that fit internal PC-computer slots provide cost-effective high-
speed multi-channel analogue inputs. Initially an Advantech PCL-818HG card was used,
with appropriate driver software to link it to LabVIEW. Various technical annoyances
were encountered and, when more channels were needed, the card was replaced with a
National Instruments PCI-6024E. Later, a second 6024E card was added and the two
have performed impeccably throughout.

The 6024E has 16 single-ended analogue inputs, suitable for measuring voltages in
ranges ±0.05 V to ±10 V. The strain gauges used within the pressure sensors and torque
load-cells, discussed later, have a very low voltage output and require differential inputs.
To accommodate these, National Instruments strain-gauge signal-conditioning boards
SC-2043-SG are used. These convert 8 of the single-ended inputs on each of the 6024E
cards into differential inputs with 10 times greater sensitivity. Again, the National
Instruments hardware has performed impeccably throughout.

4.3 Acquiring data into LabVIEW

It was initially considered that acquiring data at a rate of 1 Hz would be perfectly
adequate for the measurement of flows, pressures and so on in an RO rig. This would be
mostly true, if the voltages from the sensors were clean DC signals. In the presence of
power-electronic inverters, however, and even with diligent application of good
distancing, screening and earthing practices, the signals suffer from significant electrical
noise. This could be filtered in hardware, but was more readily dealt with in software.
The data acquisition rate is set to 1 kHz and each 1000 samples is averaged in the
software to give high-quality data at 1 Hz. A modest Pentium II PC handles these
calculations with ease, for all 32 channels.

50
4.4 Signal processing
Once in LabVIEW, the voltage data for each channel is offset, scaled, linearised and so
on to convert it to real-world units of flow, pressure and so on. In practice, it was found
convenient to perform most of these calculations at 1 kHz, before any averaging. This
allows the high-speed data to be viewed on-screen in real-world units, and again there is
no problem with computation speed.

Numerous further calculations were implemented in LabVIEW, for example calculating

the efficiencies of the various motors, pumps and so on. Viewing this data in real-time
provides immediate indication of sensor failure or calibration problems.

4.5 Pulse inputs

The flow meters and speed sensors, described later, provide pulses to the data-acquisition
system and were initially connected to dedicated hardware counter inputs of the 6024E
cards. There are two ways to use such counter inputs to measure frequency.

The first, and most obvious, is to count the number of pulses over a fixed period of time;
in our case, the obvious choice is 1 second. However, this gives very coarse results: for
example, if the frequency is 9.9 Hz, the data will appear as 10 10 10 10 10 10 10 10 10 9
10 10 10 10 10 10 10 10 10 9 etc, which is correct on average, but is not good for
viewing any dynamic effects.

The second approach is to measure the duration of one pulse. This is achieved by using
the pulse signal to gate a high-frequency signal of known frequency. Counting the gated
high-frequency pulses and taking the reciprocal gives the frequency of the unknown low
frequency. This approach also gave problems, particularly when measuring frequencies
below 2 Hz, because there are not necessarily two rising edges in a 1-second
measurement period. This means that the data from the counter channels arrives in
LabVIEW asynchronous to the other data on the analogue channels.

An additional, and unexpected, problem with counter channels was caused by electrical
noise. Inspection of the pulse waveform on a fast oscilloscope showed a finite rise time
as expected. It also showed that noise, during the rise, could cause the data-acquisition
system to see two rising edges. Using the first method of frequency measurement
described above, this led to an occasional extra count, which could easily have gone

51
unnoticed. Using the second method, the extra pulses were very prominent, because their
short duration showed up as a ridiculously high frequency.

To sidestep the complications described above, the sensors with pulse outputs were
connected instead to analogue inputs, along with the true analogue signals. Since these
channels are sampled at 1 kHz, as mentioned previously, it is straightforward to locate
the rising edges, with 1-ms precision, in LabVIEW. From this data, the software
calculates the average frequency for every pulse and then rasterises this to give analogue
data at 1 kHz. This data shows discrete steps at the pulse edges, but is generally very
smooth compared to that available from the hardware counter.

The software approach also gets round the problem of measuring frequencies below
2 Hz, because it can wait for the next rising edge, which is achieved by buffering the data
on the other channels, thus ensuring that final data is in perfect synchronism. This proved
useful for viewing flow-pressure transients.

4.6 Data logging

The data is logged to disk, usually at 1 Hz, though several other rates have been used for
particular investigations. The files are usually .csv and can be opened directly in either
Excel or MATLAB.

4.7 Control
The data-acquisition system is also used to provide control of the RO system under test.
In particular, it controls the motor speeds and the heating and cooling of the water. For
many tests, simple sequencing programs were written to step through an array of test
points automatically.

4.8 User interface

LabVIEW facilitates the construction of impressive user interfaces, and this has been
exploited to provide real-time graphic display of the system variables and to allow
manual control of the motor speeds and so on.

52
4.9 Flow measurement

4.9.1 Turbine flow meters

The initial test rig was fitted with turbine flow meters for the feed and product, as
detailed in Appendix B, section B.1. These were carefully calibrated using a bucket and
stopwatch, but repeating this calibration revealed drift of 4% and 13% respectively.
Calibration drift of this magnitude renders the data useless for the study of efficiencies
etc., and so the turbine flow meters were replaced with oval-gear flow meters.

4.9.2 Oval-gear flow meters

Three oval-gear flow meters are fitted to the new test rig. They measure the Clark pump
feed flow, the plunger pump flow and product flow respectively. They are manufactured
by Titan Enterprises Limited and offer a nominal accuracy of 0.5%. Titan provide
traceable calibration certificates for each meter, as shown in Appendix B, section B.2,
and claim repeatability of +/– 0.1%. Marcos Miranda used the data from the calibration
certificates to create a polynomial for each meter and implemented these in LabView.
Thus, the final accuracy is believed to be better than the nominal 0.5% and no further
calibration was considered necessary.

4.10 Pressure sensors

The pressure sensors used are silicon-diaphragm type, PDCR 800 series, manufactured
by Druck Limited. The datasheet (Druck 1997) claims a combined non-linearity,
hysteresis and repeatability of ± 0.1% best-straight-line and a temperature effect of
± 0.5% within the range 0 to 50 ºC.

4.10.1 Medium-pressure sensor

For the medium-pressure feed to the Clark pump, a PDCR 810-0799 sensor rated at
15 bar is used. Its calibration sheet is in Appendix B, section B.3 and indicates a full-
range output of 100.24 mV. However, the LabView data-acquisition system was set to
the nominal specification value of 100.00 mV, increasing the overall error to around
0.25%.

53
4.10.2 High-pressure sensors
For the high-pressure feed to the RO modules and for the concentrate, two PDCR 810
sensors, rated at 60-bar, are used. Using them up to 69-bar is considered acceptable,
although linearity may be slightly impaired. No calibration sheets were available for
these sensors so they were checked, at least in their low range, by comparison against the
medium-pressure sensor. This showed excellent agreement with the nominal
specification and no further calibration of the absolute measurements was considered
necessary. They were also checked against each other over the full range and this
revealed a discrepancy of 0.44% at full range. In order to maximise the accuracy of the
small delta pressure, which is calculated from the difference of the large high-pressure
and concentrate pressure, this discrepancy was removed by adjusting one sensor up
0.22% and the other down by the same amount.

4.11 Concentration measurement

The concentrations of both feed and product water are calculated from conductivity
measurements. In-line conductivity probes and industrial signal-conditioning units were
fitted by the original manufacturers of the initial test rig. These had three problems.

First, the conductivity probes were wrongly orientated and poorly positioned in the pipe
work. This can lead to lack of flow through the probe (dead space) and trapped air within
the probe. Certainly, the data became much more consistent after the probes were refitted
correctly.

Second, the 4-20-mA outputs from the signal-conditioning units were neither floating nor
properly grounded. This gave errors when connected to single-ended inputs on the data-
acquisition card. Connection to differential inputs seems extravagant for 4-20-mA
signals, but did solve the problem.

Third, the temperature compensation included in the industrial signal-conditioning units

was not accurate. To solve this it was disabled, and temperature compensation was
implemented more accurately in LabVIEW.

With hindsight, it may have been better to connect the conductivity probes directly to the
data-acquisition system. This would require a low-voltage AC supply to the probes to
prevent electrolysis; the industrial units use 1 kHz.

54
The concentration is calculated from the conductivity in LabVIEW using a polynomial
derived from data from DOW (1995 section 10.2), see Appendix B, section B.4.

For initial calibration, fixed resistors were used in place of the conductivity probes. For
final calibration, water samples are periodically compared with accurately made
solutions, and normally show good agreement.

4.12 Temperature measurement

Initially, a thermocouple temperature sensor was used, but this was very prone to
electrical noise from the power-electronic inverters. It was replaced by a thin-film
platinum element probe (RS part number: 237-1663), which provides a far better signal
and is calibrated to BS 1904, Class B, DIN437860, 100Ω±0.1Ω at 0°C. This translates to
±0.3 °C and may be expected to rise to ±0.4 °C at 50°C (RS datasheet 249-4930).

National Instruments sell signal-conditioning units specifically for platinum resistance

temperature sensors, but, to save cost, the sensor was connected in a Wheatstone-bridge
configuration feeding one of the high-sensitivity differential inputs on the SC-2043-SG.
A little circuit theory, programmed into in LabVIEW and including adjustment for the
cable resistance, provides a measure of the sensor resistance, and implementation of the
Callendar-Van Dusen equation, again in LabVIEW, provides the final temperature
measurement.

4.13 Temperature control

The performance of RO membranes is very dependent on the water temperature, as
discussed in section 2.7.5, and, in order to control this, the test rig is fitted with both an
electric heater in the tank and a titanium-plate heat exchanger in the concentrate line for
cooling. The cooling is provided by tap water, which is controlled by a solenoid valve.
The heater and the solenoid valve are controlled through digital outputs from LabVIEW.
The control uses pulse-width modulation, based on a saw-tooth at 0.05 Hz, coupled with
a P-I controller. This arrangement normally holds the measured temperature within
0.1 °C of the setpoint.

55
4.14 Torque
In order to measure torque, each motor is mounted in a swinging cradle concentric with
the shaft (gimbals) and fitted with a load-cell: Amber Instruments, model 355. The
manufacturer’s test and datasheet (for the load-cell only) is shown in Appendix B,
section B.5.1, and presents data in terms of kilograms of “load” (mass). In theory, this
could be converted to Newtons of force and then to [Link] of torque, but this
approach would rely on accurate knowledge of the effective geometry of the load-cell
mounting with respect to the motor shaft centre. In practice, measurement of this
geometry would be awkward and error-prone. As an alternative, the complete mechanism
(swinging cradle and load-cell) was calibrated by hanging mass on a horizontal arm of
known radius, as detailed in Appendix B, section B.5.2.

The load-cells are prone to electrical noise, primarily because they are very close to their
respective motors. The effect of the noise is repeatable to some extent and has been offset
as far as possible within LabVIEW. In future, it may be better to remount the load-cells
further from the motors.

4.15 Speed
The motor speeds are detected by a photocell in one case, and a magnetic sensor in the
other. They both work very well, and provide pulses that are processed as described in
section 4.5.

4.16 Voltage, current and electrical power

The measurement of electrical power consumed by the two inverters (variable-speed
drives) is achieved through high-resolution measurements of the one voltage and two
currents. In particular, the three signals are sampled at 10-kHz, then multiplied to give
power, before being averaged. This ensures that non-sinusoidal and non-DC components
are accurately measured.

4.17 Irradiance
The solar irradiance in the plane of the PV array was measured with a Kipp and Zonen
CM11 pyranometer (Appendix B, section B.6), which is connected to a separate data
logging system. It is logged at 10-second intervals and was aligned with the LabVIEW
data by way of its timestamps.

56
 Chapter 5 Component options, testing and modelling

There are innumerable possible configurations of membranes, pumps, energy recovery

mechanisms and so on that could be considered for a PV-RO system. To assess the
relative merits of different configurations, without building and testing them all in
hardware, comprehensive and flexible software modelling was developed. The test rig
and data acquisition system, already described, was used to construct models of the most
critical components, and this was supplemented by data from manufacturers and so on to
complete a set of models covering all the likely candidate components. These component
models were then used to construct system models that allowed rapid assessment of
numerous configuration options.

Of course, this strategy is pretty common throughout engineering, and there are various
software packages that have a similar aim. In the field of RO system design, the
membrane manufacturers generally offer software to system designers. DOW FILMTEC,
for example, have a program called ROSA (DOW 2000), and KOCH Membranes have
ROPRO (Koch-ROPRO 2000). These packages are similar in that they each include
component models for all the membrane elements available from the respective
manufacturer. The user can construct trains and arrays of these elements, and run
simulations to predict overall performance. Both packages include detailed modelling of
the chemical composition of the feed waters and provide chemical analysis predictions of
the product. They model flows and pressures throughout the proposed system and can be
used to estimate pump energy consumptions. Both of these programs were studied and
used in the development of the modelling described here, but neither has the scope
required for the design of a full PV-RO system. The main limitation in ROSA and
ROPRO is that they only simulate operation at a single point, and have no facility for
modelling variable flow, pressure, temperature and so on. Furthermore, they are only
accurate within the normal working range of RO membrane elements; the batteryless PV-
RO system described in this thesis sometimes operates at flow and pressure well below
this range.

57
5.1 Modelling strategy – MATLAB-Simulink
MATLAB-Simulink is a general-purpose modelling and simulation package that is used
throughout science and engineering research and design. MATLAB itself is a text-based
programming language that is particularly intended for manipulation of matrixes (hence
the name) and mathematical computing in general. Simulink is a graphical programming
language that acts as a front-end to MATLAB and allows rapid construction of dynamic
simulation models. MATLAB-Simulink was used extensively for the modelling of the
PV-RO system. MATLAB (without Simulink) was used for much of the off-line
processing of data from LabVIEW.

Simulink models are constructed as interconnected blocks that are structured

hierarchically. The programmer has complete control over the structure, and the first
challenge is to design an efficient structure for the task in hand.

Controller

1 Pa
Pa P Inverter P Motor
Plunger Pump
f P* f* Pdc T n n T

Pi Qi
Pp Iac Vac V I
Po Qo

pf f f pf

Ts
M Inverter M Motor Moineau Pump

f M* f* Pdc T n n T Clark Pump

Split

P Q Qf Pf
Pm Iac Vac V I
Combine RO Array

Ph Qh Qf Pf
pf f f pf

Qc Pc Pc Qc

Qp
Qp 1
Cp
2 T Cp 2
Ts

Figure 5-1 – Simulink model structure

The Simulink model shown in Figure 5-1 is presented here only to illustrate the general
structure and interconnections between the component models; detailed discussion of the
system model is deferred until Chapter 6, after the components have been discussed.

58
The structure of the model illustrated in Figure 5-1 may seem straightforward and
obvious. Alternative RO system configurations can be modelled simply by dragging the
blocks around and reconnecting the lines between them. Blocks can be copied and
deleted and so on, and new blocks are fairly easy to construct. Achieving this apparent
simplicity was not trivial.

The challenge in structuring a model of a physical system such as PV-RO is that

everything depends on everything else, and it is very easy to end up with a mess of
interconnecting lines, and blocks that are not identifiable with any particular hardware
component. For example, consider one pump feeding one membrane element. The flow
of water into the membrane element is a function of its pressure (and various other
things). The flow can also be determined as a function of motor speed, and, assuming
there are no leaks, the two can simply be equated to give an expression relating motor
speed to pressure. The snag is that this expression is not associated with any one physical
component. One could easily create a Simulink block to perform the calculation but its
presence in the model would spoil the one-to-one relationship between the software
blocks and the hardware components. The model could be perfectly good, in terms of
performing the right calculations, but would be very poor for the purpose of trying out
alternative RO system configurations. The blocks shown in Figure 5-1 are immediately
identifiable as the physical components of the system being modelled.

Achieving the clean structure shown in Figure 5-1 is largely a matter of choosing the
inputs and outputs of each block appropriately. For example, one might think that a
motor outputs both speed and torque to a pump and propose a model such as that shown
Figure 5-2 a). One quickly finds that this does not work, and the structure needs to be
either b) or c). The choice between b) and c) is purely a matter of preference in this
simple example, but for the full PV-RO system, which includes a loop by virtue of the
energy recovery, membrane elements with three pipe connections and a Clark pump with
four, the choices become more critical.

59
 Speed Speed
a)
Torque Torque

Motor Pump

b) Torque Speed Speed Torque

Motor Pump

c) Speed Torque Torque Speed

Motor Pump

Figure 5-2 – Simulink structure possibilities

The structures shown in Figure 5-2 b) and c), and indeed in Figure 5-1, include quantities
that are fed backwards (from right to left), implying that the model must know the
answer before it can work it out. These circular references are a necessary feature of this
type of system model. In Simulink, they are known as algebraic loops and are solved
iteratively using Newton’s method.

In conclusion, this section has introduced Simulink, which was used for modelling the
components presented in the remainder of this chapter, and has indicated how the
component models can be shaped such that they will be useful in the consideration of
alternative PV-RO system configurations presented in the next chapter.

5.2 Clark pump

The Clark pump is central to the final PV-RO system design presented in this thesis. Its
basic operation was described in section 2.8.10, but it is perhaps best understood from
the animation at ([Link]
accessed: March 04). Also available at that site is the Clark Pump Field Repair &
Rebuild Manual (Spectra 2003), which shows the internal valve mechanisms.

The Clark pump was patented in 1995 and commercialised during 1997 as part of a
complete RO system for use on yachts. The performance of the system as a whole had
been measured and its excellent efficiency was, and still is, the main selling point.

60
CREST obtained a Clark pump in August 2000, at which time, there was no performance
data available for the Clark pump itself, still less any published performance theory or
model.

5.2.1 Theory

[Link] Ideal (lossless) Clark pump

An ideal Clark pump would have no leakage, pressure or frictional losses. The reversal
(shift), at the end of the stroke, would be instantaneous and lossless. Thus, all flows and
pressures would be perfectly constant (assuming constant external conditions). Also, due
to symmetry, the flows and pressures of an ideal Clark pump can be fully understood by
considering the pistons moving in only one direction.

Qe, Pe
Qc, Pc

Qf, Pf Qh, Ph

Figure 5-3 – Ideal Clark pump

Q is the flow and P the pressure. The suffixes stand for feed, exhaust, concentrate and
high-pressure.

Flows
Equating the velocity of the pistons gives:

Qf Qe Qc Q
= = = h (1)
Ap Ap − Ar Ap − Ar Ap

where Ap is the area of the piston and Ar is that of the rod.

Thus:
Qe Qc
Qf = = = Qh (2)
1 − Rt 1 − Rt
where Rt is the theoretical recovery ratio given by:

Ar
Rt = (3)
Ap

61
Rearranging equation 2 gives:

Q f − Q f (1 − Rt ) Q f − Qc Qp
Rt = = = (4)
Qf Qf Qf
Where Qp is the product flow.

Pressures
Summing the forces acting on the piston gives:

∑f = Pf Ap − Pe ( Ap − Ar ) + Pc ( Ap − Ar ) − Ph Ap = 0 (5)

So:
Pf + Pc (1 − Rt ) = Pe (1 − Rt ) + Ph (6)

Powers
Combining equations 2 and 6 gives:

Pf Q f + PcQc = PeQe + PhQh (7)

which simply confirms that:

Power in = Power out (8)

as expected for a lossless system.

[Link] Leakages (flow losses)

Firstly, leakages past the pistons (Qfe and Qhc) and through the central rod seal (Qce) may
be considered.

QE QC

QF Qf Qe Qc Qh QH

Qfe Qce Qhc

Figure 5-4 – Leakages

62
The lowercase suffixes indicate the net flows that act on the piston (the ideal flows, as
before) and the uppercase suffixes indicate the total flows that may be observed at the
external pipe connections. Thus, for example, the net feed flow acting on the piston
would equal the total flow less the leakage: Qf = QF – Qfe.

In practice, various other leakages are possible, particularly through the valve gear, and,
in general, it is possible that any one of the four chambers could leak into any other. This
gives a total of six possible leakage flows and the following relationships.

Q f = QF + Qhf + Qcf − Q fe (9)

Qe = QE − Qhe − Q fe − Qce (10)
Qc = QC + Qhc − Qcf − Qce (11)
Qh = QH + Qhc + Qhf + Qhe (12)

The directions of the leakage flows, and hence the signs appearing in the above
equations, are chosen such that positive flow is always from the higher pressure to the
lower. Thus, all six leakage flows may be expected to be positive.

Effect on recovery ratio

Combining equations 2, 9, 10, 11 and 12 gives:

QE − Qhe − Q fe − Qce QC + Qhc − Qcf − Qce

QF + Qhf + Qcf − Q fe = = = QH + Qhc + Qhf + Qhe
1 − Rt 1 − Rt
(13)

Where, as before, Rt is the theoretical recovery ratio.

Rearranging equation 13 gives:

QC = (QF + Qhf + Qcf − Q fe )(1 − Rt ) − Qhc + Qcf + Qce

(14)
= QF + Qhf + 2Qcf − Q fe − Rt QF − Rt Qhf − Rt Qcf + Rt Q fe − Qhc + Qce

And:

QH = QF + Qcf − Q fe − Qhc − Qhe (15)

63
Thus, the measured recovery ratio is given by:

QP QH − QC
Rm = = =
QF QF

QF + Qcf − Q fe − Qhc − Qhe − QF − Qhf − 2Qcf + Q fe + Rt QF + Rt Qhf + Rt Qcf − Rt Q fe + Qhc − Qce

QF

Qhe + Qce + (1 − Rt )(Qcf + Qhf ) + Rt Q fe

= Rt − (16)
QF

Or:
QL
Rm = Rt − (17)
QF

where QL is the combined leakages:

QL = Qhe + Qce + (1 − Rt )(Qcf + Qhf ) + Rt Q fe (18)

Interestingly, Qhc should have no effect on the measured recovery ratio.

Notice also that, the other five leakages will all serve to reduce the recovery ratio from
its theoretical value (they are all positive in the expression for QL).

Also, from equation 17:

Rm QF = Rt QF − QL (19)

In words this is:

measured product flow = theoretical product flow – leakages

The leakages may be expected to include two forms: Firstly, pressure-driven leakages
that may be expected to increase with pressure. Secondly, fixed-volume-per-cycle
leakages that may be expected to increase with frequency and, therefore, with flow.

Volumetric efficiency
The volumetric efficiency ηV of a Clark pump may be defined as:

measured product flow

volumetric efficiency = (20)
theoretical product flow

64
More specifically:

QP
ηV = (21)
QF Rt

Or:

QL
ηV = 1 − (22)
QF Rt

Useful flow relationships

QP = QF − QE = QH − QC = Rm QF = Rt QF − QL (23)

QE = Q F − Q P = Q F (1 − Rm ) = Q F (1 − Rt ) + Q L (24)

QF + QC = QH + QE (25)

[Link] Pressure Losses

The feed will experience a slight pressure loss due to its own viscosity in the pipes and
valves on its way in to the chamber. Likewise, the high-pressure, on its way out of the
chamber, and so on. These pressure losses are indicated: ∆P .

PE PC
∆Pe ∆Pc
PF Pe Pc PH
∆Pf Pf Ph ∆Ph

Figure 5-5 – Pressure losses

Pf = PF − ∆Pf (26)
Pe = PE + ∆Pe (27)
Pc = PC − ∆Pc (28)
Ph = PH + ∆Ph (29)

where lowercase suffixes indicate the pressures that act on the piston and uppercase
suffixes indicate the pressures that may observed at the external pipe connections.

65
Friction F will also give rise to a pressure loss.

Thus, equation 6 becomes:

F
PF + ∆Pf + (PC − ∆Pc )(1 − Rt ) = (PE + ∆Pe )(1 − Rt ) + (PH + ∆P ) + (30)
Ap
Or:

PF + PC (1 − Rt ) − PE (1 − Rt ) − PH = PL (31)

where PL is the combined pressure losses:

F
PL = ∆Pf + (∆Pe + ∆Pc )(1 − Rt ) + ∆Ph + (32)
Ap
The combined pressure losses, including the friction, may be expected to increase with
flow, or perhaps flow squared.

Mechanical efficiency
The mechanical efficiency ηM of a Clark pump may be defined as:

effective net input pressure

mechanical efficiency = (33)
net input pressure

More specifically:

PF − PE (1 − Rt ) − PL PL
ηM = = 1− (34)
PF − PE (1 − Rt ) PF − PE (1 − Rt )

Useful pressure relationship

PF = PH + (1 − Rt )(PE − PC ) + PL (35)

66
5.2.2 Testing

[Link] Hardware configuration

In order to characterise the Clark pump, the test rig was re-plumbed so that two manually
operated valves were connected in place of the RO modules.

Exhaust
Pressure Pressure
Sensor Sensor

Flow Clark Needle Product Flow

Meter Pump Pump Valve Meter
Feed High-pressure

Ball
Valve
Pressure
Sensor Concentrate

Figure 5-6 – Configuration used for Clark pump testing

The dashed box in Figure 5-6 indicates the usual location of the modules. Testing was
carried out using freshwater, and the terms Product and Concentrate are retained simply
to identify the various flows – obviously, no actual desalination is taking place.

The needle valve is used to simulate the pressure drop of water passing through the
membranes. In practice, this pressure drop would be large (> 30 bar), since it includes
both the osmotic pressure (around 26 bar for seawater) and the mechanical pressure
required to force water through the semi-permeable membranes.

The ball valve simulates the delta pressure: the pressure drop of the water passing across
the membranes along the mesh spacer layer described in section 2.7.3. In practice, this
pressure drop would be very small (< 1 bar).

The pump used for the tests was a positive-displacement plunger pump (CAT 317). It
was driven by an induction motor on a variable-speed inverter.

Thus, the test configuration allowed independent control of feed flow, high pressure and
delta pressure.

67
[Link] Procedure
The rig was operated at a total of 55 discrete test points, including all combinations of:

• feed flow: 0.042, 0.127 and 0.211 L/s,

• high pressure: 20, 30, 40, 50 and 60 bar,

• delta pressure: 0, 2 and 4 bar.

The delta pressures chosen for testing were higher than those expected in practice, in
order to exaggerate, and thus clearly identify, any affect of delta pressure.

Marcos Miranda adjusted the valves by hand to give the chosen test pressures and the
speed setpoint on the motor inverter to give the chosen flows. Data was logged, at each
test point, for at least one minute after steady operation had been achieved.

[Link] Data analysis and modelling

Off-line analysis of the data confirmed that the Clark pump is very efficient, and that this
efficiency is maintained over a very wide range of flows and pressures. The detailed test
results and analysis are presented in Appendix C. In summary, the data was analysed
using regression calculations to determine the dominant causes of leakages and pressure
losses, which were found to be as follows.

- The leakages are dependent mainly on the product of the high-pressure and
the feed flow. The expectation had been that the leakages would increase with
the sum of pressure and flow components.

- The delta pressure (up to 4 bar) has no significant effect on leakages.

- The pressure losses are dependent on the feed flow and feed flow squared.
This was anticipated. Also, a dependence on delta pressure was observed.

The regression calculations also provided the following numerical expressions for the
leakages and pressure losses respectively.

68
 QL = 1.78 × 10 - 4 × QF PH + 1.56 × 10 - 5 (36)
PL = 49.2 × QF2 + 7.09 × 10 - 2 × ∆P + 0.528 (37)
In which, the units are litres/second and bar. And, ∆P = PH – PC

Numerous alternative formulas to calculate leakages and pressure losses were reviewed,
but equations 36 and 37 are considered to offer the most appropriate balance of precision
and simplicity.

Equations 36 and 37 are used to estimate the leakages and pressure losses in the
MATLAB-Simulink model of the Clark pump, presented next.

69
5.2.3 Modelling
The preceding theory and test results presented in Appendix C were used to build a
software model of the Clark pump in the MATLAB-Simulink environment.

1
Qf 2
0 Qh
Qh
Assumed
4 3
Qe Qc

Qf
Ql Qp
Ph
(0.875/2.75)^2
Leakages
Rt

Qf
Pl
dP
1
PressureLosses
1-Rt

2
1
Ph
Ph+(Pe-Pc)(1-Rt) Pf

3
Pc

4
Pe

Figure 5-7 – Simulink model of Clark pump

The inputs to the model are on the left of the Simulink diagram in Figure 5-7.

Qf is the feed flow, previously denoted by QF (Simulink does not allow subscripts.)

Ph, Pc and Pe are the pressures of the high-pressure, the concentrate and the exhaust
respectively.

Rt is a constant. It is the theoretical recovery ratio, calculated from the design dimensions
of the pistons and the rod (see equation 3).

The boxes labelled Leakages and Pressure Losses are subsystems that implement
equations 36 and 37 respectively.

70
The rest of the model shown in Figure 5-7 implements equations 23 and 31.

The zero shown in the box labelled Assumed represents the difference between the feed
flow and the high-pressure flow. In an ideal Clark pump, there would be no difference,
see equation 2. In practice, there probably is a slight difference but this was not
measurable since the oval-gear flow meters on the test rig are limited to 20 bar.
Fortunately, the overall efficiency is barely affected by this assumption as discussed in
Appendix C, section C.4. Thus, very little of the precision of the model is lost.

The software model was tested by using the QF, PH and PC data collected during the
original hardware testing as inputs to the model. The output data from the model was
then compared against the original test results. Unsurprisingly, the match is very good.

5.2.4 Efficiencies and conclusions

There are various ways that the overall efficiency of an energy recovery mechanism may
be defined, see section 2.8.5. The efficiency shown in Table 5-1 is calculated as the ratio
of power delivered to the feed over that available in the concentrate.

QH PH − QF PF
η Clark = (38)
QC PC − QE PE

High Concentrate
Feed Flow Efficiency
Pressure Pressure
(bar) (bar) (L/h) (%)
60 58 153 97.2
50 48 153 97.2
40 38 153 96.9
60 58 458 95.9
50 48 458 95.4
40 38 458 94.8
60 58 759 93.5
50 48 759 92.4
40 38 759 90.7

Table 5-1 – Clark pump measured efficiency

It is apparent that the overall efficiency shown in Table 5-1 decreases significantly with
increased flow and, to a lesser extent, with increased pressure. The data analysis also
showed that the volumetric efficiency, defined in section [Link], is very high throughout,

71
and that the overall efficiency is dominated by the mechanical efficiency (pressure and
frictional losses) defined in section [Link].

Anyone familiar with typical efficiencies of pumps and turbines of this small scale will
appreciate that the figures presented for the Clark pump in Table 5-1 are excellent.
Comparing these results against the experience with the Danfoss hydraulic motor
presented in Chapter 3 indicates that the Clark pump is far superior in terms of energy
efficiency. It was selected for the PV-RO system.

5.3 Plunger pump

Plunger pumps, particularly those manufactured by CAT Pumps ([Link]
accessed: March 04), are routinely used in small-scale seawater RO systems. There are
few other contenders in seawater-compatible pumps that can raise the required pressure
(up to 69 bar) in small sizes. They are mostly alternative brands of plunger and piston
pumps similar to CAT’s, and probably no better. Recently, Danfoss introduced a new
range of axial piston pumps, the APP range (Danfoss 2002), which are designed for use
in seawater RO and would be worth considering in the future.

The initial test rig, described in Chapter 3, included a CAT 317 triplex plunger pump,
and this same pump was transferred to the new test rig; it has performed impeccably
throughout and shows no sign of deterioration. The final system design, presented in
section 6.4, also includes a CAT triplex plunger pump, in particular, a CAT 237.

Geometric
Bore (mm) Stroke (mm) Plungers displacement
(L/rev)
CAT 317 20 18 3 0.01697
CAT 237 18 7 3 0.00534
Table 5-2 – CAT plunger pump geometries

Table 5-2 shows the bore and stroke, taken from the manufacturer’s datasheets, and the
geometric displacement calculated using:

2
 bore 
geometric displacement = π   × stroke × plungers
 2 

72
The CAT 237 was selected for the final system design because its smaller geometric
displacement gives a higher shaft speed, which improves the efficiency of the motor
driving it. The CAT 237 was modelled as follows.

5.3.1 Modelling

1 1/60 Qi
(rpm) (rev /s)
n 2
Ideal f low (L/s)
Geometric
(L/rev ) n
displacement Qo
eta
Pd
3
18 Volumetric
Bore (mm) efficiency
7
Stroke (mm)
3
Plungers

-K-
(m3/rad)
3
Ideal torque (Nm) T
Po
1e5 1
Pressure (Pa) n
dif f erence eta
Pd
2 (bar)
Mechanical
Pi
efficiency

Figure 5-8 – Simulink model of CAT 237 triplex plunger pump

Starting at the top left of Figure 5-8, the input n is the shaft speed in rpm, which is
converted to revolutions per second (rev/s). Multiplying this by the geometric
displacement in litres per revolution (L/rev) gives the Ideal flow (L/s). A volumetric
efficiency is then applied to provide the outputs Qi and Qo, which are the modelled
flows through the pump inlet and outlet respectively. Qi and Qo are of course equal
(assuming the pump has no external leak); they are provided as separate outputs from this
Simulink component model only to provide consistency for the connections when it is
used within the system model.

At the bottom left of Figure 5-8, inputs Pi and Po are the pressures at the pump inlet and
outlet respectively. The Pressure difference (bar) is converted to pascals (Pa) and then
multiplied by the geometric displacement in cubic metres per radian (m3/rad) to give the
Ideal torque (Nm). A mechanical efficiency is then applied to provide the output T,
which is the shaft torque that will be connected to the motor.

73
The expressions used for volumetric and mechanical efficiencies were based on
measurements taken from the CAT 317 by Marcos Miranda. He performed regression
calculations (similar to those presented in Appendix C) and provided:

volumetric efficiency = 1.00 − 8.75 × 10 - 6 Pd + 0.000173n − 2.22 × 10 - 7 Pd n

mechanical efficiency = 0.420 + 0.0225 Pd − 0.000430 Pd2

+ 2.91 × 10 - 6 Pd3 + 0.000138n − 6.97 × 10 -8 n 2

5.3.2 Modelled efficiency

100

90
~55 bar
80
~30 bar
70
Efficiency (percent)

60

50

40

30

20

10

0
0 200 400 600 800 1000 1200
Shaft speed (rpm)

Figure 5-9 – Modelled efficiency of CAT 237 plunger pump

The efficiency of a plunger pump is dependent on the operating conditions: both flow
and pressure. And, in the context of a batteryless PV-RO system, both flow and pressure
are expected to vary with the available sunlight. It is the efficiency over a broad range of
operation that is of importance.

74
The data shown in Figure 5-9 is taken from the complete system model presented in
Chapter 6. Thus, it shows the efficiency calculated by the pump model presented in
section 5.3.1, under the conditions expected in the complete PV-RO system.

The efficiency shown in Figure 5-9 is between 85 and 90 % for the majority of the
expected operating range. This is very respectable for such a small pump. And, on this
basis, the CAT 237 was selected for use in the final system design.

5.4 Moineau pump

The selection of a medium-pressure pump to feed the Clark pump is rather less
straightforward. Obviously, efficiency is paramount and, unfortunately, plunger pumps
are much less efficient at reduced pressure. Seawater compatibility is an obvious
requirement, but additionally, the pump must be able to tolerate the occasional grain of
sand, in order to minimise filtration requirements on its intake side.

A multi-stage centrifugal pump would seem to be a possibility, but, these offer very poor
efficiency (less than 40%) at the required flow/pressure. Furthermore, to achieve
optimum efficiency with a centrifugal pump, the rotor speed must be matched to the
flow/pressure operating point. This is not straightforward in a system where both the
flow and pressure must vary according to the available sunlight. Lastly, since a
centrifugal pump does not offer positive displacement, balancing the water recovery ratio
in the final two-pump system, Figure 1-3, would be rather difficult.

A wide range of diaphragm and vane pumps were also considered but finally a Moineau
(progressing-cavity) pump was selected. These offer reasonable efficiency at the required
duty and, importantly, this efficiency is generally maintained as the flow is reduced. Such
pumps are often used in PV-powered systems.

Moineau (progressing-cavity) pumps were developed by a Frenchman, Rene Moineau,

and his name has been Anglicised and Americanised by the manufacturers: Mono and
Moyno respectively. Other manufacturers include Netzsch and Orbit. Moineau pumps are
positive-displacement, employing a long metal screw rotor inside a rubber stator. They
can tolerate the odd grain of sand but are quickly ruined if run dry.

Submersible Moineau pumps are widely used and could be used in this PV-RO
application. However, if the well is shallow (less than 7 m) and the rig can be sited

75
nearby, an above-ground Moineau pump may be preferable, since this will permit use of
a much higher efficiency motor. An above-ground Moineau pump was selected for the
final design and performance predictions are based on this.

Moineau pumps can prime themselves if sufficiently wet, but can suffer from starting
problems due to the static friction between the rubber stator and the metal rotor – once
turning, the water being pumped acts as a lubricant. Starting difficulty was apparent
during the commissioning of the final test rig, but was overcome by a software routine in
LabVIEW that uses the inverter (variable-speed drive) to wobble the motor at start up.

Seeking a combination of seawater compatibility and energy efficiency, a Netzsch 021

([Link] accessed: March 04) was selected.

5.4.1 Modelling data

Within this model, the inlet pressure of the Moineau pump is taken as zero (atmospheric
pressure), representing the case where the pump is neither sucking nor being feed at
positive pressure, which is in keeping with the arrangement of the final test rig. Thus, the
pressure referred to below is equal to the outlet pressure.

The manufacturer’s performance curves (Appendix D) show that, at zero pressure, the
flow is directly proportional to shaft speed, as expected in a positive displacement pump,
In particular:

flow 3.7 m 3 h
geometic displacement = =
shaft speed 2500 rpm

As the pressure is increased, leakage between the metal rotor and the rubber stator
increases, which reduces the flow as illustrated in the performance curves and quantified
in the second column of Table 5-3.

Pressure Leakage: flow reduction Shaft power (kW)

(bar) (m3/h) at 250 rpm at 1000 rpm at 2500 rpm
4 0.0545 0.12 0.40 1.25
8 0.115 0.15 0.55 1.60
12 0.206 0.19 0.70 1.90
Table 5-3 – Netzsch 021 Moineau pump data extracted from manufacturer’s
performance curves (Appendix D)

76
Fitting a quadratic to the leakage data shown in Table 5-3 provides:

leakage = 0.000568P 2 + 0.0102 P + 0.00121

where leakage is in m3/h and P is the pressure in bar.

Using multiple regression (similar to that presented in Appendix C, section C.6) for the
shaft power shown in Table 5-3 provides:

power = 0.0167 + 0.000186n + 0.00277 P + 3.18 × 10 -5 nP + 7.11 × 10 -8 n 2

where power is in kW, n is the shaft speed in rpm and P is the pressure in bar.

5.4.2 Simulink model

[ 0.0167 0.000186 0.00277 3.18e-005 7.11e-008 ] Power coefficients

1
(rpm) n
n 1000
(kW) (W)
P
T
P*n
1
2 (Nm)
u
n2

pi/30
n (rad/s)
rpm2rad

m3/h/rpm at 0bar
3.7/2500
Ideal f low

2
(bar) Q
P 1/3.6 2
Leakage coefficients (m3/h) (L/s)

(0.00121 0.0102 0.000568)

P
2
u
P2

Figure 5-10 – Simulink model of Netzsch 021 Moineau pump

77
Figure 5-10 shows the structure of the Simulink model of the Netzsch 021 Moineau
pump constructed from the equations presented in section 5.4.1. On the left of the figure,
the inputs n and P are the shaft speed and the pressure respectively. On the right, outputs
T and Q are the shaft torque and the flow respectively.

5.4.3 Modelled efficiency

100

90

80

70
~11 bar
Efficiency (percent)

60

50

40 ~5 bar

30

20

10

0
0 100 200 300 400 500 600 700 800 900 1000
Shaft speed (rpm)

Figure 5-11 – Efficiency of Netzsch 021 Moineau pump modelled from

manufacturer’s performance curves

The efficiency of a Moineau pump is very dependent on the operating conditions: both
flow and pressure. And, in the context of a batteryless PV-RO system, both flow and
pressure are expected to vary with the available sunlight. It is the efficiency over a broad
range of operation that is of importance.

The data shown in Figure 5-11 is taken from the complete system model presented in
Chapter 6. Thus, it shows the efficiency calculated by the pump model presented in
section 5.4.2, under the conditions expected in the complete PV-RO system.

Figure 5-11 may be compared against Figure 10-10, which shows the data measured in
the new test rig.

78
5.5 Motors and inverters
In relation to PV pumping, it is often suggested that permanent-magnet motors (often
DC, but sometimes AC) offer significantly higher efficiencies than AC induction motors
(Wenham et al. 1994 page 212). However, improvements, over the last decade or so, in
the efficiency of small induction motors and inverters for mainstream industrial
application mean that this technology is now almost as efficient as any commercially
available permanent-magnet motor system, AC or DC. The MATLAB-Simulink model
of the complete PV-RO system showed that the higher cost of various alternative
permanent-magnet motor drive systems that were considered was not justified, see
section 5.6.4. Furthermore, the use of standard industrial components that can readily be
replaced is very attractive for systems intended for use in remote areas. Thus, AC
induction motors were selected.

Next, dedicated PV-pumping inverters were considered since they have the advantage
that maximum power point tracking (MPPT, introduced in section 2.5.4) is built in.
However, since it is a relatively small market, these inverters tend to be rather expensive
and offer little flexibility in control strategy. In particular, the appropriate sharing of the
available power between the two motors, which is necessary to optimise the water
recovery ratio and hence maximise the product flow, would be difficult to implement
with standard PV-pumping inverters.

The alternative, to dedicated PV-pumping inverters, is the standard industrial variable-

speed drive inverter. These are far less expensive but have no MPPT, and require suitable
selection of the PV modules so that the DC voltage falls in a relatively narrow range.

79
 AC
Supply Motor

Bridge
rectifier Inverter

Figure 5-12 – Standard industrial drive circuit

Standard industrial drives are normally intended to operate from an AC mains supply as
illustrated in Figure 5-12, but, so long as nothing is internally connected prior to the
bridge rectifier, can usually be operated directly from DC (check with the manufacturer).
Direct connection of a PV array to a standard industrial drive is a proven technique. It
requires that the PV-array voltage range lies within the safe operational voltage range of
the inverter. A typical rating is 230 VAC +/–15%, which translates to 276 to 374 VDC,
but in practice, these boundaries can often be extended (check with the manufacturer).

Standard industrial motors and inverters were selected for the PV-RO system.

5.6 Induction motors

Induction motors are used throughout industry, and account for a considerable proportion
of electricity consumption worldwide. Recognising that a general improvement in the
efficiency of induction motors brings significant reduction in electricity consumption,
many countries have introduced standards and regulations to encourage manufacturers to
improve designs with regard to energy efficiency.

5.6.1 Modelling data

The motor chosen and modelled in Simulink during the design of the PV-RO system was
the Toshiba Premium Efficient EQPIII B0026FLF2AYH, which is a 6-pole induction
motor rated at 2-horsepower (~1.5 kW) on 230-V 60-Hz. Toshiba provided the data
shown in Table 5-4, from which Table 5-5 is derived, assuming a linear torque-speed
relationship.

80
 Load Torque Speed IEEE Efficiency Power Factor
(%) ([Link]) (rpm) (%) (%)
100% 8.96 1172 88.4 66.7
75% 87.7 58.1
50% 85.3 45.6
Table 5-4 – Toshiba motor data

Output Input Active Reactive

Load Torque Speed Power Power Current Current Current
(%) (Nm) (rpm) (W) (W) (A) (A) (A)
100% 12.15 1172 1491 1687 6.35 4.23 4.73
75% 9.11 1179 1125 1283 5.54 3.22 4.51
50% 6.07 1186 754 885 4.87 2.22 4.33
Table 5-5 – Toshiba motor – derived data

5.00
Reactive current = 0.0653T + 3.9301

4.00

3.00
Current (A)

2.00
Active current = 0.3315T + 0.2039

1.00

0.00
0.00 2.00 4.00 6.00 8.00 10.00 12.00 14.00
Torque (Nm)

Figure 5-13 – Active and reactive motor currents versus torque

Figure 5-13 illustrates, as expected, that active current is proportion to torque, with an
offset for friction and windage, while reactive current (magnetising current) is more
constant.

81
5.6.2 Simulink model
f
3

60
Sy nc. speed n
2 1
Speed

6
Poles
rpm2rad/s
Slip
2.3046 w
-K-
Shaf t Power
rpm per Nm
T
1
T

0.3315*u + 0.2039 pf
Ia 3
Power Factor

I
Ia
0.0653*u + 3.9301 I 2
Ir Motor Current
Ir
Pythagoras

Efficiency
Ia
V
Activ e Power
2
Losses
sqrt(3)

Figure 5-14 – Structure of Simulink motor model

Starting on the left of the motor model shown in Figure 5-14, the input f is the frequency
of the ac voltage from the inverter. From this frequency in hertz, the synchronous speed
in rpm is calculated, for a six-pole motor.

The slip (in rpm) is calculated from the torque T, using data from Table 5-5 and
assuming a linear torque-speed relationship.

The synchronous speed less the slip provides the shaft speed n.

The active current Ia and reactive current Ir are estimated from the torque using the linear
functions illustrated in Figure 5-13, and are combined to provide the motor current I and
its power factor pf.

The remaining blocks in Figure 5-14 illustrate the calculation of motor input and output
powers, losses and efficiency.

82
5.6.3 Induction motor efficiencies
The Toshiba Premium Efficient EQPIII B0026FLF2AYH, on which the modelling just
described was based, is an American motor built to National Electrical Manufacturers
Association (NEMA) standards. NEMA is based in the US and covers the US motor
market. Motors for the US market must comply with the US Energy Policy and
Conservation Act (EPACT), which uses the IEEE 112-1992 test method B to define the
efficiency of small induction motors.

Elsewhere in the world, particularly in Europe, the efficiency of small induction motors
is classified according to the European Committee of Manufacturers of Electrical
Machines and Power Electronics (CEMEP) and is calculated according to an
International Electrotechnical Commission (IEC) standard: IEC 60 034-2.

Siemens manufacture motors for both markets, and in some cases supply the same model
to both markets. Their 1999 catalogue shows side-by-side specifications for the two
markets. For example, the 1LA9 106-6KA (which is similar to the aforementioned
Toshiba motor) is shown at 60 Hz for the US market and at 50 Hz for other markets.

50 Hz 1.5 kW 6-pole 945 rpm IEC efficiency: 83%

60 Hz 2 hp 6-pole 1150 rpm IEEE efficiency: 86.5%
Table 5-6 – Alternative specifications for the same motor (Siemens1LA9 106-6KA)

Table 5-6 shows that, for this example at least, the IEEE efficiency is some 3.5 % higher
than the IEC efficiency, and illustrates the need for caution in comparing motor
efficiencies.

5.6.4 Value of motor efficiency

In order to quantify the value of motor efficiency within the PV-RO system, and
therefore provide a basis to compare motors of different costs and efficiencies, the final
system model, presented in Chapter 6, was used as follows.

The efficiencies of the two motors in the system model were temporarily adjusted, first
up 1 %, then down 1 %. In particular, multiplication factors of first 1.01 then 0.99 were
inserted in the active current Ia line of Figure 5-14.

83
 Motor efficiency Active current Annual water
factor production (m3)
Base case 1 1424
+1% 0.99 1435
–1% 1.01 1413
Table 5-7 – Modelled results for motor efficiency changes

Table 5-7 shows that 1 % increase (or decrease) in motor efficiency gives an increase
(decrease) in annual water production of 11 m3, which is 0.77 %. The total capital cost of
the PV-RO system, also estimated in Chapter 6, is £18272. Thus, the value of motor
efficiency within the PV-RO system is 0.77 % × £18272 = 141 £/%.

5.6.5 Motor selection

By fortunate coincidence, the two selected pumps (Moineau described in section 5.3 and
plunger described in section 5.4) have very similar shaft speed and torque requirements,
and can be well served by two motors of the same type; in particular, a 6-pole 1.5 kW
induction motor.

Various 6-pole 1.5 kW motors were considered during the system design, including those
shown in Table 5-8.

Cost Efficiency Adjusted cost

Make and model
(£) (%) (£)
IEEE IEC Relative
Toshiba B0026FLF2AYH 309 88.4 0 309
Siemens 1LA9 106-6KA 165 86.5 83.0 -1.9 433
WEG W21 HDA 100 B3 98 80.1 -4.8 775
FKI Marelli MA112M6 139 75.3 -9.6 1493
Baldor EM3664T 388 88.5 85.2 0.1 374
Table 5-8 – Costs and efficiencies of various 6-pole 1.5 kW induction motors

The cost shown in Table 5-8 for the Toshiba motor was based on its list price US$440 at
the time (early 2001). The other costs were taken from specific quotations. The
efficiencies shown are from the manufacturer’s data and are a mix of IEEE and IEC as
discussed in section 5.6.3. The relative efficiency shown takes the Toshiba as the base
case, and takes into account the difference between IEEE and IEC efficiencies. The last
column in Table 5-8 shows the cost adjusted according to the relative efficiency at
141 £/%. The minimum adjusted cost is clearly that of the Toshiba and illustrates why it
was selected for the final design presented in Chapter 6.

84
5.6.6 Test rig motors
When the new test rig was being built, later in 2001, it was found that the selected
Toshiba motor was not readily available in the UK and that importing it would inflate its
cost to around £870. The next choice, according to adjusted costs in Table 5-8, would be
the Baldor, but this has a high unadjusted cost (which was relevant for building a test rig)
and had a long lead-time. Considering this the third choice, a Siemens 1LA9 106-6KA
was purchased for the test rig. This has a quoted efficiency 1.9 % lower than the Toshiba;
however, there is no guarantee that the efficiencies quoted by the two manufacturers can
be compared with this accuracy.

The new Siemens motor drives the Moineau pump in the new test rig. The plunger pump
is driven by 4-pole 3-kW motor left over from the earlier test rig.

5.7 Standard industrial variable-speed drive inverters

Various different industrial drive inverters were used throughout the project. There was
little to choose between the manufacturers. One of the inverters included sensorless
vector control. This gave more accurate speed control, which is largely irrelevant for PV-
RO, but consumed power in a rather greedy and erratic fashion, particularly at low
speeds, which is not good in a PV-powered system.

The inverters modelled and subsequently bought and fitted to the new test rig, were
supplied by FKI Industrial Drives. In particular, a 1.5-kW FID1000 FKI-12150 is used
for the Moineau pump on the test rig and for both pumps in the system model. A slightly
larger 2.2-kW FID1000 is used for the plunger pump on the test rig. It is oversized only
to allow flexibility in testing.

They are V-over-F type inverters, which means that the AC output voltage is directly
proportional to its frequency, so that V-over-F is a constant. V-over-F type inverters are
usually cheaper than those with vector control, and the FID inverters have performed
well.

V-over-F type inverters often provide the option to apply slip-compensation. When
selected, this option applies a slight increase to the frequency of the inverter’s ac output
voltage, to offset the estimated slip in the induction motor, so that the shaft speed more

85
closely matches the speed setpoint. Slip-compensation is of no benefit for the PV-RO
system in hand and was turned off.

5.7.1 Modelling
With slip-compensation turned off, the frequency of the inverter’s ac output voltage is
directly proportional to the speed setpoint input. Thus, this input can be expressed as a
speed setpoint n* in rpm, as it is in Chapter 9, or as frequency setpoint f* in hertz as it is
in the Simulink model shown in Figure 5-15.

f* f
1 3

Vac
2
230/60
V-over-F

2 Motor activ e power Pdc

Iac 1
sqrt(3)

3
pf Losses
u
0.42*u^2 + 4.7*u + 7.7

Figure 5-15 – Structure of Simulink inverter model

Starting at the top left of Figure 5-15, the input f* is the frequency setpoint and is passed
straight through to the output f, which is the frequency of the inverter’s ac output, as
explained above.

Multiplying f by the constant V-over-F provides Vac, which is the voltage of the
inverter’s ac output. In this case, V-over-F = 230V/60Hz, corresponding with the Toshiba
motor described in section 5.6.1.

Again on the left of Figure 5-15, inputs Iac and pf are the motor current and power
factor respectively, which will be connected to the outputs from the motor model
described in section 5.6.2.

86
The motor active power is simply the product of motor voltage, current and power
factor, with further factor of root-3 because it is a three-phase motor.

The output Pdc represents the DC power drawn by the inverter and is the sum of the
motor active power and inverter losses, which are modelled as follows.

FKI, the inverter manufacturers, provided measured loss data, for the 1.5-kW FID1000
FKI-12150, at three current levels and a curve was fitted as shown in Figure 5-16.

45
Manufacturer's data
40 Fitted curve

35

30
Inverter loss (W)

25

20

15

10

0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Motor current (A)

Figure 5-16 – Inverter losses versus motor current

The inverter losses were expected to include components in I 2 (resistive losses),

I (switching losses) and a constant. The manufacturer’s data was consistent with this and
a 2nd order polynomial was fitted:

Inverter loss = 0.42 I 2 + 4.7 I + 7.7

where I is the motor current in Amps and the inverter loss is in Watts.

The maximum motor current shown in the complete system model was 4.9 A, which, as
illustrated in Figure 5-16, required modest extrapolation of the data.

87
5.7.2 Parameter adjustments
During commissioning, it was found that the detailed settings of control parameters
within the inverters can have a significant effect on energy efficiency. Since these
inverters are normally used for industrial purposes and supplied by mains electricity, the
instruction manuals tend to focus more on starting performance and speed control than on
energy efficiency. Using the instrumentation and data-acquisition system attached to the
test rig, it was possible to observe energy efficiency in real-time and make adjustments
accordingly.

The first adjustment was to the V-over-F ratio. It was found that, for one motor, it was
best left at the default: 230 V at 50 Hz (the test rig has 50-Hz nominal motors). But for
the other motor, energy efficiency could be improved by ~2% by reducing the setting to
180 V at 50 Hz. This degraded the speed control due to the increased slip, but that is of
little consequence in a PV-RO system.

Next, the starting boost was adjusted. This is a parameter commonly found on industrial
inverters that increases the voltage at low frequencies in order to provide higher starting
torque. This is useful in an industrial context because it ensures that machinery starts
decisively at low speeds. However, it does consume extra power and, in the context of
PV-pumping, including PV-RO, it can do more harm than good. Testing showed that for
both the plunger pump and the Moineau pump the starting boost should be set to a
minimal value in order to start the motors with minimum power. When commissioning
an inverter with a Moineau pump, one might imagine that a generous starting boost
should be applied in order to overcome stiction, but this is not the case; it is better to
overcome stiction by wobbling the motor, if the software permits.

Next, the inverters used had a facility to bend the V-over-F characteristic, allowing the
voltage at mid frequencies to be adjusted. This gave no efficiency improvement.

Lastly, different switching frequencies were tested. The default was 16 kHz. Changing
this to 8 kHz, improved efficiency by ~1.2 %, because of the reduced switching losses,
but the 8 kHz was highly audible, bordering on painful, and the setting was put back to
the default 16 kHz for all further tests.

88
5.8 Solar irradiance
The main input to the PV-RO system model is of course the solar irradiance striking the
PV array, which depends on the proposed location. Meteonorm ([Link]
accessed: March 04) is a software package that predicts irradiance at any chosen location.
It uses algorithms, based on measured data from many sites around the world, together
with some basic characteristics of the chosen location.

Meteonorm was used to provide data for a sample location at Massawa, Eritrea. The data
is hourly and covers one whole year. Meteonorm aims to provide data for a typical year,
as opposed to the long-term average for many years.

89
5.8.1 Solar-trajectory tracking
The system was modelled with a fixed PV array and with single-axis and dual-axis solar-
trajectory-tracking arrays. Meteonorm provided global irradiance data in the plane of the
array for both the fixed array and the array with dual-axis tracking. Rowbottom (Dulas
Limited) then used PVSyst ([Link] accessed: March 04) to adjust this
Meteonorm data to provide the data for the array with single-axis tracking. Figure 5-17
shows the expected bell-shape daily profiles for these three sets of data averaged over the
whole year.

900
DualAxis
800 SingleAxis
FixedArray

700
Annual average irradiance (W/m2)

600

500

400

300

200

100

0
0 2 4 6 8 10 12 14 16 18 20 22 24
Time of day (hours)

Figure 5-17 – Annual average global irradiance versus time of day for a fixed PV
array and for single-axis and dual-axis solar-trajectory tracking arrays

90
 Time of day Annual average global irradiance
(hours) in the plane of the fixed array
(W/m2)
Fixed Single-axis Dual-axis
6 0 0 0
7 23 109 114
8 182 451 461
9 386 627 637
10 577 738 755
11 738 815 839
12 837 855 881
13 867 868 893
14 814 843 869
15 694 786 807
16 531 709 725
17 336 595 605
18 131 367 376
19 9 33 34
20 0 0 0

Average 255 325 333

Percentage gain 27% 2%
Table 5-9 – Data for Figure 5-24

The percentage-gain figures in Table 5-9 suggest that, at this location (Eritrea), the
additional costs (both capital and maintenance) of a single-axis tracker may well be
justifiable, but that the higher costs of a dual-axis tracker are probably not. And on this
basis, a single-axis tracker was assumed throughout the design process. The figures in
Table 5-9, however, represent the irradiance, which is the input to the PV-RO system; the
choice of tracking arrangement was later checked with reference to the water production
predicted by the final system model presented in section 6.5.1.

It should also be noted that the relative merits of single- and dual-axis tracking are very
dependent on latitude.

5.8.2 Hourly irradiance

As noted earlier, Meteonorm aims to provide hourly data for a typical year, as opposed to
the long-term average for many years. Thus, the data for a particular day is affected by
local atmospheric conditions (particularly cloud cover) simulated within Meteonorm and
may well differ from the bell-shaped average day.

91
 1200

1000

800
Irradiance (W/m2)

600

400

200

0
1 2 3 4 5 6 7
Day of year

Figure 5-18 – Hourly global irradiance in the plane of a single-axis tracking PV

array for the first week of the year-long data set

Figure 5-18 shows the hourly irradiance data for only one week of the year-long data set.
The simulations of the PV-RO system were generally carried out on an hour-by-hour
basis for the whole year (8760 hours). In order to present the input and results data here,
they are averaged daily and monthly.

92
5.8.3 Average daily irradiance

500

450

400
Average daily irradiance (W/m2)

350

300

250

200

150

100

50

0
0 50 100 150 200 250 300 350
Day of year

Figure 5-19 –Average daily global irradiance in the plane of a single-axis tracking
PV array for the whole of the year-long data set

93
 Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
1 90 141 376 337 406 381 330 379 326 376 428 338
2 284 187 392 270 452 296 163 359 395 379 381 379
3 349 399 410 252 403 314 297 341 431 359 319 369
4 364 381 291 433 316 313 331 47 386 384 353 268
5 344 329 424 436 255 300 401 49 404 384 381 269
6 154 325 393 261 185 322 218 380 387 322 286 268
7 363 168 424 370 343 340 367 197 287 292 358 280
8 252 362 392 362 418 297 270 397 304 277 328 353
9 222 403 450 423 456 364 339 367 285 312 309 345
10 246 385 437 428 457 149 362 373 202 384 327 132
11 354 362 43 361 428 389 205 241 401 246 180 191
12 344 342 451 203 110 175 301 126 373 417 228 267
13 368 362 203 402 317 217 333 330 296 382 316 360
14 267 365 384 410 396 283 382 115 366 327 350 366
15 275 409 410 411 266 403 320 254 220 375 315 153
16 392 405 373 427 361 166 344 345 311 314 319 126
17 314 377 483 362 335 358 360 348 202 361 225 368
18 392 414 408 249 199 317 371 406 309 273 265 388
19 388 397 411 418 196 345 367 392 382 357 258 369
20 379 399 392 349 440 254 72 397 412 287 381 347
21 383 75 454 324 317 327 284 333 370 272 331 354
22 103 392 336 403 211 323 361 407 366 321 315 371
23 388 363 370 426 379 347 284 398 332 343 228 100
24 398 410 259 405 388 339 346 398 279 328 277 348
25 397 434 361 261 384 282 125 350 290 172 118 220
26 313 435 242 398 287 295 222 316 290 378 275 235
27 213 436 329 404 401 334 392 248 383 356 206 336
28 343 399 375 382 415 405 217 72 378 386 355 330
29 370 435 397 374 389 215 358 227 386 387 329
30 114 429 428 430 255 235 397 81 110 326 335
31 268 434 342 334 387 217 328
Table 5-10 – Data for Figure 5-19

94
5.8.4 Average monthly irradiance

500

450

400
Average monthly irradiance (W/m2)

350

300

250

200

150

100

50

0
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

Figure 5-20 – Average monthly global irradiance in the plane of a single-axis

tracking PV array

Irradiance (W/m2)
Jan 303
Feb 357
Mar 366
Apr 376
May 337
Jun 306
Jul 300
Aug 301
Sep 330
Oct 319
Nov 308
Dec 295
Annual Average 325

Table 5-11 – Data for Figure 5-20

95
The monthly averaging used to present Figure 5-20 and Table 5-11 (and for subsequent
monthly figures and tables) is based on twelve equal months of 730 hours each. Thus, the
figures shown in Table 5-11 will not exactly match the averages of the columns of
Table 5-10. The annual average is of course consistent at 325 W/m2.

5.9 Ambient temperature

Meteonorm was also used to provide ambient temperature data for the same site in
Eritrea and for the same typical year. The data is hourly and was used within the
Simulink model to estimate the hourly PV array temperature, as described in
section 5.10.2.

5.9.1 Hourly ambient temperatures

40

35
Ambient temperature ( °C)

30

25

20

15

10
1 2 3 4 5 6 7
Day of year

Figure 5-21 – Hourly ambient temperature for the first week of the year-long data
set

Notice that the cloud cover (modelled within Meteonorm) that led to the low irradiance
figures during days 1 and 6, shown in Figure 5-18, has also led to warmer nights, as
shown in Figure 5-21.

96
5.9.2 Average monthly ambient temperatures

35

30
Average ambient temperature ( °C)

25

20

15

10

0
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

Figure 5-22 – Average monthly ambient temperature

Average monthly ambient temperature

(ºC)
Jan 24
Feb 24
Mar 26
Apr 28
May 30
Jun 33
Jul 34
Aug 34
Sep 32
Oct 29
Nov 27
Dec 25
Annual Average 29

Table 5-12 – Data for Figure 5-22

97
5.10 PV array

5.10.1 Two-diode model

IPH RS I
ID1 ID2
RP V

Figure 5-23 – Two-diode model of a PV cell (Gottschalg 2001 page 60)

The two-diode model shown in Figure 5-23 is commonly used to represent an individual
PV cell. IPH is the photo current, which is the part we actually want. The other
components represent losses within the cell; in particular: ID1 represents recombination in
the bulk material, ID2 represents recombination in the space charge region, RP represents
parallel leakage losses and RS represents the series resistance. I and V are the terminal
current and voltage respectively.

Applying Kirchoff’s laws and exponential representations of the diodes yields:

e(V + IRS ) e(V + IRS ) V + IRS

I = I PH − I 01 (exp − 1) − I 02 (exp − 1) −
n1 kT n2 kT RP

(Gottschalg 2001 page 60)

in which:

I PH = AC (C 0 + C1T )G

eV gap
I 01 = AC C 01T 3 exp(− )
kT
5
eV gap
I 02 = AC C 02T exp(−
2
)
2kT

T is the temperature of the cell in ºK.

AC is the area of the cell and Vgap is the band gap voltage of the material. These and the
other parameters were kindly provided by AstroPower ([Link] accessed:
March 04), describing their 120-Wp polycrystalline modules: AP-1206.

98
5.10.2 Simulink model
The two-diode model just described was implemented in Simulink, as shown in
Figure 5-24.

1
TcK
TcK
Iph
2 G
G

TcK
ID1
V+IRs

3 1
V TcK I
ID2
V+IRs

Rs

Rp
IC
[1]

Figure 5-24 – Two-diode model of a PV cell in Simulink

In Figure 5-24, the input TcK is the temperature of the cell in ºK, previously denoted as T
in section 5.10.1. The other symbols are defined as they were in that section.

The model requires an iterative solution, as indicated by the feedback of the current
signal I. Simulink performs this almost transparently using Newton’s method. The block
labelled IC provides an initial value for the iteration.

Next, the PV cell model shown in Figure 5-24 was used to construct a model of a PV
module as shown in Figure 5-25.

99
 1 TaC
TcK TcK
TaC G

NCOT
2 G I 1
Module current
G I
3
Module v oltage V
V PV_CinS
Cells in
PV Cell
series

Figure 5-25 – Simulink model of a PV module

In Figure 5-25, the input TaC is the ambient temperature in ºC, as provided by
Meteonorm discussed in section 5.9. Input G is the irradiance, also from Meteonorm, as
discussed in section 5.8.

The cell temperature TcK is estimated from the ambient temperature and the irradiance
using the Normal Operating Cell Temperature (NOCT) method (Markvart 1999 page
88).

Dividing the module voltage by the number of cells in series provides the cell voltage. A
higher layer in the Simulink model hierarchy calculates the array voltage from the
module voltage in a similar fashion. This approach makes no allowance for mismatch
losses, which leads the PV-module model to give higher output predictions than those
shown on the manufacturers’ data sheet. It also means the array predictions are on the
optimistic side. Measurements and modelling of PV arrays composed of undamaged
production-run modules operating at uniform temperature and irradiance (totally
unshaded) showed mismatch losses below 0.53 % (Chamberlin et al. 1995).

100
 2 TaC
TaC Impp
1
1 G I em
G

V
0.02
PV Module P Vmpp
Solve
em f (z) z 2
f(z) = 0

Figure 5-26 – Simulink programme to locate maximum power point (MPP)

Figure 5-26 illustrates how Simulink was used to locate the maximum power point (see
section 2.5.4). The algorithm relies on the fact that, at the maximum power point,
dP/dV = 0. The Solve block, towards the right of Figure 5-26, is a standard Simulink
block and uses Newton’s method to adjust its output z until its input f(z) equals zero. In
this application, its output z is the module voltage and is fed back round to the left of the
diagram. An arbitrary small constant 0.02 is subtracted from this voltage and the thick
vertical bar combines the two signals into a vector of two voltages, indicated by the thick
arrow line connected to the input V of the PV Module block. When presented with this
vector input, the PV Module block simply produces a vector output I, which contains the
two corresponding currents. Multiplying the two voltages by the two currents provides
two powers in the vector P. The difference of these powers is fed to the solve block. The
solve block locates the voltage where the difference of the two powers calculated at the
two voltages is zero. This is where dP/dV = 0.

Since the I-V characteristic of the modelled PV module is smooth, the algorithm just
described has no difficultly in locating the theoretical maximum power point. This again
leads the model to be optimistic, since it makes no allowance for error in the maximum
power point tracker. Chapter 9 will discuss maximum power point tracking in practice.

5.10.3 Electrical power available from PV

The irradiance and ambient temperature data, described in sections 5.8 and 5.9
respectively, was used as input to the Simulink model just described. The two-diode
model parameters were kindly provided by AstroPower ([Link] accessed:

101
March 04), describing their 120-Wp polycrystalline modules: AP-1206, and the results
presented in this section are for an array of 20 modules, ie. a 2.4-kWP PV array, which is
the size chosen for the completed design shown in section 6.4.

The results presented here represent the available power from the PV; it is always
possible, though wasteful, to use less than the available power.

The model performs the calculations on an hour-by-hour basis and gave an annual total
of 5713 kWh.

[Link] Electrical power available – hourly

2500

2000
Electrical power available (W)

1500

1000

500

0
1 2 3 4 5 6 7
Day of year

Figure 5-27 – Electrical power available – hourly for the first week of the year-long
data set

The electrical power available from the PV array, shown in Figure 5-27, is a function of
the irradiance, shown in Figure 5-18, and the ambient temperature, shown in Figure 5-21.

102
[Link] Power available versus irradiance

2500

2000 Ambient temperature: minimum 15°C

Electrical power available (W)

1500
Maximum 42°C

1000

500

0
0 200 400 600 800 1000 1200
2
Irradiance (W/m )

Figure 5-28 – Electrical power available versus irradiance

Figure 5-28 illustrates that power available from the PV is roughly proportional to
irradiance, but decreases significantly as the ambient temperature increases.

103
[Link] Affect of ambient temperature

2500

2000
~1000 W/m2
Electrical power available (W)

1500 ~800 W/m2

~600 W/m2
1000

~400 W/m2

500
~200 W/m2

0
15 20 25 30 35 40 45 50
Ambient temperature ( °C)

Figure 5-29 – Electrical power available versus ambient temperature

In Figure 5-29, the irradiance levels 200, 400, 600, 800 and 1000 W/m2 are chosen for
illustration only. The data plotted for each level includes +/– 10 W/m2. Fitting a straight
line to the 1000-W/m2 data shows a temperature coefficient of around – 0.7 % per Cº.

104
[Link] Average monthly electrical power available

800

700
Average electrical power availible (W)

600

500

400

300

200

100

0
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

Figure 5-30 – Average monthly electrical power available from PV array

Electrical power available from PV array

(W)
Jan 623
Feb 730
Mar 747
Apr 759
May 674
Jun 605
Jul 585
Aug 585
Sep 653
Oct 638
Nov 625
Dec 605
Annual Average 652

Table 5-13 – Data for Figure 5-30

105
Figure 5-30 illustrates, as expected, that the monthly average power available from the
PV is has a similar profile to the monthly average irradiance, shown in Figure 5-20, but is
depressed during the summer due to the high ambient temperatures shown in
Figure 5-22.

5.11 Seawater
The energy consumption of real-world seawater RO systems is affected by the
concentration of the seawater feed, which varies from place to place and from time to
time. Even in laboratories, the concentration of the seawater has no universally accepted
standard, and is complicated by the variety of units employed.

5.11.1 Concentration: ppm versus mg/L

Oceanographers tend to express the concentrations of salts in seawater in weight-to-
weight ratios, such as percent (%), parts per thousand (‰ or g/kg) or parts per million
(ppm). Chemists, on the other hand, use weight-to-volume ratios, such as mg/L. The
desalination industry uses both.

To convert from ppm to mg/L, one should multiply by the density (specific gravity), but,
since the density of water is close to 1 kg/L, concentrations in ppm and mg/L are
numerically similar, and the conversion is often overlooked. At low concentrations,
associated with the product water, or even brackish feed water, the error is small and has
little consequence in the field of desalination. With seawater however, the density is
around 1.024 kg/L at 20 ºC (El-Dessouky et al. 2002 page 527), so for example
35,000 ppm is 35,840 mg/L. Interchanging the units without performing the calculation
is perhaps acceptable when quoting ballpark figures to the nearest 1000, but for testing
and modelling of energy efficiency the conversion should be done.

5.11.2 Salinity versus TDS

During the 1800’s, salinity was defined as the total amount of dissolved material in
grams in one kilogram of seawater. Because the units are in the definition, salinity is
dimensionless, and is correctly written without units. Nonetheless, salinity by this
definition is obviously numerically equal to the total dissolved solids (TDS) in parts per
thousand.

106
Unfortunately, total dissolved solids is very difficult to measure directly, and so
oceanographers have redefined salinity to suit available indirect measurement methods,
and have done so several times over the last century as more accurate measurement
methods have been developed (Stewart 2002). The result is that salinity is approximately
0.995 of TDS in ‰ (Millero 1996; Tonner 1999). Thus, seawater with 35,000 ppm TDS
has a salinity of around 34.825. The difference between TDS and salinity is rarely a
concern in desalination, and the word salinity is frequently used when discussing figures
quoted in TDS.

5.11.3 Typical concentrations of seawater

The average salinity of the world’s oceans is 34.72 (Snoke 2002), which equates to
35,750 mg/L TDS at 20 ºC, and the vast majority of seawater worldwide is close to this
concentration: TDS between 35,300 and 36,100 mg/L. There are various descriptions of
standard seawater in the literature (Millero 1996; Tonner 1999; DOW 1995; ASTM
1998), and most lie in this range. Costal waters can have significantly lower
concentrations, particularly near rivers, but such areas are unlikely to need water through
desalination. Some seas have significantly higher concentrations, notably the Red Sea
and Arabian Gulf, which can be up to 45,000 mg/L (DOW 1995 section 3.2).

5.11.4 Composition of seawater

Although the total dissolved solids in seawater can vary significantly, as discussed
above, the relative concentrations of the major constituents, shown in Table 5-14, are
virtually constant throughout all of the world’s oceans.

107
 g/kg mg/l
Chloride 18.9799 19441.1
Sulfate 2.6486 2713.0
Bromide 0.0646 66.2
Bicarbonate 0.1397 143.1
Fluoride 0.0013 1.3
Sodium 10.5561 10812.6
Magnesium 1.272 1302.9
Calcium 0.4001 409.8
Potassium 0.38 389.2
Silica 0.00201 2.1
Strontium 0.0133 13.6
Orthoborsic Acid 0.026 26.6
Others 0.00135 1.4
Total Dissolved Solids 34.48496 35322.9
Water 965.51504
Total 1000 1024.3
Salinity 34.325
Specific Gravity @ 20°C 1.0243
Table 5-14 – The major constituents of seawater (Lyman and Fleming via Tonner
1999)

5.11.5 Osmotic pressure

For the study of energy consumption in a reverse-osmosis system, the critical
characteristic of the feed water is its osmotic pressure. Rudimentary texts suggest that
osmotic pressure may be calculated by van’t Hoff’s equation:

π = nRT

where π is the osmotic pressure, n is the concentration, R is the gas constant, and T is the
temperature. Unfortunately, this is only accurate at very low concentrations (Reid 1966).
For stronger solutions containing mixed species, such as seawater, the calculations are
more complex. ROPRO (Koch Membranes’ RO systems design software: Koch-ROPRO
2000) uses experimentally determined osmotic pressure coefficients for each species.
These are multiplied by the species concentration and summed to provide an estimate of
the overall osmotic pressure of the solution. ROPRO was used to calculate the osmotic
pressures of three standard seawaters and one other.

Lyman and Fleming’s data (Table 5-14) is at the low-concentration end of typical
seawater: TDS = 35323 mg/L and gives 25.4 bar.

108
Data for seawater at the other end of the range (Millero 1996) has TDS = 36121 mg/L
and gives 26.0 bar.

Standard seawater according to DOW (1995) has TDS = 35617 mg/L and gives 25.6 bar.

Water from a particular borehole well in Massawa, Eritrea, off the Red Sea has
concentration much higher than the average. It has a TDS of 40149 mg/L and gives
28.85 bar.

5.11.6 Straight NaCl solution

There is a recipe for the preparation of artificial standard seawater (ASTM 1998), but for
the study of energy consumption in an RO system, it is sufficient to work with straight
NaCl solution, provided the concentration is adjusted accordingly. Indeed, several of the
manufacturers (Koch, Dow and Osmonics) specify their membranes in terms of straight
NaCl solution. In particular, Koch Membranes are specified with NaCl at 32,800 mg/L,
which is isosmotic with ASTM standard seawater.

Using ROPRO with NaCl at 32,800 mg/L gives 25.8 bar which lines up with the
standard seawaters calculated previously.

To convert this to ppm, the density of straight NaCl solution (not seawater) at 20 ºC is
given, by:

ρ = 0.6985 × C + 0.99829

where ρ is the density in kg/L and C is the concentration in mg/L (Weber-Shirk 2003).

Thus for C = 32,800 mg/L, ρ = 1.021 kg/L, and 32800 mg/L / 1.021 kg/L = 32,125 ppm.

5.11.7 Summary
The average salinity of seawater is 34.72, which equates to 35,750 mg/L TDS at 20 ºC,
and has an osmotic pressure of around 25.7 bar. For testing and modelling RO energy
consumption, Koch use straight NaCl solution at 32,800 mg/L, which is 32,125 ppm and
has a very similar osmotic pressure. Various concentrations have been used for testing
and modelling of the PV-RO system and will be noted where appropriate.

109
5.12 Feed-water temperature

Figure 5-31 – Red Sea water temperature profile provided by Dulas Limited

A temperature profile for water in the Red Sea was provided by Dulas Limited in the
form of the bitmap shown in Figure 5-31. This was converted to numeric data by treating
the bitmap image as a matrix within MATLAB. Since the PV-RO system is designed to
operate from a beach well, rather than an open-sea intake, it was assumed that the short-
term temperature fluctuations would be smoothed out and that only the seasonal
variations were important. The six years of data were superimposed and a sine curve was
fitted, as shown in Figure 5-32.

 day of year − 118 

Feed water temperature = 25 + 8 sin  2π 
 365 

110
 40

35
Seawater Temperature ºC

30

25

20

15

10
1-Jan 1-Apr 1-Jul 1-Oct 31-Dec

Figure 5-32 – Red Sea water temperature: data derived from Figure 5-31 and fitted
sine curve

5.13 Reverse osmosis membrane elements

The general principle of reverse osmosis was outlined in section 1.2, and an introduction
to the technology was presented in section 2.7. This section will discuss the development
of a Simulink model for use in the design of the PV-RO system.

5.13.1 Model structure

In order to predict the flow and concentration of product water for the complete PV-RO
system, a model of an RO membrane element is required. The required model is similar
to those included in the membrane manufacturers’ software packages such as ROSA
(DOW 2000) and ROPRO (Koch-ROPRO 2000), except that it must fit within the
complete Simulink PV-RO system model, so that operation at variable flow and pressure
can be predicted. Also, it should be accurate over the full range of flows and pressures
that occur in a batteryless PV-RO system.

111
The flow and concentration of product water from an individual membrane element is a
function of the flow, pressure, concentration and temperature of the feed water, but, in
order that completed membrane-element models can be connected together to model an
array, care is required in selection of inputs and outputs of the Simulink block
representing the RO element.

Element 1 Product Mix

Pf Q1
Qf
C1 Q
Qp Q2
Element 2 C2
Cf Cp Q3
C3
Pf
Qc Qf Q4 C
C4
Pc Qp
Cc
Element 3
Cf Cp
Pf
Qc Qf
Pc Qp
Cc
Element 4
Cf Cp
Pf
Qc Qf
Pc Qp
Cc
Cf Cp

Qc
Pc
Cc

Q Flow
Capital letters C Concentration
P Pressure
f feed
Lowercase p product
c concentrate
Figure 5-33 – Structure of the Simulink model of the array of RO elements

The chosen structure is shown in Figure 5-33, and thus the RO element model has:

Inputs Outputs:
Qf Flow of the feed, Qc Flow of the concentrate,
Cf Concentration of the feed, Cc Concentration of the concentrate,
Pc Pressure of the concentrate, Pf Pressure of the feed,
Qp Flow of the product,
Cp Concentration of the product.

112
The pressure of the product Pp and the temperature T are also inputs to the RO element
model, but since they are global variables, they are not passed from one block to the next.

5.13.2 Calculations
Internally, the model follows the structure outlined in the equations set out by
DOW(1995 section 4.11), and upon which ROSA is based. Initially, the model was
constructed exactly according to those equations, but it was found that various
adjustments were required, first, to make the results line up with ROSA, second, and
more importantly, to match up with measured data obtained from the test rig. Throughout
the project, the Simulink model has been refined, with assistance from Marcos Miranda.
He tested various membrane elements over a very wide range of flows, pressures,
temperatures and feed concentrations. Within normal flow and pressure ranges, the
results lined up reasonably well with ROSA and ROPRO, but at very low flows and
pressures, large differences were observed. The model used in the performance
predictions presented in Chapter 6 is as follows.

All pressures are in bar, flows in L/s and concentrations in mg/L.

The model first calculates the small pressure drop in the feed/concentrate channel:

∆P = – 0.26 + 7.61Qf /T + 11.3Qf 2 + 3.36×10-6Cf + 326(T×Pc)-1 – 1.22×105(T×Pc)-2

The pressure of the feed Pf is simply Pc + ∆P, and the average pressure in the
feed/concentrate channel is Pfc = Pc + ½∆P. The osmotic pressure is calculated using:

0.002654 × C × (T + 273.15)
πf = (ASTM 1985).
1000 − C / 1000

In the early Simulink models, the calculation of C included Cc and concentration

polarisation, following the DOW equations. But this creates an algebraic loop and it was
found better to simply to calculate the osmotic pressure of the feed and to accommodate
the concentration increase through extra terms in the product flow calculation. The net
driving pressure (neglecting concentration increase) is then given by:

Pnd = Pfc – πf – Pp.

113
The product flow is then given by:

Qp = 0.00307 + 0.000193Pnd – 1.61×10-7Cf + 5.49×10-5T + 2.01×10-5Pnd×T

+ 2.44×10-5Pnd×Qf×T + 7.9×10-7Cf/T – 3.25×10-10Pnd×T×Cf

The salt passage through the membrane is the product of the product flow and its
concentration and is given by:

Qp×Cp = 5.26 – 7.55×10-5Cf – 0.0209Pnd – 0.422T + 0.00551T 2 + 7.52×10-6Cf×T

+ 0.00163Pnd×T + 5.79Qp/Qf

Lastly, a simple mass balance is applied: Qc = Q f − Q p and Q f C f = Q pC p + QcCc .

5.14 Product tank

1
z
Tank Save value from previous step
capacity
Volume already in tank (m3)
6 0
min
Qp max
1 -K- Volume
Volume in tank
to add 3
L/s tom3/h
this step Consumption During
(m3) (m3/d) daylight
Ov erf low
Shortf all

Integral
Integral Integral

Cumulative Cumulative Cumulative

Volume Shortfall
Overflow Volume that
actually came
Volume that out (m3)
actually
went in (m3)
Mass of salt
Concentration in tank
Mass of salt that came out
that went in
2
Cp
Mass
Save value from previous step of salt
Mass of salt already in tank
in tank
1
z

Figure 5-34 – Structure of Simulink model of product tank

Figure 5-34 illustrates the Simulink model used to represent the product tank. Starting on
the left, Qp represents the flow of product water into the tank, in L/s. This is converted to
m3/h, which is numerically equal to the Volume to add this step in m3, because the

114
time-step used to run the model is 1 hour. This volume is integrated, as the model runs, to
provide the Cumulative Volume.

Next, the Volume to add this step is added to the Volume already in the tank from
the previous step. If this exceeds the Tank capacity, (6m3) an Overflow is calculated
and this is also integrated to provide the Cumulative Overflow.

Continuing to the right of Figure 5-34, the Consumption shown is 3m3/day; the During
daylight block simply doubles this between 8AM and 8PM and gives zero during the
night. The consumption is then subtracted to give the (new) Volume in tank, and any
shortfall is calculated.

The lower half of Figure 5-34 calculates the salt concentration in the tank and, therefore,
supplied to the consumer. Starting again on the left, Cp represents the concentration of
product water as it comes from the RO rig. Multiplying this by the volume of water that
actually went in (after allowing for overflow) gives the Mass of salt that went in. And
dividing this by the volume now in the tank gives the Concentration in tank, which is
the concentration of the water consumed. Multiplying this by the volume of water that
actually came out gives the Mass of salt that came out, allowing the remaining Mass
of salt in tank to be calculated and saved for the next modelled time-step.

115
 Chapter 6 System modelling, optimisation and
performance predictions

6.1 System design strategy

The Simulink component models presented in the previous chapter were assembled into a
complete system model, which will be discussed in this chapter. First, the structure of the
completed model will be discussed, followed by an explanation of how it calculates the
capital cost of the system and annual performance figures. These figures are used later in
the chapter, in particular for optimising the RO and PV array sizes with respect to the
capital cost.

6.1.1 System model structure

The general strategy for interconnecting the component models within Simulink was
discussed in section 5.1. Now, the completed model will be discussed, starting with the
top layer of the hierarchy shown in Figure 6-1.

Hourly data PV Array RO Rig Tank

G(W/m2) G

Pa Pa Qp Qp
Ta (degC) Ta

Ts (degC) Ts Cp Cp
Sea Temperature

Figure 6-1 – Top layer of the hierarchical Simulink model of the complete system

In Figure 6-1, the Hourly data block includes the irradiance G and the ambient
temperature Ta, both derived from Meteonorm as described in sections 5.8 and 5.9. This
data feeds the PV Array model, described in section 5.10, which calculates the available
power Pa. The seawater temperature Ts is modelled as described in Section 5.12. The
RO Rig block uses the available power and the seawater temperature to predict the flow

116
 of product water Qp and its concentration Cp. The tank integrates this flow of product
water, taking into account the water consumption, as described in section 5.14.

Controller

1 Pa
Pa P Inverter P Motor
Plunger Pump
f P* f* Pdc T n n T

Pi Qi
Pp Iac Vac V I
Po Qo

pf f f pf

Ts
M Inverter M Motor Moineau Pump

f M* f* Pdc T n n T Clark Pump

Split

P Q Qf Pf
Pm Iac Vac V I
Combine RO Array

Ph Qh Qf Pf
pf f f pf

Qc Pc Pc Qc

Qp
Qp 1
Cp
2 T Cp 2
Ts

Figure 6-2 – Simulink model of the RO rig

Figure 6-2 shows the contents of the RO Rig block shown in Figure 6-1. The Controller
shares the power available Pa between the two inverter/motor/pumps in order to
maximise the water production, as described in section 6.2.3. It does this by providing a
frequency setpoint signal f* to each of the two inverters. In return, the inverter models
provide signals Pdc that represent the DC power drawn. The controller model ensures
that the sum of these two equals the power available: Pp + Pm = Pa.

Each inverter is modelled as described in section 5.7.1 and provides signals Va and f
representing the ac voltage and its frequency. These signals, together with the torque T
from the respective pump, are inputs to the motor model blocks, which are as described
in section 5.6.

117
Each motor block has an output n, which is the shaft speed and is fed to the respective
pump. The motor blocks also provide outputs I and pf, which are the ac current drawn by
motor and its power factor respectively, and are fed back to the inverter model blocks.

The Moineau pump, Plunger pump, Clark pump and RO Array are as shown in
sections 5.4.2, 5.3.1, 5.2.3 and 5.13.1 respectively.

The blocks labelled Split and Combine may be understood with reference to Figure 1-3.
In particular, the Split performs a simple subtraction, representing the fact that the feed
flow to the Clark pump equals the flow through the Moineau pump less that through the
Plunger pump. Likewise, the Combine block represents the addition of the Plunger
pump outlet flow to that of the Clark pump.

6.1.2 Modifying the system model

The completed system model just described represents the final design, which was used
for the performance predictions presented later in this chapter (section 6.5) and from
which the new test rig was built. In order to develop this design, the system model was
modified many times. A critical feature of the model structure is that components, such
as pumps and RO modules etc., can be added, deleted and reconnected simply by
manipulating the blocks and interconnections of the graphical interface. In order to assess
the costs and benefits of alternative configurations, two additional MATLAB scripts
(programmes) were written and linked to the Simulink model. The first deals with the
capital costs, while the second provides annual performance figures and relates these to
the capital cost.

6.1.3 Capital cost modelling

Since there are no fuel costs, the total cost of water in a PV-RO system is largely
determined by the capital cost and the main aim of the optimisation presented in this
thesis was to minimise the capital cost with respect to the predicted annual water
production.

Each Simulink component model (each block in Figure 6-2 for example) contains a value
representing its cost. Each time the model is run, a MATLAB script collates the costs for
all the components in use and provides a summary table such as Table 6-1.

118
 20 PV modules at £369 each £7,380
4 RO membrane elements at £330 each £1,320
4 pressure vessels at £369 each £1,476
Clark Pump - Spectra £1,200
Inverter (for Moineau) - FID 1.5kW £94
Motor (for Moineau) - Toshiba PE £309
Moineau Pump - Netzsch 021 £710
Inverter (for Plunger) - FID 1.5kW £94
Motor (for Plunger) - Toshiba PE £309
Plunger Pump - Cat 237 £380
Miscellaneous Costs £5,000

Complete system capital cost £18,272

Table 6-1 – Capital cost modelling estimates summary table

The data presented in Table 6-1 is that for the completed design. The costs shown were
estimated in early 2001 as follows.

The estimated cost of the PV modules (£369 each) was obtained from Dulas Limited. It
was for a typical 120 Wp module (no particular manufacturer) and includes an element to
cover a single-axis solar-trajectory tracking system.

The cost shown for the RO membrane elements (£330 each) was based on a quotation for
Koch Seawater High-flow TFC1820HF 4” by 40” elements. The price paid later in 2001
for elements of this type for the new test rig was £270 each plus VAT.

The cost shown for the pressure vessels (£369 each) was based on a quotation for 4” by
40” vessels. The price paid later in 2001 for four such vessels for the new test rig was
£296 each plus certification, delivery and VAT. In the future, it would be well worth
considering use of longer vessels that would hold either two or four membrane elements
each, since they are much cheaper overall.

The cost shown for the Clark pump was based on discussions with the manufacturers
Spectra Watermakers Inc. The Clark pump is not normally sold separately.

The cost shown for the inverters (£94 each) was based on a quotation for standard 1.5kW
industrial drives, in particular the FID1000 FKI-12150. The price paid later in 2001 for
an inverter of this type was £157 plus VAT.

The cost shown for the motors (£309 each) was discussed in sections 5.6.5 and 5.6.6.

119
The cost shown for the Moineau pump (£710) was based on a quotation for a Netzsch
NM021SY02S12B in 316 stainless steel. The price paid later in 2001 for a pump of this
type was £745 plus VAT.

The cost shown for the Plunger pump (£380) was based on a quotation for a Cat 237.

The £5000 shown in Table 6-1 for Miscellaneous Costs is an estimate provided by Dulas
Limited and is intended to cover:

• pipe work, filters and interconnections,

• a frame on to which the components may be mounted,
• sensors, wiring and the controller,
• manufacture.

Lastly, Table 6-1 shows the Complete system capital cost, which is simply the sum of the
above components. It was this figure that was minimised with respect to the predicted
annual water production in the optimisation presented in this thesis.

6.1.4 Annual performance figures

In order to assess alternative system configurations and component sizes, the Simulink
system model was run many times. Each run included hourly calculations for one whole
year (8760 steps) of operation. The input data for all runs was that for Massawa, Eritrea,
as discussed in sections 5.8 (irradiance), 5.9 (ambient temperature) and 5.12 (feed-water
temperature), but with a feed concentration of 35,000 mg/L NaCl, as discussed in
section 6.5. The output data includes hourly values for every input and output of every
component (discussed in Chapter 5) and for the system as a whole. This data was
summarised (by a Matlab script) after each run of the Simulink model, to provide a table
of annual performance figures such as that shown in Table 6-2.

120
 Annual energy available from the PV 5713 kWh

Annual energy consumption

Insufficient to start 20 kWh
Excess (not used) 701 kWh
Losses
Inverter (for Moineau) 163 kWh
Inverter (for Plunger) 153 kWh
Motor (for Moineau) 365 kWh
Motor (for Plunger) 369 kWh
Moineau pump 854 kWh
Plunger pump 200 kWh
Clark pump 330 kWh
Membranes 1446 kWh
Desalination 1113 kWh

Minimum monthly-average product flow 3.37 m3/d

Annual-average recovery ratio 33 %
Tank concentration mean 300 mg/L
Tank concentration maximum 576 mg/L

Annual product volume 1424 m3

Complete system capital cost 18272 £
Capital cost per daily water production 4685 £/m3/day

Table 6-2 – Annual performance figures summary table

The data presented in Table 6-2 is that for the completed design. All other designs
considered had a higher Capital cost per daily water production (last row).

The first row of Table 6-2 shows the annual energy available from the PV is 5713 kWh,
which agrees with that shown in section 5.10.3.

The second section of Table 6-2 shows how this energy is consumed throughout the
system. The 20 kWh recorded as Insufficient to start is the annual sum of the energy not
used during the hours (typically at dawn and dusk) during which the power available
from the PV is insufficient to run the RO rig. Likewise, the Excess recorded is the annual
energy not used because the power available from the PV during some hours (typically
around midday) is above the maximum that the RO rig can utilise, because pressure or
flow limits are reached. The losses shown for the inverters, motors, pumps and
membranes are those calculated from the component models presented in Chapter 5. And
lastly, the figure shown for Desalination may be regarded as the useful output from the
whole system; it is the theoretical minimum energy required to desalinate the predicted

121
volume of product water from seawater, as discussed in 2.3.1; it is calculated from the
osmotic pressure of the feed and the flow of the product.

Insufficient to
start Excess (not
used)
Desalination
Inverter (for
Moineau)
Inverter (for
Plunger)

Motor (for
Moineau)

Motor (for
Plunger)

Membranes

Moineau pump

Clark pump Plunger pump

Figure 6-3 – Pie chart of annual energy consumption in the completed system design

The pie chart of Figure 6-3 illustrates the annual energy consumption data from
Table 6-2 and clearly shows the proportions of the total available energy consumed in
losses in the various components. Considering the data in this way gives a better view of
component performance within the system than considering the individual component
efficiencies.

Returning to Table 6-2, the next section includes: Minimum monthly-average product
flow, Annual-average recovery ratio, Tank concentration mean and Tank concentration
maximum. These parameters were not optimised during the design process but were
checked, for each of the designs considered.

The last section of Table 6-2 shows the data that was used to optimise the system design.
The Annual product volume is the sum of the hourly product flow data from the Simulink
model discussed in section 6.1.1. The Complete system capital cost is calculated as

122
described in section 6.1.3. And lastly, the Capital cost per daily water production is
Complete system capital cost
given by: . It was this figure that the design process sought
Annual product volume 365
to minimise.

6.2 Review of configuration ideas

The early designs were based on the Danfoss hydraulic motor as described in Chapter 3,
but this was dropped in favour of the Clark pump, primarily because its water-to-water
efficiency is substantially higher as discussed in section 3.3. Additional reasons were that
its energy efficiency is maintained over a very wide range of operation, which is
especially important in a batteryless PV-RO system, and that the Clark pump was
designed for seawater compatibility from the outset. The configurations discussed in this
chapter are all based on the Clark pump.

The Clark pump is a high-tolerance precision machine (Spectra 2003); its cost reflects
this, and is a significant part of the final design cost, see section 6.1.3. Thus, the design
process is largely a matter of building a system around a Clark pump that makes best use
of that Clark pump. The Clark pump is manufactured in only one size, though it is now
available with different diameter rods. The standard Clark pump has a rod-to-piston area
ratio of 10 % and Spectra recommend that the feed flow should not exceed 3.5 US
gallons per minute, which is 0.22 L/s.

6.2.1 Single feed pump

The simplest configuration of a Clark pump in an RO system was shown in Figure 2-11.
It requires only one motorised pump, and a single 2.5-inch by 40-inch membrane
element. A second similar membrane element may be added to increase water
production, but given the 10 % fixed ratio and the feed flow limit mentioned above, the
product flow is limited to 0.022 L/s, which is 1.9 m3/day assuming 24-hour-a-day
production. In a PV-powered system without batteries, production would typically be
less than 0.6 m3/day, assuming that the PV array is sized so that it is well utilised. A
larger PV array would extend the average equivalent working day beyond 8 hours but
would be under utilised around midday due to the flow limit. Referring to Table 5-9, the
ratio of average irradiance to midday peak is around 0.26 for a fixed array and 0.37 for a
tracking array. However, there will be reduction of pumping efficiency at dawn and dusk

123
and so an average equivalent of 0.33 (8 hours a the day) is a reasonable first estimate of
best performance.

The sheer simplicity of the system in shown in Figure 2-11 is very attractive, but, in
order to increase product flow per Clark pump and to achieve low specific energy
consumption over a wide operating range, a second motorised pump is required.

6.2.2 Delayed injection – single motor

High-
pressure
pump

Motor
RO membranes 1-3

RO membranes 4-6
Medium- Clark
pressure pump
pump

Figure 6-4 – Delayed injection – single motor

Recognising the need to increase product flow per Clark pump, Spectra proposed the
configuration shown in Figure 6-4. The introduction of the second pump increases both
the feed flow to the membrane array and the recovery ratio. The new recovery ratio can
be adjusted by the ratio of the pump displacements, assuming they are positive-
displacement pumps. The feed from the second pump in Figure 6-4 does not go to the
first membrane element, but rather is delayed and injected before the fourth. The thinking
here is that this reduces the build up of concentration through successive membrane
elements.

The delayed injection arrangement was modelled in Simulink. The results presented here
are for a system with a total of five 4” by 40” membrane elements.

124
 Lowest
Annual Annual average
module
product product Highest module
concentrate
volume concentration recovery ratio
flow
(m3) (mg/L)
(L/s)
Injection before
1409 308 12.8 0.22
1st module
Injection after
1408 311 16.4 0.21
1st module
Injection after
1402 318 0.18
2nd module
Table 6-3 – Delayed injection modelling results

The results shown in Table 6-3 illustrate that best overall performance, in terms of both
the product volume and its concentration, is achieved with the injection before first
module, that is at the beginning of the train, without any delay. The reduction of product
volume and the increase of its concentration, caused by the delay of the injection until
after the first module, is insignificant; but, the increase in highest module recovery ratio
is not desirable, and is a direct result of the first module having reduced cross-flow. This
undesirable reduction is also indicated by the reduction in lowest module concentrate
flow. Delaying the injection until after the second model only worsens the situation. The
concentration builds up more quickly in the first two modules, and, as far as the second
three are concerned, it did not much matter whether the injected water has been through
the first two modules or not. The idea of delaying the injection was dropped.

The model also showed that the fixed recovery ratio imposed by having the two pumps
on the same motor was not ideal for optimising performance over a broad range. The
model was modified to give each pump a separate motor. Being smaller, these two
motors had lower efficiency than the previous one, but the ability to vary the recovery
ratio allowed the overall efficiency, over a range of flows, to be improved.

The next alteration was to move the feed for the second pump to the outlet of the first.
This gives the first pump the sole responsibility of getting water to the rig; it would allow
the first pump to be a submersible borehole pump if necessary and simplifies filtering
arrangements and so on.

It seems that Spectra had followed a similar course of development (Smith 2000
Figure 3).

125
6.2.3 Two motor – variable recovery ratio
High-
pressure
pump

Medium-
pressure
pump RO membranes

Clark
pump
Motor

Figure 6-5 – Two motor – variable recovery ratio

A system using two variable-speed motors is shown in Figure 6-5, and this is the general
arrangement finally adopted for the PV-RO system. The independent control of the two
pump speeds provides control over the water recovery ratio, and this is especially
valuable in a batteryless PV-RO system because it enables the water production to be
maximised as the available sunlight varies through the day. The next step was to
determine how the recovery ratio should be varied in order to achieve this maximum.

To investigate this, the Simulink model was programmed to step through a 3-D array of
permutations of speeds, for the two pumps, and of feed-water temperatures. Sorting the
results according to the product flow for a given electrical input power revealed that the
recovery ratio should be varied as shown in Figure 6-6.

126
 40

35

Increasing
30 feedwater temperature
17°C - 33°C
Recovery ratio (%)

25

20

15

10

0
0 200 400 600 800 1000 1200 1400 1600
DC power available from PV (W)

Figure 6-6 – Optimum recovery ratio versus DC power available from PV

Figure 6-6 shows how the water recovery ratio must be varied, in order to maximise the
product water flow as the available power varies. When the available power is low, 200-
300 Watts, the system operates with a low recovery ratio: 10 %. This is achieved by the
running only the medium-pressure pump. As the available power increases, the high-
pressure pump is started and its speed ratio, against that of the medium-pressure pump, is
increased. The optimum recovery ratio also depends on feed-water temperature, as
illustrated towards the right of Figure 6-6. Variation of feed-water concentration and
membrane condition will have a similar effect.

6.2.4 Solar-thermal feed-water pre-heating

The final PV-RO design presented in this thesis does not include pre-heating of the feed
water, but it was considered, as follows.

It is well known that the product flow of an RO system increases with an increase of feed
water temperature (Byrne 1995 page 12). The product concentration also increases, but,
if this can be tolerated, then pre-heating of the feed water can be useful. Indeed, pre-
heating is a recognised technique in RO systems on boats, where waste heat from the

127
main engine may be used. With this in mind, the potential benefit of solar pre-heating of
the feed water in a PV-RO system was considered. It was quickly apparent that a very
large area solar collector would be required to significantly raise the temperature of the
feed water. For example, to raise the temperature of 0.5 L/s of seawater from 15 °C to
35 °C requires:

0.5 L/s × 20 °C × 4.0 kJ/L°C = 40 kW

The 4.0 kJ/L°C is the specific heat of typical seawater (35,840 mg/L,
35,000 ppm) around 25 ºC (El-Dessouky et al. 2002 page 529).

To address this, a system of heat exchangers to recover heat from the concentrate and
product flows was drafted and modelled.

Freshwater 50° <1bar Solar Collectors

35°
35° <1bar <1bar

H/E Cat 35°

15° 30° 70bar
15° 35°
Moineau
35°
10bar H/E Clark Membranes
45° 70bar
20° 20°
Concentrate

Flows = line thickness: 10, 20 or 30 L/m at maximum

Figure 6-7 – A solar-thermal pre-heating configuration that was considered

In the absence of a brine-stream energy recovery mechanism, the model indicated that
the quantity of product water would be increased by around 60 % by pre-heating the feed
water from 15 °C to 35 °C. This suggested that the cost and additional complexity of the
solar-water heater and heat exchangers would be justified. However, the improved
efficiency occurs because the recovery ratio is increased and so less energy is wasted in
the brine-stream. Once brine-stream energy recovery in the shape of the Clark pump was
added to the model, the effect of feed water temperature was very much reduced, and it

128
became obvious that pre-heating of the feed water was much less significant. The effect
of feed water temperature within the completed system design is shown in Figure 6-11.

6.3 RO and PV array sizing

Having selected the general arrangement shown in Figure 6-5, pumps and motors were
chosen as described in sections 5.3 through 5.7. The remaining task was to adjust the
sizes of the RO and PV arrays to minimise the cost of water.

6.3.1 RO array sizing

When designing an RO system that will operate on mains electricity or diesel power
there is a tendency to specify the smallest number of membrane elements that will
provide the required product flow. This keeps capital costs down but also leads to the
specific energy (kWh/m3) and hence energy costs being higher than they could be.
Systems designed with energy efficiency in mind will usually have a greater number of
membrane elements, because increasing the membrane area reduces the required driving
pressure and therefore the energy consumption. When sizing an RO array intended to be
PV-powered, the optimum size is larger still, because of the dominant effect of the PV
array cost.

The Simulink model of the complete PV-RO system was used to predict the performance
of systems with varying sizes of RO array. Starting with six 2½-inch by 40-inch
membrane elements, the model showed that adding elements consistently reduced the
Capital cost per daily water production, at least as far as eleven elements total. At this
point it was clear that 4-inch by 40-inch membrane elements should be considered.

Number of Annual Annual average

Complete system Capital cost per
4” by 40” product product
capital cost daily water production
membrane volume concentration
(£) (£/m3/day)
elements (m3) (mg/L)
3 1189 17,573 5,397 250
4 1424 18,272 4,685 300
5 1482 18,971 4,673 376
Table 6-4 – RO array size modelling results

Table 6-4 shows that the Capital cost per daily water production is considerably lower
with four 4-inch by 40-inch membrane elements than with three; the reduction is around
13 %. Whereas adding a fifth element gives almost no further improvement: ~ 0.25 %.

129
Table 6-4 also shows that adding membrane elements has a significant effect in
increasing the concentration of the product water. This was not an overriding concern at
the time because in-house membrane testing had always shown very good product
quality in relation to the requirements discussed in section 2.1.3. Nonetheless, it clearly
outweighed the minor cost reduction and an array of four 4-inch by 40-inch membrane
elements was adopted for the design. It is roughly twice the size that would normally be
selected to give the same maximum water production, in a petrol-powered or grid
connected RO system.

6.3.2 PV array sizing

The required size of the PV array is obviously very dependent on solar irradiance at the
site of installation. The data for Massawa, Eritrea, discussed in section 5.8, was taken as
a specific example and the Simulink model was used to determine the size of PV array
that would minimise the Capital cost per daily water production.

Number of Annual Annual average

Complete system Capital cost per
120-Wp product product
capital cost daily water production
PV volume concentration
(£) (£/m3/day)
modules (m3) (mg/L)
16 1275 16,796 4807 329
17 1322 17,165 4737 319
18 1362 17,534 4699 310
19 1396 17,903 4682 305
20 1424 18,272 4685 300
21 1447 18,641 4704 295
22 1466 19,010 4732 291
23 1484 19,379 4767 289
24 1499 19,748 4810 286
Table 6-5 – PV array size modelling results

130
 1600

1550

1500
Annual product volume (m3)

1450

1400

1350

1300

1250

1200
16 17 18 19 20 21 22 23 24
Number of 120-W p PV modules

Figure 6-8 – Annual water production versus PV array size

Figure 6-8 shows, as expected, that increasing the PV array size increases the annual
water production, but the curvature indicates a diminishing return.

131
 4900

Capital cost per daily water production (£/m3/day)

4850

4800

4750

4700

4650

4600
16 17 18 19 20 21 22 23 24
Number of 120-W p PV modules

Figure 6-9 – Capital cost per daily water production versus PV array size

Figure 6-9 shows that the Capital cost per daily water production is minimised with an
array of nineteen 120-Wp PV modules. The vertical scale is chosen to exaggerate the
curve; the cost penalty for having a slightly under- or over-sized PV is actually very
small. In particular the array of twenty modules chosen for the final design increases the
Capital cost per daily water production by less than 0.1 %.

Having twenty modules, rather than nineteen, facilitates mounting, particularly when
solar-trajectory tracking is employed. Also, it would allow modules to be configured in
parallel strings, although, if standard industrial drives are to be employed, it is more
likely that all modules would be in series; see the discussion on drive voltage
requirements in section 5.5.

Even with the impressive efficiency improvements made by the Clark pump and
throughout the RO system, the PV array remains the largest single cost in the PV-RO
system, as shown in Table 6-8. It represents roughly one third of the total capital cost. It
will, however, offset all fuel costs for the whole life of the equipment.

132
6.3.3 Life-cycle costing
The calculations just described, considered only the capital costs of the components. If
full life-cycle costing were employed it is possible that the membrane area would be
reduced slightly, but since the membrane life in an intermittently-operated system is
unknown, section 2.9.3, accurate life-cycle costing is not possible.

133
6.4 Completed design
Photovoltaic
Array

Inverter Controller Inverter Product

Tank
Solar
Tracker
Induction Induction
Motor Motor
Pulsation
Dampener

Pressure
Plunger Relief
Valve
Coarse Fine Pump
Filter Filter
Moineau Reverse-Osmosis
Pump Modules

Pulsation Clark
Dampener Pump

Beach-well
Seawater Concentrate
Intake Discharge

Figure 6-10 – Completed design

Photovoltaic array 2.4 kWp, twenty AstroPower AP-1206 modules,

with single-axis solar-trajectory tracking
Moineau pump Netzsch NM021SY02S12B in 316 stainless steel
Nominal 1.2 m3/h at 900 rpm, 10 bar
Plunger pump CAT 237 in nickel aluminium bronze
Nominal 8.7 L/min at 1725 rpm, 105 bar (used up to 69)
Motors (both) 1.5-kW 6-pole high-efficiency induction motors,
Siemens 1LA9 106-6KA10
or Toshiba B0026FLF2AYH
Inverters 1.5-kW standard industrial drives
FID1000 FKI-12150
Reverse osmosis Four 4” by 40” seawater RO membrane elements
modules Koch TFC1820HF
with 69-bar (1000-psi) pressure vessels
Clark pump Spectra Watermakers
Product tank > 6 m3
Table 6-6 – Major parts list for completed design

134
6.5 Performance predictions
The performance of a PV-RO system is greatly affected by:

• the solar irradiance,

• the concentration of the seawater feed,
• the temperature of the seawater feed and
• the ambient temperature.

The correlation of the above factors is also important, particularly between the seawater
temperature and the solar irradiance.

The site at Massawa, Eritrea was used to provide data for the following performance
predictions. Hour-by-hour solar irradiance and ambient temperature data was derived
from Meteonorm, as discussed in sections 5.8 and 5.9. The seawater was represented by
straight NaCl solution at 35,000 mg/L, which is isosmotic with seawater at around
38,000 mg/L. This is significantly higher than typical seawater ~35,700 mg/L, but is still
short of that measured at a seawater well in Massawa: 40,150 mg/L. An annual
temperature profile for the Red Sea was used, showing a summer maximum of over
33 °C and a winter minimum of around 17 °C, as discussed in section 5.12.

135
6.5.1 Product water flow

500

450

400

350 Increasing
ambient temperature Increasing
Product flow (L/h)

300 15°C - 42°C feed-water temperature

17°C - 33°C
250

200

150

100

50

0
0 200 400 600 800 1000 1200
2
Irradiance (W/m )

Figure 6-11 – Predicted product water flow versus irradiance

Figure 6-11 illustrates that the hourly product flow is broadly proportional to irradiance
over a wide working range. According to the model, production starts at around
100 W/m2 and increases steadily up to around 800 W/m2. Above 800 W/m2, production
is limited by the pressure and flow limits of the Clark pump. The two temperature effects
are also indicated in Figure 6-11. In the linear region, between 100 W/m2 and 800 W/m2,
the ambient temperature has a significant effect on the power available from the PV, as
discussed in section [Link], and this causes the scatter in the data. In the region limited
by pressure and flow, the feed-water temperature has the dominant effect. The data in
Figure 6-11 is shaded according to feed-water temperature.

136
 500

450

400

350
Increasing
Product flow (L/h)

300 feed-water temperature

17°C - 33°C
250

200

150

100

50

0
0 500 1000 1500 2000 2500
DC power availible from PV (W)

Figure 6-12 – Predicted product water flow versus DC power available from the PV

As discussed in section [Link], the electrical power available from the PV is roughly
proportional to irradiance, but decreases significantly as the ambient temperature
increases. Thus, Figure 6-12 is similar to Figure 6-11 but with the scatter caused by the
variation of ambient temperature removed.

137
 5

4.5

3.5
Product flow (m3/day)

2.5

1.5

0.5

0
1 2 3 4 5 6 7 8 9 10 11 12
Months

Figure 6-13 – Predicted product water flow – Monthly averages

The total annual predicted water production is 1424 m3, which is an annual average of
just over 3.9 m3/day. The main bars in Figure 6-13 show how this is distributed though
the year. The increase in the summer is due to the increase in seawater feed temperature,
not solar resource.

The dots shown in Figure 6-13 are the results predicted with a fixed PV array; that is
without the single-axis solar-trajectory tracking discussed in section 5.8.1 and which has
been assumed throughout the design. The model was also run with dual-axis tracking and
the results are presented in Table 6-7.

Fixed Single-axis Dual-axis

3
Annual product volume (m ) 1160 1424 1434
Percentage gain 23% < 1%
Table 6-7 – Predicted annual product volume for a fixed PV array and for single-
axis and dual-axis solar-trajectory tracking arrays

138
The annual water production shown in Table 6-7 is slightly less affected by the tracking
arrangement than is the annual average global irradiance shown in Table 5-9.
Nonetheless, the selection of single-axis tracking is confirmed.

5.5

4.5

3.5
Product flow (m3)

2.5

1.5

0.5

0
0 50 100 150 200 250 300 350
Days

Figure 6-14 – Predicted product water flow – Daily totals

Figure 6-14 shows the same data, on a daily basis. There are occasional cloudy days on
which very little water is desalinated and in order to meet a continuous demand, a
generous product tank is required. By adjustment of the modelled tank capacity, shown in
section 5.14, the simulations showed that a tank of at least 6 m3 is required. In practice, it
would be prudent to have a significantly larger tank to allow for even more erratic
weather conditions, unplanned system downtime and variability in consumption.

139
 6

5
Product volume in tank (m3)

0
0 50 100 150 200 250 300 350
Days

Figure 6-15 – Predicted volume in product water tank

Figure 6-15 predicts the volume of water in a product tank of 6 m3, assuming that
consumption is a constant 3.0 m3/day. In practice, consumption is likely to be seasonal
and could be influenced by the level in the tank.

140
6.5.2 Product water quality
At full power, the predicted product water concentration is in the range 170 to 330 mg/L,
depending on the feed water temperature. As the available power reduces, the
concentration of the product water is greatly increased, but, at the same time, the product
flow is reduced. This small volume of poor-quality water will have little effect on the
overall quality of water in a product storage tank.

1000

900

800
Concentration in product tank (mg/L)

700

600

500

400

300

200

100

0
0 50 100 150 200 250 300 350
Days

Figure 6-16 – Predicted salt concentration in product water tank

Figure 6-16 shows the predicted salt concentration in the same 6 m3 product water tank
and with the same constant 3.0 m3/day consumption. This graph assumes that all of the
product water is directed into the tank (no concentration-controlled diverter valve is
included). It shows that the predicted quality of the water in the product tank is generally
good, except after two consecutive cloudy days in August. The concentration of the
product water can be expected to rise slightly as the membranes age. There are various
straightforward techniques that could be used to limit the concentration in the product
water tank, should this prove necessary as discussed in section 10.6.

141
6.5.3 Predicted energy flows – Sankey diagram
Inverter Motor Plunger Pump

1.70
1.33

4.01 3.51 0.11 0.26 0.14

1.42
0.23

Moineau
Inverter

Pump
Motor
RO Membranes
Clark Pump
0.01 0.49 0.78
1.81 0.61
Unused

0.11 0.26 0.60

0.07
0.95
All figures in kWh/m3

3.33

0.23

Figure 6-17 – Predicted energy flows – Sankey diagram

The numbers and line widths shown in Figure 6-17 represent the annual-average energy
flow in kWh per m3 of product water. These figures can readily be converted to annual
kWh simply by multiplying by 1424 m3, which is the predicted annual product volume.

Starting on the left, 4.01 kWh/m3 (5710 kWh) is available from the PV array.

A tiny proportion of this, 0.01 kWh/m3, is unused, representing the rare occasions when
there is insufficient power available to start the rig.

Next, a further 0.49 kWh/m3 is unused, representing the times when there is an excess of
power available: the rig is operating at its maximum flow and pressure. A reduction in
the size of the PV array would reduce this figure and so improve the annual-average
energy efficiency; however, as already noted, the PV array size is chosen to minimise the
overall cost of the water, which is a slightly different goal.

142
Neither of the above unused powers are necessarily losses, since they represent electrical
power that is available for other purposes, such as battery charging for lighting.

Thus, the 3.51 kWh/m3 may be regarded as the total electrical specific energy
consumption for the desalination system.

Next, it can be seen that the inverter and motor efficiencies are good, particularly
considering that these relate to very small induction motors, each delivering around
500 W on average.

The plunger pump is also very efficient. The additional 0.23 kWh/m3 that feeds the
plunger pump represents the work done by the Moineau pump in raising the feed water to
medium pressure.

The Moineau pump itself is less efficient. Its selection is discussed in section 5.4.

Two losses are shown in relation to the RO membranes themselves. First, the
0.07 kWh/m3 represents the slight pressure drop found between the feed and the
concentrate flows. This includes pressure drops in the interconnecting pipe work.
Second, the 0.95 kWh/m3 represents the viscous losses associated with driving the
product water through the membrane.

The 0.78 kWh/m3, on the far right of the diagram, represents the actual desalination.

The massive 3.33 kWh/m3, also emanating from the membranes, represents the energy
contained in the pressurised concentrate water.

The Clark pump recovers this energy and delivers it directly back into the feed flow. It
achieves an annual average of 93 % efficiency, which is excellent.

143
6.5.4 Power usage – average day
Power availible from PV
1800 DC Power used
AC Power to Motors
Shaft Power to Pumps
1600 Hyrdaulic Power to Membranes
Useful Power
1400

1200
Power (Watts)

1000

800

600

400

200

0
0 2 4 6 8 10 12 14 16 18 20 22 24
Time (hours)

Figure 6-18 – Predicted analysis of power usage over an average day

Figure 6-18 illustrates how the power available from the PV array is used throughout an
average day. The average is created from hourly values across the whole year.

As discussed under Figure 6-17, a proportion of the power available from the PV is
unused: either it is insufficient to start the rig or it is in excess because the rig is operating
at its maximum flow and pressure. Thus, the DC Power used in Figure 6-18 is
significantly below that available. The AC Power to Motors is the sum of that going to
the two motors and is below the DC Power used because of losses in the two inverters.
Likewise, the Shaft Power to Pumps is the sum for the two motor-driven pumps, and is
below the AC Power to Motors because of the losses in the two motors. The difference
between the Shaft Power to Pumps and the Hydraulic Power to Membranes is due to
losses in all three pumps: Moineau, plunger and Clark. Lastly, the Useful Power
represents the actual desalination calculated from the osmotic pressure and the product
flow. The difference between the Hydraulic Power to Membranes and the Useful Power
is due to the losses in the membranes themselves.

144
6.5.5 Specific energy

5
Specific energy (kWh/m3)

0
0 200 400 600 800 1000 1200 1400 1600
Electrical input power (Watts)

Figure 6-19 – Predicted specific energy vs. electrical input power

Figure 6-19 shows that the specific energy consumption is typically between 3.2 and
3.7 kWh/m3. Moreover, it is near constant except under very low-power operation. Thus,
the flow of product water is virtually proportional to the power available from the PV
array over a very wide range.

These predictions are very impressive when compared to traditional reverse osmosis
systems. The credit for this goes firstly to the Clark pump but also to the selection of all
the other system components and the recovery ratio control strategy, discussed in
section 6.2.3.

The slight spread of specific energy values for a given electrical input power (apparent in
Figure 6-19) is due to the variation in feed seawater temperature (17 to 33 °C). This
variation in specific energy is actually very small, again due primarily to the Clark pump.

145
6.6 Manufacturer’s cost analysis
A full cost analysis of the completed design was conducted by Dulas Limited (Gwillim
2001). A summary was made public in a joint report (Thomson et al. 2001) and is
repeated below.

Item Cost
Photovoltaic (PV) Array £ 7,319
Reverse osmosis elements and pressure vessels £ 2,796
Pumps (all three) £ 2,290
Motors and Inverters £ 806
Miscellaneous £ 2,537
Total Components and Materials £15,748
Manufacturer’s mark-up on Components and Materials £ 3,150
Manufacturer’s Labour at cost including overheads £ 2,906
Mark-up on Labour £ 872
Total £23,055
Table 6-8 – Capital cost Dulas estimates summary table

The costs shown in Table 6-8 may be compared against those used in the modelling
presented earlier in Table 6-1. The costs are very similar, except for the miscellaneous
category: Table 6-1 showed £5000, while the sum of miscellaneous and manufacturing in
Table 6-8 is £9465. This largely accounts for the difference between the modelled
Complete system capital cost: £18,272 and final estimate: £23,055, stated in the abstract
and introductory chapter.

In order to estimate the cost of water over the lifetime of the equipment, Dulas assumed a
20-year life for the system as a whole, with pump replacements at 5, 10 and 15 years.
They also assumed that a complete new set of RO membrane elements would be required
every 12 months. The estimated overall cost of water, including full maintenance and
using an 8% discount rate, was £2.00 per m3. This figure will reduce significantly if
longer membrane lifetimes prove possible in practice.

A fuller discussion of these cost estimates was presented in the joint report (Thomson et
al. 2001) and is repeated for convenience in Appendix E.

146
 Chapter 7 New test rig hardware

Photovoltaic
Array Reverse-Osmosis Modules

Inverter Controller Inverter

Induction Induction
Motor Motor

Plunger
Pump
Filter
Moineau
Pump

Clark
Pump

Concentrate
“Seawater” Discharge Fresh water product
Intake

Tank

Figure 7-1 – Test rig configuration

The test rig described in Chapter 3 was inflexible and not suited to accommodating the
new design presented in section 6.4. A new test rig was constructed during 2002, with
much assistance from Marcos Miranda. The arrangement is shown in Figure 7-1, and is
very similar to the design, except for the following details.

A tank holds the “seawater”, which is actually straight NaCl solution. Both concentrate
and product water are fed directly back into the tank so as to maintain the feed
concentration during extended testing and the tank is continuously stirred to ensure that
concentration is even. Despite this, the feed concentration does vary due to the variation
of concentration of water held within the RO modules, which is affected by the recovery
ratio. The new test rig has four 4-inch by 40-inch membrane elements, and these hold a

147
significant volume of water compared to the tank. (The volume of four 4-inch by 40-inch
pressure vessels is 33 L; the tank held 75 L.)

The tank is fitted with a heating and cooling system, as discussed in section 4.13, to
ensure that the temperature of the feed water to the reverse-osmosis membranes is held
constant: 25 °C throughout the tests presented here.

The plunger pump is the CAT 317, taken from the earlier test rig. It is rather larger than
the 237 called for in the design, and therefore, it revolves rather slowly at the required
flow. The motor is a 4-pole induction machine, again left over from the earlier test rig,
and would be very inefficient at such low speeds. So, the test rig employs pulleys and a
toothed rubber belt to allow the motor to run faster. Ideally (as in the design), a smaller
plunger pump would be used, directly connected to a 6-pole induction motor.

The PV array used for testing was the BP Saturn array on the CREST building at
Loughborough. It is a fixed array (no tracker) comprising thirty-two 85 Wpeak BP Solar
Saturn BP585F mono-crystalline silicon modules, but only eighteen were used, as
discussed in Chapter 9.

The Controller shown in Figure 7-1 was implemented in LabVIEW, alongside the data
acquisition system. It provides the speed setpoint signals for the two inverters, and the
development of the control algorithms is described in the next two chapters.

148
 Chapter 8 Recovery-ratio control algorithm

The requirement to operate the batteryless PV-RO system with a variable recovery ratio
so as to maximise the water production was discussed in section 6.2.3, and is to be
achieved through the control of the two pump speeds. The Controller, shown in the
completed design Figure 6-10, provides speed setpoints to the two inverters connected to
the two pump motors.

Likewise, in the new test rig Figure 7-1, the controller provides the speed setpoint
signals. The controller for the test rig was implemented in LabVIEW, alongside the data
acquisition system, and this chapter describes the design of the recovery ratio control
algorithm.

The design of the algorithm was based on measurements taken from the test rig while
using mains electricity. These measurements characterised the system in terms of product
flow in response to the two speed setpoints.

8.1 System characterisation

Using mains electricity for the test (unlimited power supply), the two pumps can be run
at any speed from zero up to their respective maximums. However, the other system
components impose limits as indicated in Figure 8-1. The pressure limit is 69 bar
(1000 psi) and is imposed primarily by the rating of the pressure vessels that house the
RO membranes. The Clark pump flow limit is 795 L/h (3.5 USgal/m). We will see later
that these limits are rarely encountered if the system is operated for best efficiency.
LabVIEW was employed to implement a test procedure that stepped through allowable
combinations of the two pump speed setpoints, whilst respecting the pressure and flow
limits. Operation at each test condition was maintained until the key system variables had
stabilised. The steady-state results were then logged and the test moved to the next point.

149
 481 488
400 443 449 456 463
s i) 410 417 424 431
0p
350 00
r (1 371 379 386 393 400 406
Plunger pump speed setpoint (rpm)

9 ba
6 it 338 346 354 361 367 375 381
300 lim
u re 293 306 314 321 329 335 342 348

)
ss

l/ m
Pre 257 272 281 289 296 304 310 317 323

ga
250

US
229 240 249 256 264 271 278 285 291

5
(3.
200 196 207 216 224 231 239 246 253 260 266

L/h
140 173 183 190 198 206 214 221 228 234 240

95
150

it 7
113 136 150 157 165 172 181 188 196 203 209

lim
97 115 124 131 138 147 155 163 170 177 184

w
100

f lo
75 88 96 103 111 119 127 135 143 150

p
um
50 51 60 69 75 84 93 101 107 115 122

kp
18 26 33 43 50 58 65 72

ar
Cl
0 17 25 34 43 51 58 65 72

0 100 200 300 400 500 600 700 800 900

Moineau pump speed setpoint (rpm)

Figure 8-1 – Measured product flow (L/h) against pump speed setpoints

Inspection of Figure 8-1 confirms that increasing the speed setpoint of either pump
increases the product water flow. Note that these are the speed setpoints, not the actual
speeds, which will be lower due to slip in the induction motors. In particular, this
accounts for the second row up, where the plunger pump speed setpoint is 28 rpm but the
motor is stalled. It is perfectly possible to start the motor at this speed by introducing a
voltage boost in the inverter settings; however, this can reduce efficiency and may or
may not provide an overall benefit, as discussed in section 5.7.2.

8.2 Control strategy

Figure 8-1 shows that the maximum water production is around 488 L/h, and that the
system can be operated over a wide range of pump speeds. But, Figure 8-1 does not
provide any obvious means of optimising the two pump speeds to maximise the water
production with respect to the power consumed. With this in mind, it is interesting to
look at the specific energy consumptions (kWh/m3) shown in Figure 8-2.

150
 3.65 3.71
400 3.67 3.67 3.64 3.66

s i) 3.70 3.68 3.63 3.62

350
00p 3.80 3.74 3.67 3.62 3.58 3.63
(1 0
Plunger pump speed setpoint (rpm)

ar
9b 3.85 3.70 3.65 3.63 3.58 3.57 3.58
300 t6
l imi 4.16 3.84 3.77 3.67 3.60 3.57 3.55 3.59

)
l/ m
re
ssu 4.31 4.01 3.78 3.67 3.62 3.60 3.55 3.57 3.63

ga
250 Pre

US
4.34 4.08 3.89 3.72 3.64 3.60 3.60 3.59 3.60

5
(3.
200 4.64 4.21 4.00 3.80 3.69 3.62 3.61 3.60 3.59 3.65

L/h
5.95 4.56 4.12 3.94 3.85 3.73 3.68 3.66 3.66 3.68 3.74

95
it 7
150
6.70 5.26 4.46 4.18 4.02 3.86 3.74 3.78 3.76 3.77 3.83

lim
6.85 5.17 4.47 4.32 4.13 4.05 3.93 3.92 3.88 3.92 4.00

w
f lo
100
7.22 5.36 4.90 4.63 4.49 4.29 4.20 4.19 4.13 4.20

p
um
7.58 6.38 5.55 5.38 4.99 4.79 4.65 4.64 4.57 4.64

kp
50

ar
22.46 13.78 10.36 8.46 7.66 7.08 6.75 6.66

Cl
0 26.30 8.46 6.66 5.69 5.29 5.06 4.96 5.01
0 100 200 300 400 500 600 700 800 900
Moineau pump speed setpoint (rpm)

Figure 8-2 – Measured specific energy consumptions (kWh/m3) against pump speed
setpoints

In Figure 8-2 it can be seen that efficient operation (less than 5 kWh/m3) is available over
most of the area set by the pressure and flow limits, and that the most efficient operation
is obtained in the central region, well away from those limits. Imagining this data in three
dimensions, one can picture a valley of best efficiency, as indicated by the bold diagonal.
Further plots (not presented here) and manipulation of the data confirmed that following
this valley provides a good control strategy for maximising water production at powers
between 400 and 1800 W. For powers less than 400 W, it is best to use the Moineau
pump alone, as indicated in bold across the bottom of Figure 8-2.

This simple control strategy was implemented in LabVIEW. In particular, the speed
setpoint for the plunger pump was derived from the Moineau pump setpoint, and thus,
the two independent speed setpoints were reduced to one.

151
8.3 Measured performance

6 Measured
Specific Energy Consumption (kWh/m3) Predicted

0
0 200 400 600 800 1000 1200 1400 1600
DC Power (Watts)

Figure 8-3 – Measured specific energy consumption against DC power

Slowly sweeping the Moineau speed setpoint, with the plunger speed under control as
described above, gave the results plotted as points in Figure 8-3. The line is the predicted
characteristic from the model described in section 6.5, and, in particular, derived from the
data in Figure 6-19, but with the spread caused by temperature removed. Clearly, the
measured results align very closely with the predicted. A discussion of differences is
presented in section 10.4.

152
 Chapter 9 Maximum power point tracking (MPPT)
algorithm

The tests described in the previous chapter were conducted using mains electricity to
power the system. The next stage was to operate the rig from the real PV array.

As outlined in section 5.5, the strategy was simply to connect the standard industrial
variable-speed drive inverters directly to the PV array. The first step was to configure the
PV array to provide a suitable voltage. The inverters used in the experimental rig specify
a maximum DC-bus voltage of 400 VDC and this allows use of a maximum of 18 of the
BP-Saturn PV modules available on the roof of the CREST building to be connected in
series. Since the individual Saturn modules are rated at 85 Wp, 18 modules provide a total
array power of 1530 Wp, which is only 64 % of the design value for the PV-RO system.
Other brands of PV modules are readily available that have similar voltage but more
power, and could normally be chosen to provide a better match to the RO system.

Connection of a PV array to a standard industrial inverter is common practice – the

challenge lies in achieving maximum power point tracking (MPPT) as introduced in
section 2.5.4. The next section will outline some of common MPPT algorithms used in
PV systems, and after that, the implementation of MPPT with standard industrial
inverters will be discussed

153
9.1 Common MPPT algorithms

9.1.1 Constant voltage

Perhaps the simplest way to get close to the MPP is to control the current drawn from the
PV array such that the voltage remains constant.

1500

2
/m
W
00
10
2
1000 = /m
W
ce
0
an

80
Power (W)

di
a

2
Irr

m
0 W/
60
2
500 /m
40 0W

2
/m
200 W

0
0 100 200 300 400
Voltage (V)

Figure 9-1 – Indicative power curves for a PV array at 25°C

Inspection of Figure 9-1 suggests that operation at a constant 310 volts will ensure
operation close to the MPP, regardless of variations in irradiance. Indeed it will, but it
will not perform well when the cell temperature changes, as illustrated by Figure 9-2.

154
 1500

2
/m
1000

W
00
10
Power (W)

=
n ce
ia
ad
Ir r

500

0
0 100 200 300 400
Voltage (V)

Figure 9-2 – Indicative power curves for a PV array at 50°C

9.1.2 Constant voltage with temperature compensation

The addition of a temperature sensor attached to one of the modules and wired into the
controller allows the voltage setpoint to be adjusted and makes a useful improvement to
the basic system.

9.1.3 Percentage of open-circuit voltage

When MPPT is implemented in a microprocessor, it is fairly straightforward to briefly
open-circuit the array at regular intervals and to measure the open-circuit voltage. A
fixed percentage of this voltage, usually around 79 %, is then used as the operating
voltage setpoint until the next measurement is made. This technique allows for variations
in cell temperature without need of a sensor.

155
9.1.4 Perturb and observe (hill climbing)
A basic hill-climbing maximum power point tracker requires feedback of the power
being drawn from the PV array; usually this is derived by multiplying measured current
and voltage. The tracker perturbs (makes a small change to) the array voltage (or current)
and observes any change in the power; if the power has increased, it adjusts the voltage
(or current) a little further in the same direction; otherwise, it adjusts it in the opposite
direction. Thus, the tracker steps towards the MPP regardless of which side it starts. Once
it reaches the MPP, it will move back and forth about the MPP, but remain close to it. If
the MPP should move, the tracker will follow it.

9.1.5 Incremental conductance

The incremental conductance method is an alternative implementation of hill climbing.
Instead of multiplying the measured current and voltage together at each end of the
perturbation, the incremental conductance method compares dI/dV against I/V in order to
decide which way to go (Wilamowski et al. 2002).

9.2 Standard industrial variable-speed drive inverters

In the preceding descriptions of MPPT algorithms, it was assumed that the MPPT
controller had direct control of the current drawn from the PV array, and through this, it
could control the voltage. Direct control of the voltage would serve equally well, but with
a standard industrial variable-speed drive inverter, we have neither. All we have is the
speed setpoint input, which provides some control of the current drawn from the PV
array, but not complete control.

Consider a standard industrial inverter and induction motor driving a pump, with the
speed setpoint held constant. If the supply voltage is reduced slightly, either the voltage
to the motor will be reduced, or the inverter will draw more current from the supply in
order to maintain the voltage to the motor. Most modern industrial inverters will do the
latter, at least by default, and this creates a problem for operation from PV. The problem
turns into an opportunity, but we will come back to that later.

Consider that the standard industrial inverter, motor and pump are operating from a PV
array. As the speed setpoint is gradually increased, the motor will accelerate and the
current drawn from the PV array will increase, and as it does so, the voltage will fall

156
slightly as dictated by the I-V curve. The operating point will move upwards and
leftwards along the I-V curve, and similarly along the P-V curve, as indicated by the
arrows in Figure 9-3.

I P
MPP MPP

V V

Figure 9-3 – I-V and P-V curves

All is well, until the operating point reaches the MPP, which obviously corresponds to
the maximum available speed and is exactly the point at which we wish to be.

But, if the speed setpoint is increased slightly further, the extra current drawn by the
inverter will cause an excessive reduction in voltage, and the power (the product of I and
V) will be reduced, as indicated to the left of the MPP in the P-V curve. The inverter will
respond to the reduction in voltage, (no more increase of setpoint is necessary) by again
increasing the current drawn, which will reduce the voltage further, and so on, causing
the voltage to collapse.

Thus, an industrial inverter of this type is only stable while operating on the right of the
MPP. It is possible to operate on the left by rapid control of the speed setpoint, and this
was demonstrated during commissioning of the PV-RO rig, but since the objective is to
be at the MPP, neither left nor right, an alternative strategy was pursued.

9.3 New MPPT algorithm

If the voltage is steady, gently increase the speed setpoint.

If the voltage is falling, rapidly reduce it.

Thus, the problem of voltage collapse is turned into an opportunity to locate the MPP.

157
The word new, in the title of this section is perhaps an overstatement, but it does reflect
the fact that this particular strategy has not been found in the literature. There are
hundreds of academic papers describing MPPT algorithms, many of them claiming to be
new. On close inspection, most of them turn out to be subtle variations of the approaches
outlined earlier in section 9.1. Indeed, the algorithm presented here can be described as a
hill-climber, in which one side of the hill is sheer cliff, so perhaps it should be called the
cliff-hanger algorithm.

An important feature of the algorithm is that it requires only the voltage as an input
signal; the hill-climbers outlined in section 9.1 require the current as well.

9.4 Basic implementation

The algorithm was implemented in LabVIEW using a 100 ms program loop. On each
cycle of the loop, the program measures the PV array voltage v, which is the DC bus
voltage of the inverter, and gives a new speed setpoint n* to the inverter.

Computer

LabVIEW

Buffer DAQ Card

n*
+

v PV Inverter Motor
Array C
–

Figure 9-4 – MPPT hardware/software arrangement

Practical note: As described in section 4.16, the voltage is actually sampled at 10 kHz.
The 100 ms program loop takes 1000 samples at a time, and averages these to provide
good quality data at 10 Hz. In common with other hill-climbing algorithms, good-quality
data is essential. Speeding up the program loop would have little benefit.

158
 Discrete
integrator
∆ N+

Non-linear n*
gain ∆n
d
dt ∆n–
G

v PV Inverter Motor
Array C
–

Figure 9-5 – MPPT basic implementation

Figure 9-5 shows the structure of the software used for a basic implementation of the
algorithm, which was previously stated:

If the voltage is steady, gently increase the speed setpoint.

If the voltage is falling, rapidly reduce it.

∆N+ is a constant that defines the rise rate, while the voltage is steady. The integrator
turns this constant into a ramp, and provides the speed setpoint n* that is applied to
inverter. Thus, the speed ramps up.

The voltage v is fed back and differentiated in software. If dv/dt is zero (steady voltage)
or positive (rising voltage), it has no effect. But if it is negative (falling voltage), it is
passed through to the integrator, which reduces n*, and prevents the voltage from
collapsing.

The if statement is implemented by the non-linear gain box, in which the G represents the
gain when the input is negative. This defines by how much n* is reduced.

Ideally, ∆N+ should be set as large as possible, so that the controller ramps up quickly to
the MPP. Meanwhile, G should be set as small as possible, just high enough to prevent
the voltage from collapsing. If G is set too large, the controller becomes too cautious and
stays a little to right of the MPP. To understand this, bear in mind that the voltage is

159
never perfectly steady; each time it dips a little, the controller reduces the speed setpoint
a little.

In practice, the controller shown in Figure 9-5 did work, but was not ideal. In particular,
it became apparent that G needed to be set to different values under different operational
conditions. When the irradiance was high, G needed to be set to a high value so that
speed setpoint would be reduced sufficiently to prevent voltage collapse. But when the
irradiance was low, such a high value of G made the controller too cautious. Setting G by
trial-and-error was not succeeding and a more rigorous approach was sought.

9.5 Enhanced implementation

Discrete
integrator
∆ P+
C Non-linear dn n*
gain ∆p dp ∆n
d
dt pC ∆p–
G

v PV Inverter Motor
Array C
–

Figure 9-6 – MPPT enhanced implementation

The rate at which the voltage collapses is dependent on the capacitance C, illustrated
between the PV array and the inverter in Figure 9-6. This capacitance includes the
capacitance built in to the industrial inverter and any external capacitance added.

The voltage collapses when the inverter draws more power than the PV can supply.
During the collapse, the additional power is supplied by the capacitor, and calculating
this power will indicate by how much the inverter power must be reduced.

To maintain familiar sign conventions, power into the capacitor is treated as positive and
dv dv
is given by: pC = v iC . We also know that iC = C , and thus pC = v C . This
dt dt
calculation is implemented on the left of Figure 9-6.

160
So, the inverter power must be adjusted by an amount theoretically equal to pC , but the

dn *
inverter has a speed setpoint input, not power. To provide the conversion, is
dp
required, and this of course depends on the load, in our case, the RO rig.

400

350

300
Power consumption p (W)

250
p = 0.50n* + 42
200

150

100

50

0
0 100 200 300 400 500 600
Speed setpoint n * (rpm)

Figure 9-7 – Inverter power consumption vs. speed setpoint

The power consumption of the RO test rig was measured: first, running only the Moineau
pump, and second, running both pumps as discussed in Chapter 8. Mains electricity was
used for the test and the data for the Moineau pump only is shown in Figure 9-7.

dp dn *
The linear trendline in Figure 9-7 shows that = 0.50 , and thus, = 2.0 .
dn * dp

dp
Alternatively, a quadratic trendline may be used. In which case, is a function of n*,
dn *
but it is sufficient to use n* from the previous control-program cycle. This was
demonstrated during commissioning of the PV-RO rig, but gave no discernible
performance improvement, and was dropped. It would be worth reconsidering, if the

161
algorithm were being applied to a centrifugal pump, for example, where the power-speed
characteristic was more curved.

Returning now to Figure 9-6, the components are complete and, in theory, G may be set
to unity to ensure that the operating point is brought back to the MPP, whenever the
voltage starts to collapse. In practice, a safety margin is required and setting G to 2 gave
good performance.

9.6 Two inverters

In the interest of clarity, the preceding sections have described the MPPT algorithm with
respect to a single inverter. Of course, the RO rig has two inverters, and their speed
setpoints are to be coordinated as discussed in Chapter 8. The overall arrangement of the
two inverters and their control is shown in Figure 9-8.

Discrete
integrator
∆ P+
ΣC Non-linear dn n*
gain ∆p dp ∆n
d
dt pC ∆p– Recovery-ratio
G control

v PV Inverter Motor
Array C C
–

Inverter Motor
C

Figure 9-8 – Complete two-inverter MPPT control system structure

162
9.7 In practice

1400

1200

1000

800
Power (W)

600

400

200 Start

0
290 300 310 320 330 340 350 360 370
PV Array Voltage (V)

Figure 9-9 – Trajectory of MPPT operation, power vs. voltage

To illustrate the operation of the MPPT algorithm, a 3-minute period of 1-second data is
shown in Figure 9-9. At the start of the period the sun was behind a cloud, and less than
200 watts was available from the PV array, just enough to run the Moineau pump. Then
the sun came out and the voltage rose from 354 V to 368 V in just over 10 seconds. The
tracker recognised this immediately, and started to ramp up the speed. The trajectory
followed the classic P-V curve until the MPP was reached and then tracker stayed close
to the MPP for the rest of the 3-minute period. The same data is plotted against time, in
Figure 9-10.

163
 1400

1200

1000

800
Power (W)

600

400

200

0
11:04 11:05 11:06
Time (hh:mm)

Figure 9-10 –MPPT operation, power vs. time

Close inspection of Figure 9-10 reveals a slightly soft (curved) start to the ramp. This is
partly due to the 10 seconds taken for the sun to come out, but also due to a slight
modification to the algorithm, whereby ∆P+ is deliberately reduced at low powers to
improve stability.

A slight pause is apparent, just below 400 W. This is where the plunger pump is started.

At the top of the ramp, there appears to be a slight overshoot, but in fact, this effect is due
to the heating of the PV array.

The 3-minute section of data presented above was selected because it so cleanly
demonstrates the operation of the controller. It shows the controller’s response to an
almost step change in irradiance, followed by a steady condition. The sun is rarely so
obliging, and a more typical period of operation is shown in Figure 9-11.

164
 1000 2000
Irradiance (W/m2)

Irradiance

500 1000
Power

Power (W)
0 0
11:38 11:40 11:42 11:44 11:46 11:48 11:50
Time (hh:mm)

Figure 9-11 – MPPT operation over a 15-minute period

The performance of the tracker over a 15-minute period is shown in Figure 9-11. The
thin trace is the irradiance in the plane of the PV array, measured with a pyranometer, as
described in section 4.17. The thick trace is the DC power extracted from the array,
measured at the input to the inverters. Some clouds passed over towards the end of the
period shown and the tracker responded effectively to track the changing power
available.

9.8 Discussion
No quantitative assessment of the tracker was attempted, but its operation under varying
conditions was carefully observed over many hours and under different weather
conditions. In particular, an X-Y plot in LabVIEW allowed on-line observation of the
trajectory of the operating point in the power-voltage plane, rather like Figure 9-9. This
indicated, for example, whenever the controller was being too cautious and operating to
the right of the MPP. The following discussion is based on those observations.

For the most part, the tracker did appear to sit very close to the MPP. Two observations
support this.

165
First, looking again at Figure 9-9, it is clear from the shape of the curve that the tracker is
sitting at the top of the hill, where dP/dV = 0. The on-line LabVIEW display regularly
showed similar behaviour. Furthermore, disabling the tracker and ramping the speed
setpoint down over a period of a few seconds caused the operating point to cleanly trace
out the right-hand side of the classic P-V curve, again showing that it had been tracking
at the very top.

Second, when close to the MPP, the operating point could be seen looping around it. The
looping is caused by the capacitor, and the fact that the power displayed includes the
external capacitor power, which allows the displayed power to deviate transiently from
the static P-V curve. The looping indicates that the voltage is starting to collapse, and that
the controller is acting as intended.

Once the controller parameters G and ∆P+ were suitably adjusted, the tracker was
effective over a full range of steady-sun conditions: irradiance = 150 to 1100 W/m2, but
the rise time remained a concern.

Figure 9-10 showed that the tracker took approximately 1 minute to respond to a near
step change in irradiance. During this time, the tracker was not at the MPP and power
was being wasted. This is a natural characteristic of hill-climbing algorithms. The total
energy loss that it causes over a long period of operation depends largely on the type of
clouds. During the testing of the PV-RO system at Loughborough, there were periods of
patchy clouds, during which some energy loss was apparent. Such loss may well be
significant in many implementations of hill-climbing MPPT, but go unnoticed. In the
case of the RO system, the sound of the motors running is a give away.

The rise time can be reduced by increasing the setting of ∆P+. This pushes the operating
point up the P-V curve more rapidly, but also tends to push it too far, causing the voltage
to collapse. If the voltage does collapse, the controller resets and then has to climb the
curve again, incurring further energy loss. Use of a variable ∆P+ within the controller
improved the performance of the demonstration system, as mentioned in discussion of
Figure 9-10. Developing this approach further could substantially reduce the overall rise
time.

166
9.9 Pros and cons
The MPPT algorithm described in this chapter works directly with standard industrial
variable-speed drive inverters, which have much lower cost than specialised PV
inverters.

LabVIEW was used for the demonstration, but the algorithm is simple, and could readily
be implemented in a very low-cost micro-controller.

The algorithm requires feedback of only the PV voltage, or the inverter DC-bus voltage,
which is virtually the same. Many industrial inverters can be configured to provide this as
a ready-conditioned analogue signal. No feedback of the PV current is required. The
algorithm tracks array temperature changes, without requiring a temperature sensor or
periodically open-circuiting the array.

The performance of the demonstration MPPT was critically dependent on the careful
adjustment of several control parameters and required an estimate of the load
characteristic dp/dn*. With experience, these adjustments could be rationalised, possibly
to a point that no manual commissioning is needed. No particular knowledge of the PV
characteristics is required, and the algorithm should perform equally well with
amorphous-silicon PV, provided the lower MPP voltage is within the range acceptable to
the industrial inverter.

The algorithm appeared to perform well under steady sun, but needs further refinement to
track unsteady conditions more rapidly.

The algorithm could be immediately applied to solar pumping applications, and it would
be interesting to compare its performance with more traditional techniques.

167
 Chapter 10 Batteryless PV-RO demonstration system
measured performance

The PV-RO test rig outlined in Chapter 7, coupled with the recovery-ratio control of
Chapter 8 and the MPPT algorithm just described in Chapter 9, was tested for two
complete days, and some of the measured results are presented and discussed here.

The feed water was straight NaCl solution nominally at 32,800 mg/L, which is isosmotic
with ASTM standard seawater, as discussed in section 5.10.3. The concentration varied
by plus and minus 500 mg/L (±3% of the feed water concentration) during the tests
because of variation of concentration of water held within the RO modules, which is
affected by the recovery ratio.

The temperature was held at 25 ºC, plus and minus 0.1 ºC for most of the duration.

10.1 Irradiance and product flow

1000

Irradiance
Irradiance (W/m2)

Product Flow

500

400
Product Flow (L/h)

200

0 0
06:00 10:00 14:00 18:00 22:00 02:00 06:00 10:00 14:00 18:00
Time (hh:mm)

Figure 10-1 – Measured irradiance and product flow June 9th and 10th 2003

168
The thin trace in Figure 10-1 is the irradiance in the plane of the PV array, measured with
a pyranometer, as described in section 4.17. The thick trace is the product flow, which
totalled 2.93 m3 over the complete two-day test.

The system is intended to produce more than twice this amount, but bear in mind that this
testing was performed in the UK, with many passing clouds, and that the PV array is just
1530 Wp, which is only 64 % of the design value.

10.2 Product flow versus irradiance

350

300

250
Product Flow (L/h)

200

150

100

50

0
0 100 200 300 400 500 600 700 800 900 1000
2
Irradiance (W/m )

Figure 10-2 – Product flow versus irradiance

Figure 10-2 uses the same data as Figure 10-1 and shows that water production is
roughly proportional to irradiance, and, importantly, that it starts at very low irradiance.
This graph uses data averaged at 10-minute intervals, and shows considerable scatter.
Looking ahead to Figure 10-3 shows that the scatter in Figure 10-2 occurs primarily in
the conversion from irradiance to DC power. There are two likely causes. First, the
temperature of the PV array, which varies during the day and with passing clouds, will
have an effect on the DC power available from the array for any given irradiance. In
particular, the measured array temperature varied by up to 26 ºC, at 200 W/m2, which

169
was the worst case. This will affect the power available by approximately 13%, taking
the coefficient of power at 0.5% per Cº (BP-Solar 2002). The scatter in Figure 10-2 is
clearly much greater than 13% and is caused mainly by the maximum power point
tracker, and in particular its slow rise time, as discussed in section 9.8. Assuming that
this can be resolved by further development of the tracker, the points scattered below the
diagonal of Figure 10-2 will be pushed upwards and the total water production will be
correspondingly increased.

10.3 Product flow versus DC power

400

350

300
Product Flow (L/h)

250

200

150

100

50

0
0 200 400 600 800 1000 1200 1400 1600
DC Power (W)

Figure 10-3 – Product flow versus DC power

In order to see the performance of the RO rig itself, Figure 10-3 shows the product flow
against DC power coming from the PV array, measured at the input to the inverters. This
graph and all subsequent graphs in this chapter use data averaged at 1-minute intervals,
covering the whole of the two-day test.

Figure 10-3 shows that water production starts at just 100 W and then increases more or
less proportionally to the available power.

170
10.4 Specific energy consumption (kWh/m3)

6 Measured
Predicted

5
Specific Energy (kWh/m3)

0
0 200 400 600 800 1000 1200 1400 1600
DC Power (W)

Figure 10-4 – Specific energy (kWh/m3)

Plotting the same data, but this time as specific energy consumption, gives the points in
Figure 10-4. The line is the predicted characteristic from the model described in section
6.5, and, in particular, from the data in Figure 6-19, but with the spread caused by
temperature removed.

For perspective, bear in mind that a standard mains-powered RO unit would probably not
even appear on this graph, and if it did, it would only offer a single fixed operating point.

At low input power (towards the left of Figure 10-4), it is apparent that the test rig is
actually more efficient than was predicted. This is probably due to the membrane element
models, which were extrapolated from measurements taken from slightly different
elements. In particular, the measurements were taken from high-rejection elements;
whereas the test rig is fitted with high-flow elements, see section 2.7.6.

At higher input power (moving to the right in Figure 10-4), it is apparent that the test rig
becomes less efficient than was predicted. This is largely due to the plunger pump and its

171
motor, which do not match those in the design model. Recall that both the pump and
motor were taken from the old test rig and that they are coupled by a toothed belt, as
discussed in Chapter 7. As shown later in the chapter, the old motor has significantly
lower efficiency than the new motor used on the Moineau pump, especially at low
speeds.

10.5 Water recovery ratio

40

35

30
Recovery Ratio (percent)

25

20

15

10

0
0 200 400 600 800 1000 1200 1400 1600
DC Power (W)

Figure 10-5 – Water recovery ratio

The variable recovery ratio discussed in Chapter 8 is clearly apparent in Figure 10-5. The
section at the bottom left is where the Moineau pump only is running and the recovery
ratio is around the fixed value of 10 % as determined by the Clark pump. The section
across the top is where both motorised pumps are running. The scatter in the figure
occurs because the recovery ratio varies rapidly in response to the varying irradiance,
while the data used for this plot is averaged at 1-minute intervals.

Although the two modes of operation are very apparent in Figure 10-5, looking back at
Figure 10-3 shows that the transition is in fact very smooth. In practice, the plunger
pump was started and stopped and re-started very frequently with apparently no ill effect.

172
10.6 Product concentration

6000

5000

4000

3000 Product
Concentration
(mg/L)

2000

1000

Product Flow (L/h)

0
06:00 18:00 06:00 18:00
Time (hh:mm)

Figure 10-6 – Product concentration

The product concentration, measured via its conductivity as described in section 4.11 and
averaged at 1-minute intervals, is shown in Figure 10-6. Recall, from section 2.1.3, that:

The palatability of water with a TDS level of less than 600 mg/litre is generally
considered to be good; drinking-water becomes significantly unpalatable at TDS
levels greater than 1200 mg/litre. (WHO 2003)

Clearly, the concentration of product water from the PV-RO rig is, at times, greatly in
excess of this. Fortunately, the high concentration generally coincides with low flow, and
what matters is the average concentration of water in the product tank.

Of course, the test rig has no product tank, but the data has been aggregated to illustrate
the volume and concentration that there would be.

173
 3000
2930

2500

2000 Product Volume (L)

1500

1120
1000 Product Concentration (mg/L)

500

0
06:00 18:00 06:00 18:00
Time (hh:mm)

Figure 10-7 – Product tank volume and concentration, basic case

If all of the product water were directed to a tank, it would accumulate as shown in
Figure 10-7, which is obtained simply by integrating the measured data shown in
Figure 10-6. This shows the total production of 2.93 m3 over the two days, as mentioned
previously. It also shows that the concentration is getting close to the significantly
unpalatable level.

A significant part of the total salt in the product tank is due to a small volume of very
high concentration water coming from the membrane elements at startup (usually at
sunrise). The salt is that which has diffused though the membranes while the pumps were
stopped (overnight).

174
 3000
2904

2500

2000 Product Volume (L)

1500

1000
Product Concentration (mg/L)
790

500

CP > 6000mg/L : Diverted Volume (L)

26
0
06:00 10:00 14:00 18:00 22:00 02:00 06:00 10:00 14:00 18:00
Time (hh:mm)

Figure 10-8 – Product tank volume and concentration, diverting > 6000 mg/L

A simple diverter valve, fitted in the product line prior to the tank, could be used to
significantly reduce the concentration of the water in the tank. The valve would need to
operate automatically and reject product water exceeding a chosen concentration
threshold. The effect of such a valve, with a threshold set at 6000 mg/L, is illustrated in
Figure 10-8. This discards less than 1 % of the product volume but brings the final
concentration in the tank to below 800 mg/L.

The concentration of the product water actually delivered by the test rig is markedly
higher than that predicted by the model. This is partly because the volume of production
of good quality water is reduced by the small PV array and the passing clouds, but also
because the model was based on measurements taken from membrane elements with
high-rejection, whereas the test rig is fitted with high-flow elements, see section 2.7.6.
High-rejection elements could certainly be used in the PV-RO system; they have similar
cost and would improve the quality of the product, but at the expense of quantity.

175
 10.7 Pressures

70

High Pressure
60 Product Flow

50
High Pressure (bar)

40
350
30 300

Product Flow (L/h)

250
20 200
150
10 100
50
0 0
06:00 10:00 14:00 18:00 22:00 02:00 06:00 10:00 14:00 18:00
Time (hh:mm)

Figure 10-9 – Membrane feed pressure

The pressure of the feed water at the inlet of the first RO membrane element is shown in
Figure 10-9. Production starts at around 40 bar and the pressure varies all the way up to
the maximum 69 bar allowable by the membrane pressure vessels. Pressure is maintained
during short interruptions (passing clouds), but decays almost to zero during the night.

The medium pressure, between the Moineau pump and the Clark pump, behaves
similarly, but in the range 5 to 9.6 bar.

176
10.8 Pump efficiencies

100

90

80

~9.6 bar
Moineau Efficiency (percent)

70

60

50

40

~4.4 bar
30

20

10

0
0 100 200 300 400 500 600 700 800
Moineau Speed (rpm)

Figure 10-10 – Moineau pump efficiency

The Moineau pump meets its expectations. The efficiency shown in Figure 10-10 is
certainly far higher than that attainable by a centrifugal pump of the same size. Vane
pumps offer an alternative, but must be seawater compatible. Spectra Watermakers have
had success with vane pumps. The scatter in Figure 10-10 occurs because the operating
point of the pump varies rapidly in response to the varying irradiance, while the data used
for this plot is averaged at 1-minute intervals.

The plunger pump efficiency measured between 75 and 85 % regardless of flow and
pressure.

The energy losses in the Clark pump, while operating within the complete PV-RO
system, were less than the data acquisition system could reliably measure. The Clark
pump was tested in isolation, as described in section 5.2, and its efficiency has proved
excellent throughout.

177
10.9 Inverter and motor efficiencies
The inverter and motor efficiencies are combined here because measurement of the high-
frequency switched power between the two cannot readily be achieved with sufficient
accuracy.

100

90

80
Combined Efficiency (percent)

70

60

50

40

30

20

10

0
0 100 200 300 400 500 600 700 800
Inverter and Siemens Motor Speed (rpm)

Figure 10-11 – Combined efficiency of inverter and motor driving the Moineau
pump

Figure 10-11 shows the newer of the two motors, and its efficiency is very good
considering that it is operating typically between 100 and 400 W shaft output power.

178
 100

90

80
Combined Efficiency (percent)

70

60

50

40

30

20

10

0
0 100 200 300 400 500 600 700 800
Inverter and Brook Motor Speed (rpm)

Figure 10-12 –Combined efficiency of inverter and motor driving the plunger pump

Figure 10-12 shows the combined efficiency for the older motor and its inverter, which is
about 10 % lower than the new motor. Replacement of this motor and the associated
plunger pump would bring significant improvement to the overall system performance.

179
 Chapter 11 Practical considerations

11.1 Clark pump

The excellent energy efficiency of the PV-RO system reported in the previous chapter is
due primarily to the Clark pump. The Clark pump is a high-tolerance precision machine
(Spectra 2003), and will typically require more maintenance than a simple needle valve.
Indeed, the Clark pump at CREST required some maintenance during the course of this
project. Spectra, the manufacturers, provided excellent support and have altered the
design over the years to improve the reliability of new Clark pumps.

11.2 Membrane fouling

Membrane fouling can cause a large reduction of product flow in an RO system, and a
consequent increase in specific energy consumption. But, the effect can be compared to a
reduction in temperature, and therefore, as discussed in section 6.2.4, the increase in
specific energy consumption is very much less in systems with efficient brine-steam
energy recovery.

Nonetheless, the main operating cost of the batteryless PV-RO system is expected to be
the replacement of the RO membrane elements when they become excessively fouled
and can no longer be effectively cleaned. And, as noted in section 2.9.3, there is concern
that intermittent operation of an RO system will accelerate membrane fouling. Indeed,
this is probable, but has not been quantified, and the relative costs of replacing batteries
or membrane elements are not yet known.

Membrane life expectancy in a well-maintained 24-hour-a-day RO plant is around five

years. The cost estimates presented in section 6.6 for the batteryless PV-RO system
assumed that a complete new set of RO membrane elements would be required every
12 months. The membrane replacement rate in practice will be critically dependent upon
feed-water quality, and this depends to a large extent on the feed water intake
arrangement.

180
11.3 Feed water intake arrangement – beach wells
The Moineau pump in the PV-RO system can be above ground, as it is in the test rig, and
suck water from a shallow well, or may be submersible and some distance from the main
RO rig. The above-ground approach offers better energy efficiency because it allows use
of a high-efficiency motor; submersible seawater-compatible motors are available but
only at lower efficiency. On the other hand, sucking water from a well is limited to about
7 metres height, and provision must be made to ensure that the Moineau pump does not
run dry.

Many large RO plants have open-sea intakes, which often provide feed water of poor
quality that requires elaborate pre-treatment before it may be fed to the actual RO
membranes. A far preferable arrangement is the use of a beach well, since this can
provide pre-filtered water and greatly reduce pre-treatment requirements (Koch 2000;
Schwarz 2000). Where a traditional beach well is not practical, alternative belowground
intakes can be used to similar effect (Cansdale 2001; Kunczynski 2003).

A beach well intake or similar, providing very good quality feed water, may be
considered an essential prerequisite for the PV-RO system described in this thesis, in
order that it may operate intermittently and without chemical additives. The use of
electrostatic pre-treatment in the form of a Zeta Rod (Romo et al. 1999) may also be
worth considering.

11.4 Membrane flushing

Most RO systems include provision for flushing of the membranes at shut down. Often,
this involves feeding product water, made earlier, through the membranes, which
removes most of the salt and prevents precipitation. As an alternative, Koch Membrane
Systems (2000) suggest use of seawater for the shutdown flush, in which case, the flush
is simply a matter of reducing the recovery ratio to zero. The PV-RO system has no
provision to automatically flush the membranes, but it does reduce the recovery ratio to
just 10 % as the sun goes down. And, this 10 % is spread over a very generous membrane
area, meaning that concentration polarisation is minimal. It is hoped, therefore, that the
omission of a shutdown flush mechanism will not unduly affect membrane life. In its
favour, the PV-RO system can be expected to run nearly every day, assuming it is located

181
in an area with a good solar resource. Thus, it should rarely stand still for much over 16
hours.

11.5 Product tank chlorination

At the time it is produced, the product water from an RO system should be almost free of
biological activity, but it is very prone to re-infection, particularly in the storage tank.
Modest chlorination is suggested to maintain safety in the tank and to provide a residual
presence in the water, which will continue to be effective throughout any pipes, storage
bottles or whatever.

11.6 Osmotic suck-back

When shutdown, most RO systems will suck water back from the product line because
(forward) osmosis occurs. This can be used as a way to provide a shutdown flush,
provided chlorine is not introduced, since chlorine will damage the membrane. But, suck-
back does not occur in the PV-RO system, because the pumps are positive-displacement
and do not release the pressure.

182
 Chapter 12 Conclusions

This thesis has presented the design and demonstration of a photovoltaic-powered

reverse-osmosis (PV-RO) seawater desalination system without batteries. The method of
design, in particular the structure of the Matlab-Simulink model, was developed by the
author and may be of value in the development of similar variable-flow/pressure RO
systems. The use of variable recovery ratio to achieve energy efficiency over a wide
range of input power has been demonstrated and serves to support the case that PV-RO
systems can operate efficiently without batteries. Lastly, a novel maximum power point
tracking algorithm has been developed, which, with further refinement, could be of value
in PV-RO and other PV-pumping systems.

12.1 Renewable-energy powered desalination

The rapid worldwide uptake of desalination over the last thirty years, the expected
continuation of this trend and the associated consumption of fossil fuel lead to
considerable interest in renewable-energy powered desalination. The coupling of
renewable energy and desalination technologies was reviewed in section 2.4, with
reference to more comprehensive studies, and showed the significance of small-scale
batteryless PV-RO, both in its own right and as a precursor to larger variable-flow RO
systems for use in connection with PV and other intermittent renewable-energy sources.

12.2 Batteryless PV-RO

The case against using batteries in a PV-RO system was discussed in section 2.9.3 but
not proven, since it relies on un-quantified reports of very poor battery performance in
many other PV systems. The counter argument, that RO systems are best run at constant
flow, was also presented, but again is un-quantified because of lack of documented
experience of RO systems with intermittent or variable flow.

This thesis presents the design and testing of the batteryless approach, and provides
estimates of performance and capital costs of such systems. Meanwhile, other researchers
are developing PV-RO systems that do include batteries, and it would now be possible to
compare cost estimates for the two approaches. Such a comparison would, however, be

183
critically reliant on assumptions regarding membrane lifetime under variable-flow
conditions.

12.3 Variable-flow
The testing of RO membranes under variable-flow conditions, presented in Chapter 3 and
Chapter 10, was successful and revealed no immediate problems. It was, however, very
brief and used straight NaCl in place of seawater. The testing presented in Chapter 10
demonstrated energy efficiency over a broad range of operation, but did not provide any
indication of membrane lifetime under real operating conditions. It is recommended,
therefore, that long-term membrane testing under intermittent and variable flow
conditions should now be conducted. This could be within a field trial of batteryless PV-
RO or carried out under laboratory conditions at an established RO plant. The latter
approach would perhaps yield better-calibrated data and a better comparison against
traditional continuous operation. Such testing is also highly relevant to the coupling of
RO with other intermittent renewable-energy sources, in particular windpower, and
should take account of variable recovery ratio as discussed below.

12.4 Product concentration

Operation of RO membranes under variable-flow conditions generally implies a
reduction in average flow, which in turn causes an increase in the concentration of the
product water. The results of the initial testing (section 3.1) and the software modelling
(section 6.5.2), both of which were based on high-rejection membrane elements,
suggested that this increase would be perfectly acceptable. However, the results
measured from the new test rig, which was fitted with high-flow membranes, did show
rather high concentrations at times, particularly at dawn and dusk (Figure 10-6). This
indicates the need for some caution in designing RO systems to operate with variable-
flow but the problem is not insurmountable. Use of high-rejection membranes or a simple
diverter valve to reduce product concentration is discussed in section 10.6.

Furthermore, the acceptable concentration of product water is primarily a matter of taste

and depends greatly on who the consumers are. The 500-mg/L TDS limit, often stated in
desalination literature, may not always be appropriate (section 2.1.3).

184
12.5 Energy recovery – The Clark pump
Brine-stream energy recovery is critical for the efficiency of seawater RO, but is often
neglected in small systems because suitable mechanisms are only now becoming
available. Testing of a hydraulic motor, presented in Chapter 3, showed that it gave a
water-to-water efficiency of typically below 60 % and raised doubts regarding its
seawater compatibility. The Clark pump was identified as an alternative and a theory
defining its efficiencies was developed in section 5.2.1. A Clark pump was tested
(section 5.2.2 and Appendix C) and showed water-to-water efficiency typically above
90 %. Furthermore, this efficiency was maintained over a very wide range of operation,
making it well suited for a batteryless PV-RO system. The Clark pump is central to the
design presented in this thesis.

12.6 Variable recovery ratio

In order to make best use of the Clark pump, a system design employing two motor-
driven pumps was adopted (section 6.2.3), and, to provide efficient operation over a wide
range of input power, use of a variable recovery ratio was proposed and modelled
(Figure 6-6). The new test rig was commissioned with a variable recovery ratio as
described in Chapter 8, and efficient operation over a wide range of input power was
demonstrated (Figure 8-3).

The use of a variable recovery ratio is recommended for RO systems operating from
variable energy sources. Use of a variable recovery ratio also has implications for the
membrane lifetime, since operation at a very low recovery ratio is akin to a seawater
flush (section 11.4). It is recommended that the membrane testing, suggested above,
should also consider variable recovery ratio.

12.7 Maximum power point tracking and industrial inverters

A novel maximum power point tracking (MPPT) algorithm specifically for use with low-
cost industrial drive inverters was proposed and demonstrated as discussed in Chapter 9.
The algorithm appeared to perform well under steady sun, but needs further refinement to
track unsteady conditions more rapidly. The algorithm could readily be applied to PV
water pumping applications, and it would be interesting to compare its performance with
existing MPPT techniques used with industrial inverters.

185
12.8 Batteryless PV-RO hardware demonstration
The design features outlined above were brought together in hardware, and seawater RO
operating from a PV array without batteries was successfully demonstrated (Chapter 10).

The rate of water production varied according to the irradiance, as shown in Figure 10-1,
and the variable recovery ratio can be seen in Figure 10-5. The overall water production
was in line with expectations, given that the testing was performed in the UK, with many
passing clouds, and that the PV array used was only 64 % of the design size. These
factors are removed when one considers the specific energy consumption shown in
Figure 10-4. It is a little higher than predicted because one pump and one motor were
taken from the earlier test rig and do not quite match the design. Despite this, efficient
operation is achieved over a wide range of input power, enabling the system to make
good use of the available power from the PV.

12.9 System complexity and reliability

The energy-efficient operation of the system just described is achieved, in part, through
the addition of components that undeniably increase its complexity and may adversely
affect reliability. The Clark pump required some maintenance during the course of this
project, and the implementation of variable recovery ratio introduces a second inverter,
motor and pump to the system. Clearly, reliability is a critical factor for any desalination
system intended to provide drinking water in remote locations, and it is recommended
that reliability be treated as a priority alongside energy-efficiency in the future
development of PV-RO.

12.10 Instrumentation
The instrumentation of the test rig centred on National Instruments’ data-acquisition
hardware and LabView software, both of which are excellent and are recommended for
future work of this type. Turbine flow meters were found to be inadequate and were
replaced by oval-gear flow meters, which are much more satisfactory. Concentration
measurements were initially error prone but were improved as discussed in section 4.11.
Torque measurements remain prone to electrical noise from the inverter-driven motors,
and it is recommended that the load cells be mounted further from the motors in future.

186
12.11 Software modelling
The use of Matlab-Simulink for the modelling of the complete PV-RO system proved
successful. The model includes everything from the solar irradiance striking the
photovoltaic panels through to the water in the product tank. It models the variations of
flows and pressures throughout the system with respect to variations in irradiance and
feed-water temperature, and it can, for example, perform an hour-by-hour simulation of a
whole year of operation. A critical feature of the model is its structure, described in
sections 5.1 and 6.1, which allows the components, such as motors, pumps and RO
modules, to be added, deleted and reconnected simply by manipulating the blocks and
interconnections of the graphical interface.

The model was used to evaluate alternative system configurations and to minimise the
Capital cost per daily water production, as shown in section 6.3. The components of the
model were developed from in-house testing supplemented by manufacturer’s data, and
the performance predications it gave showed good agreement with measurements taken
later from the test rig. The use of Simulink and the model structure described is
recommended for future modelling of variable-flow RO systems.

12.12 Market identification

The system described in this thesis is one of many seawater PV-RO demonstrations
worldwide, and it appears that the technology could soon be commercialised, as has
already happened with brackish-water PV-RO. However, interest in PV-RO today seems
dominated by technology push, with rather less evidence of market pull. A precise
identification of the market for such systems is now required.

187
 Published Papers

Desalination
Thomson, Murray, Marcos S. Miranda and David Infield (2003). A small-scale seawater
reverse-osmosis system with excellent energy efficiency over a wide operating
range. Desalination 153(1-3): 229-236. Also available at:
[Link] (accessed: March 04)

Thomson, Murray and David Infield (2003). A photovoltaic-powered seawater reverse-

osmosis system without batteries. Desalination 153(1-3): 1-8. Also available at:
[Link] (accessed: March 04)

Thomson, Murray and David Infield (2003). A Reverse-Osmosis System for the
Desalination OF Seawater Powered by Photovoltaics Without Batteries.
Proceedings of Renewable Energy Sources for Islands, Tourism and Water
Desalination Conference. Crete, Greece, May 2003, EREC (European Renewable
Energy Council) page 551

Thomson, Murray, Jo Gwillim, Andrew Rowbottom, Ian Draisey and Marcos Miranda
(2001). Batteryless Photovoltaic Reverse-Osmosis Desalination System,
S/P2/00305/REP, ETSU, DTI, UK. Also available at:
[Link]
(accessed: March 04)

Others .

Thomson, Murray (2000). Automatic voltage control relays and embedded generation –
Part 1. IEE Power Engineering Journal 14(2): 71-76

Thomson, Murray (2000). Automatic voltage control relays and embedded generation –
Part 2. IEE Power Engineering Journal 14(3): 93-99

188
 References

Abufayed, A.A. (2003). Performance characteristics of a cyclically operated seawater

desalination plant in Tajoura, Libya. Desalination 156: 59-65
Al Suleimani, Zaher and V. Rajendran Nair (2000). Desalination by solar-powered
reverse osmosis in a remote area of the Sultanate of Oman. Applied Energy 65(1-
4): 367-380
Alawaji, Saleh, Mohammed Salah Smiai, Shah Rafique and Byron Stafford (1995). PV-
Powered Water Pumping and Desalination Plant for Remote Areas in Saudi
Arabia. Applied Energy 52(2-3): 283-289
Andrews, W. T., Wil F. Pergande and Gregory S. McTaggart (2001). Energy
performance enhancements of a 950 m3/d seawater reverse osmosis unit in Grand
Cayman. Desalination 135(1-3): 195-204
Andrews, William T. and David S. Laker (2001). A twelve-year history of large scale
application of work-exchanger energy recovery technology. Desalination 138(1-
3): 201-206
Arrindell, E. M. and J. D. Birkett (2002). Aruba's Early Experiences in Desalination.
International Desalination and Water Reuse Quarterly 12(3): 30-37
Assimacopoulos, D., R. Morris and A. Zervos (2001). Water, Water everywhere..:
Desalination Powered by Renewable Energy Sources. REFOCUS, Elsevier. Also
available at: [Link] (accessed: March 04)
ASTM (1985). D4516-85 Standard Practice for Standardizing Reverse Osmosis
Performance Data
ASTM (1998). D1141-98e1 Standard Practice for the Preparation of Substitute Ocean
Water.
Avlonitis, S. A., I. Poulios, N. Vlachakis, S. Tsitmidelis, K. Kouroumbas, D. Avlonitis
and M. Pavlou (2003). Water resources management for the prefecture of
Dodekanisa of Greece. Desalination 152(1-3): 41-50
BP-Solar (2002). BP 585 photovoltaic module datasheet, 02-3009-2B 6/02. Also
available at: [Link] (accessed: May 03)
Bray, D. T. (1966). Engineering of Reverse-Osmosis Plants. Desalination by Reverse
Osmosis. U. Merten. Cambridge, Massachusetts, M.I.T. Press
Buros, O.K. (2000). The ABCs of Desalting. Topsfield, Massachusetts, USA,
International Desalination Association. Also available at:
[Link] (accessed: March 04)
Byrne, Wes (1995). Reverse Osmosis - A practical guide for industrial users. Littleton,
CO, USA, Tall Oaks
Cabassud, Corinne and David Wirth (2003). Membrane distillation for water
desalination: how to choose an appropriate membrane? Desalination 157: 307-
313

189
Cansdale, Richard (2001). The installation of a new sub-sand sea water intake for La
Source, Grenada. International Desalination and Water Reuse Quarterly 11(3):
40-43
Chamberlin, Charles E., Peter Lehman, James Zoellick and Gian Pauletto (1995). Effects
of mismatch losses in photovoltaic arrays. Solar Energy 54(3): 165-171
Childs, Willard D. and Ali E. Dabiri (1998). Vari-Ro Desalting Pilot Plant Testing And
Evaluation Final Technical Report, Water Treatment Technology Program
Report No. 30, Assistance Agreement No. 14255-FC-Sl-20410, U. S. Department
of the Interior, May 1998, Science Applications International Corporation, San
Diego, CA
Childs, Willard D. and Ali E. Dabiri (1999). Hydraulic Driven RO Pump & Energy
Recovery System. International Desalination and Water Reuse Quarterly 9(2):
21-29
Childs, Willard D. and Ali E. Dabiri (2000). VARI-RO Solar-powered Desalting Study,
MEDRC Project: 97-AS-005a, March 2000, MEDRC: The Middle East
Desalination Research Center, Muscat, Sultanate of Oman. Also available at:
[Link] (accessed: March 04)
Childs, Willard D., Ali E. Dabiri, Hilal A. Al-Hinai and Hussein A. Abdullah (1999).
VARI-RO solar-powered desalting technology. Desalination 125(1-3): 155-166
CRES (1998). Desalination Guide Using Renewable Energies, THERMIE - DG XVII,
European Commission Report, ISBN 960-90557-5-3., CRES, Greece
Danfoss (2002). Danfoss Water Pumps type APP 0.6/1.0 , APP 1.8/2.2 and APP
5.1/6.5/7.2/8.2, Danfoss, Nordborg, Denmark
Delyannis, E. and V. Belessiotis (2001). Research and Technological Applications in
Solar Desalination. International Journal of Island Affairs - The Water-Energy
Binomial 10(1): 7-11
Doman, D.G., Bowie G. Keefer and J.L. Richardson (1982). Performance of energy
recovery triplex pumps developed for reverse osmosis desalination. 10th Annual
Conference of the Water Supply Improvement Association. Honolulu, July 1982
Doujak, Eduard and Bernhard List (2003). Application of PIV for the Design of Pelton
Runners for RO-Systems. Proceedings of Renewable Energy Sources for Islands,
Tourism and Water Desalination Conference. Crete, Greece, May 2003, EREC
(European Renewable Energy Council) pp 533-540
DOW (1995). FILMTEC Membrane Elements - Technical Manual, April 1995, Dow
Liquid Separations. Also available at:
[Link] (accessed: March 04)
DOW (2000). ROSA, Version 4.2. Dow Liquid Separations. Also available at:
[Link] (accessed: March 04)
Druck (1997). PDCR 800 SERIES Datasheet, 12/1997, Druck Limited, Fir Tree Lane,
Groby, Leicester LE6 0FH England. Also available at: [Link]
(accessed: March 04)
Dunlop, James P. and Brian N. Farhi (2001). Recommendations For Maximizing Battery
Life In Photovoltaic Systems: A Review Of Lessons Learned. Proceedings of
Forum 2001, Solar Energy: The Power to Choose, Washington, DC

190
El-Dessouky, Hisham T. and Hisham M. Ettouney (2002). Fundamentals of Salt Water
Desalination. Amsterdam, Elsevier
El-Nashar, Ali M. (2001). The economic feasibility of small solar MED seawater
desalination plants for remote arid areas. Desalination 134(1-3): 173-186
EPA (2004). National Secondary Drinking Water Regulations, US Environmental
Protection Agency. Also available at: [Link]
(accessed: March 04)
Espino, Tomas, Baltasar Penate, Gonzalo Piernavieja, Dirk Herold and Apostel Neskakis
(2003). Optimised desalination of seawater by a PV powered reverse osmosis
plant for a decentralised coastal water supply. Desalination 156: 349-350
Fabre, Aurélie (2003). Wind turbines designed for the specific environment of islands
case study: Experience of an autonomous wind powered water desalination
system on the island of Therasia, Greece - Vergnet SA, France. Proceedings of
Renewable Energy Sources for Islands, Tourism and Water Desalination
Conference. Crete, Greece, May 2003, EREC (European Renewable Energy
Council) pp 247-252
Garcia-Rodriguez, Lourdes (2002). Seawater desalination driven by renewable energies:
a review. Desalination 143(2): 103-113
Gocht, W., A. Sommerfeld, R. Rautenbach, Th. Melin, L. Eilers, A. Neskakis, D. Herold,
V. Horstmann, M. Kabariti and A. Muhaidat (1998). Decentralized desalination
of brackish water by a directly coupled reverse-osmosis-photovoltaic-system - A
pilot plant study in Jordan. Renewable Energy 14(1-4): 287-292
Gotor, Antonio Gomez, gnacio De la Nuez and Celso Argudo Espinoza (2003).
Optimization of RO desalination systems powered by renewable energies.
Desalination 156: 351-351
Gottschalg, Ralph (2001). Module/Cell Performance Evaluation. Photovoltaic
Technology for Bangladesh. A. K. M. S. Islam and D. G. Infield, Bangladesh
University of Engineering and Technology, Dhaka
Green, Donna (2003). Thailands solar white elephants an analysis of 15yr of solar battery
charging programmes in northern Thailand. Energy Policy
Green, Martin A. (1995). Silicon Solar Cells - Advanced Principles and Practice.
Sydney, Australia, Centre for Photovoltaic Devices and Systems, Univeristy of
New South Wales
Gwillim, Jo (1996). Village Scale Photovoltaic Powered Reverse Osmosis, March 1996,
Dulas Limited, Machynlleth, Wales
Gwillim, Jo (2001). Pesonal Communication. Dulas Limited, Machynlleth, Wales
Hasnain, Syed M. and Saleh A. Alajlan (1998). Coupling of PV-powered RO brackish
water desalination plant with solar stills. Desalination 116(1): 57-64
Herold, D., V. Horstmann, A. Neskakis, J. Plettner-Marliani, G. Piernavieja and R.
Calero (1998). Small scale photovoltaic desalination for rural water supply-
demonstration plant in Gran Canaria. Renewable Energy 14(1-4): 293-298
Herold, Dirk and Apostolos Neskakis (2001). A small PV-driven reverse osmosis
desalination plant on the island of Gran Canaria. Desalination 137(1-3): 285-292

191
IEA (1999). Lead acid battery guide for stand alone photovoltaic systems, IEA-PVPS 3-
06:1999, December 1999, International Energy Agency Photovoltaic Power
Systems Programme
IEA (2002). Testing of batteries used in Stand Alone PV Power Supply Systems, IEA
PVPS T3-11: 2002, October 2002, International Energy Agency Photovoltaic
Power Systems Programme
IPCC (2001). Climate Change 2001: The Scientific Basis. Contribution of Working
Group I to the Third Assessment Report of the Intergovernmental Panel on
Climate Change. J. T. Houghton, Y. Ding, D.J. Griggs, M. Noguer, P.J. van der
Linden, X. Dai, K. Maskell, and C.A. Johnson (eds.), Cambridge University Press
Jensen, Thomas Lynge (2000). Renewable Energy on Small Islands, Forum for Energy
and Development (FED)
Johnson, James S., Lawrence Dresner and Kurt A. Kraus (1966). Hyperfiltration
(Reverse Osmosis). Principles of Desalination. K. S. Spiegler. New York,
Academic Press
Kalogirou, Soteris (1997). Survey of solar desalination systems and system selection.
Energy 22(1): 69-81
Kapp, Gisbert (1888). Society of Telegraph Engineers and Electricians
Keefer, Bowie G. (1980). Reverse osmosis method and apparatus. United States Patent
4,187,173
Keefer, Bowie G. (1984). Multi-cylinder reverse osmosis apparatus and method. United
States Patent 4,434,056
Keefer, Bowie G., R. D. Hembree and F. C. Schrack (1985). Optimized matching of solar
photovoltaic power with reverse osmosis desalination. Desalination 54: 89-103
Koch (2000). Seawater Design Guide - Reverse Osmosis Membrane Systems.
Wilmington, Koch Membrane Systems
Koch-ROPRO (2000). ROPRO Version 6.1. Fluid Systems, Koch Membrane Systems, .
Kolhe, Mohanlal, Sunita Kolhe and J. C. Joshi (2002). Economic viability of stand-alone
solar photovoltaic system in comparison with diesel-powered system for India.
Energy Economics 24(2): 155-165
Koschikowski, Joachim, Matthias Rommel and Marcel Wieghaus (2003). Solar thermal-
driven desalination plants based on membrane distillation. Desalination 156: 295-
304
Kunczynski, Yan (2003). Development and Optimization of 1000-5000 GPD Solar
Power SWRO. IDA World Congress on Desalination and Water Reuse. Bahamas,
Sept 28-Oct 3 2003
Laborde, H. M., K. B. Franca, H. Neff and A. M. N. Lima (2001). Optimization strategy
for a small-scale reverse osmosis water desalination system based on solar
energy. Desalination 133(1): 1-12
Lührs, I. (2003). ENERCON Desalination: Sustainable solutions for drinking water
production. Proceedings of Renewable Energy Sources for Islands, Tourism and
Water Desalination Conference. Crete, Greece, May 2003, EREC (European
Renewable Energy Council) pp 221-222

192
MacHarg, John P. (2001). Exchanger Tests Veryfy 2.0 kWh/m3 SWRO Energy Use.
International Desalination and Water Reuse Quarterly 11(1): 42-46
MacHarg, John P. (2002). The Evolution of SWRO Energy-Recovery Systems.
International Desalination and Water Reuse Quarterly 11(4): 48-53
Manth, Thomas, Michael Gabor and Eli Oklejas Jr. (2003). Minimizing RO energy
consumption under variable conditions of operation. Desalination 157: 9-21
Markvart, Tomas (1999). Solar electricity. Chichester, Wiley
Maslin, Anthony, David Annandale and Marc Saupin (2003). Decentralised water
solutions Island infrastructure project. Proceedings of Renewable Energy Sources
for Islands, Tourism and Water Desalination Conference. Crete, Greece, May
2003, EREC (European Renewable Energy Council) pp 287-294
Mathew, Kuruvilla, Stewart Dallas, Goen Ho and Martin Anda (2001). Innovative Solar-
Powered Village Potable Water Supply. Women Leaders on the Uptake of
Renewable Energy Seminar. Perth, Australia, June 2001. Also available at:
[Link] (accessed: March 04)
Merten, Ulrich and (Editor) (1966). Desalination by Reverse Osmosis. Cambridge,
Massachusetts, M.I.T. Press
Millero, F.J. (1996). Chemical Oceanography, CRC Press. Also available at:
[Link] (accessed: March 04)
Miranda, Marcos S. (2003). Small-Scale Wind-Powered Seawater Desalination Without
Batteries, Ph.D Thesis (Forthcoming), Loughborough University
Norris, W. T. (1999). The hundred year hunt. Power Engineering Journal 13(3): 184
Oklejas, Eli (2002). Energy Efficiency Considerations for RO plants: A Method for
Evaluation. International Desalination and Water Reuse Quarterly 11(4): 26-39
Oldach, Rolf (2001). Matching renewable energy with desalination plants, MEDRC
Project: 97-AS-006a, September 2001, IT Power Ltd. Sponsored by MEDRC:
The Middle East Desalination Research Center, Muscat, Sultanate of Oman. Also
available at: [Link] (accessed: March 04)
Parish, Oliver (1999). Economic aspects. Stand-Alone Photovoltaic Applications:
Lessons Learned. G. Loois and B. V. Hemert. London, James and James pp 10-13
Permar, Clark (1995). Liquid treatment apparatus for providing a flow of pressurized
liquid. United States Patent 5,462,414
Pinkerton, Harry E. (1979). Fluid handling system. United States Patent 4,178,240
Rahal, Zeina (2001). Wind powered desalination, Ph.D Thesis, Loughborough University
Reid, C. E. (1966). Principles of Reverse Osmosis. Desalination by Reverse Osmosis. U.
Merten. Cambridge, Massachusetts, M.I.T. Press
Richards, Bryce S. and Andrea I. Schafer (2003). Photovoltaic-powered desalination
system for remote Australian communities. Renewable Energy 28(13): 2013-2022
Roberts, Simon (1991). Solar electricity : a practical guide to designing and installing
small photovoltaic systems. New York, Prentice Hall International
Rodriguez-Girones, R J., Miriam Rodriguez Ruiz and Jose M. Veza (1996). Experience
on Desalination with Renewable Energy Sources, Work Package 1, EURORED,

193
 APAS RENA-CT94-0063, March 1996, Universidad de Las Palmas de Gran
Canaria
Romo, Rodrigo F. V. and M. Michael Pitts (1999). Application of Electrotechnology for
Removal and Prevention of Reverse Osmosis Biofouling. Environmental
Progress 18(2): 107
Sardi, L. (1996). R.O. Deslainators Powered by PV Systems for Small/Medium Italian
Islands. Proceedings of the Mediterranean Conference on Renewable Energy
Sources for Water Production. Santorini, Greece, June 1996 pp 36-44
Sauer, D. U., M. Bachler, G. Bopp, W. Hohe, J. Mittermeier, P. Sprau, B. Willer and M.
Wollny (1997). Analysis of the performance parameters of lead/acid batteries in
photovoltaic systems. Journal of Power Sources 64(1-2): 197-201
Schwarz, Joshua (2000). Beach well intakes for small seawater reverse osmosis plants,
MEDRC 97-BS-015, August 2000, TAHAL Consulting Engineers Ltd, Tel Aviv,
Israel. Also available at: [Link] (accessed: March 04)
Smith, David (2000). Battery-powered desalter relies on unique pressure pump.
International Desalination and Water Reuse Quarterly 10(3): 18-22
Snoke, J. A. (2002). GEOL 4104: Properties of Seawater Physical Oceanography. Also
available at: [Link] (accessed:
March 04)
Snyder, A. E. (1966). Freezing Methods. Principles of Desalination. K. S. Spiegler. New
York, Academic Press
Spectra (2003). Clark Pump Field Repair & Rebuild Manual, 22/08/03, Spectra
Watermakers, San Rafael, California. Also available at:
[Link] (accessed: March
04)
Spiegler, K.S. and Y.M. El-Sayed (1994). A Desalination Primer Introductory book for
studcms and newcomers to desalination, Balaban Desalination Publications
Stewart, Robert H. (2002). Introduction To Physical Oceanography, Department of
Oceanography
Texas A & M University. Also available at:
[Link]
tml (accessed: March 04)
Thomson, Murray, Jo Gwillim, Andrew Rowbottom, Ian Draisey and Marcos Miranda
(2001). Batteryless Photovoltaic Reverse-Osmosis Desalination System,
S/P2/00305/REP, ETSU, DTI, UK. Also available at:
[Link]
(accessed: March 04)
Tonner, Jodie (1999). Standard Seawater, 13 April, 1999, World Wide Water. Also
available at: [Link]
[Link]/PDFs/Std%20SW%20ions%[Link] (accessed: March 04)
Tzen, Eftihia and Marios Sigalas (2003). Design and Development of a Hybrid
Autonomous System for Seawater Desalination. Proceedings of Renewable
Energy Sources for Islands, Tourism and Water Desalination Conference. Crete,
Greece, May 2003, EREC (European Renewable Energy Council) pp 231-238

194
UNESCO (2003). The United Nations World Water Development Report - Water for
People - Water for Life (Executive Summary), UNESCO Publishing / Berghahn
Books
Wagner, Jørgen (2001). Membrane Filtration Handbook - Practical Tips and Hints,
Osmonics. Also available at: [Link]
(accessed: March 04)
Wangnick, Klaus (2002). IDA Worldwide Desalination Plants Inventory, Report No. 17,
Wangnick Consulting GmbH
Weber-Shirk, Monroe L. (2003). Laboratory Measurements and Procedures. Also
available at:
[Link]
[Link] (accessed: June 03)
Weiner, Dan, David Fisher, Eduard J. Moses, Baruch Katz and Giora Meron (2001).
Operation experience of a solar- and wind-powered desalination demonstration
plant. Desalination 137(1-3): 7-13
Wenham, Stuart R., Martin A. Green and Muriel. E. Watt (1994). Applied Photovoltaics,
Centre for Photovoltaic Devices and Systems, Australia
WHO (2003). Guidelines for Drinking-Water Quality (Draft) 3rd Edition, 17 February
2003, World Health Organization
Wilamowski, Bogdan M and Xiangli Li (2002). Fuzzy System Based Maximum Power
Point Tracking for PV System. IEEE
Wilson, Leslie P. S. (1983). Reverse osmosis liquid purification apparatus. United States
Patent 4,367,140
Wiseman, Robin (2002). IDA Desalination Inventory: Installed capacity doubles in less
than two years. International Desalination and Water Reuse Quarterly 12(3): 10-
13

195
 Appendix A Details of testing carried out using the
initial test rig

This appendix provides details of the testing outlined in Chapter 3.

A.1 Instrumentation and data-acquisition

The instrumentation and data-acquisition system used with the initial test rig was based
on National Instruments hardware and LabView software as described in Chapter 4.
Analogue signals were sampled at 1 kHz and then averaged and logged at 1 Hz. Pulse
signals were converted to frequency and also logged at 1 Hz. The sensors were broadly
as described in Chapter 4, though the following details should be noted.

Electrical input power to the inverter was measured using a Elcontrol VIP96-APQ power
meter, which has an integral digital display of real and reactive power and provides
analogue outputs representing each. These were connected to analogue inputs of the
LabView-based data-acquisition system, which was calibrated simply to match the digital
display on the Elcontrol power meter. Unfortunately, the calibration of the Elcontrol
meter itself was undermined by the fact that the current drawn by the inverter is non-
sinusoidal and thus the collected data should be viewed with a degree of caution. This
matter was addressed later by dispensing with the Elcontrol power meter and replacing it
with instantaneous voltage and current sensors, sampled at 10-kHz directly into
LabView, as described in section 4.16.

The flow meters used for this testing were the turbine flow meters detailed in section B.1,
which were found to be problematic and were later replaced by the oval-gear flow
meters.

Thus, the data presented in this appendix should be viewed with caution. Nonetheless,
the testing revealed various matters that were valuable in the design of the new test rig
and its instrumentation, as discussed below.

196
A.2 Procedure and results
The data shown in Table 3-1 was collected from a single run, 3¾ hours long, during
which the speed setpoint was ramped up and down three times in steps of 30 rpm, as
shown in Figure A-1.

1200

1000

800
Speed setpoint (rpm)

600

400

200

0
9:15 9:45 10:15 10:45 11:15 11:45 12:15 12:45
Time (hh:mm)

Figure A-1 – Speed setpoint profile

197
 1200

1000

800
Speed (rpm)

600

400

200

0
9:15 9:45 10:15 10:45 11:15 11:45 12:15 12:45
Time (hh:mm)

Figure A-2 – Measured speed

The measured speed shown in Figure A-2 closely matches the setpoint shown in
Figure A-1, except for the irregular steps between 300 and 400 rpm. These are due to a
mode change within the inverter software.

198
 1600

1400

1200
Inverter input power (W)

1000

800

600

400

200

0
9:15 9:45 10:15 10:45 11:15 11:45 12:15 12:45
Time (hh:mm)

Figure A-3 – Measured inverter input power consumption

Figure A-3 shows the input power to the inverter, and shows the aforementioned mode
change more markedly. The inverter in use during this test was a Siemens Micromaster
standard industrial drive, which has one mode for normal operation and another for low-
speed operation. This improves the performance of the drive in normal industrial
applications, where the addition power drawn in the low-speed mode would normally go
unnoticed. But for operation direct from a PV array, as discussed in this thesis, it would
considerably complicate maximum power point tracking. The inverters used later, with
the new test rig, were FKI brand, as described in section 5.7, and did not exhibit any
mode-change affects.

199
 0.35

0.3

0.25
Feed flow (L/s)

0.2

0.15

0.1

0.05

0
0 200 400 600 800 1000 1200
Speed (rpm)

Figure A-4 – Measured feed flow versus pump speed

Figure A-4 shows that the feed flow through the plunger pump is approximately equal to
its shaft speed, as expected with a positive displacement pump. The scatter in the figure
is due primarily to the inaccuracy of the turbine flow meter in use at the time.

200
 30

28

26
Water temperature (ºC)

24

22

20

18
9:15 9:45 10:15 10:45 11:15 11:45 12:15 12:45
Time (hh:mm)

Figure A-5 – Measured water temperature

The water for the test rig is circulated through the tank as shown in Figure 3-1and heats
up by virtue of the energy introduced by the two pumps. At the time of the testing
described here, there was no cooling mechanism and water temperature rose as shown in
Figure A-3. The peaks around 10:20 and 11:45 are where the water is starting to flow
again after being stationary for while, see Figure A-2, and yet being heated by the
constant-speed centrifugal pump.

201
 70
Feed

60

50
Pressure (bar)

40

30 Concentrate

20

10

0
9:15 9:45 10:15 10:45 11:15 11:45 12:15 12:45
Time (hh:mm)

Figure A-6 – Measured feed and concentrate pressures

The measured pressures in Figure A-6 indicate a delta pressure of up to 4 bar, which is
unusually high and was caused partly the poor condition of the membranes and partly
because they are small-diameter (2½-inch) membrane elements.

202
 0.03
Product flow (L/s)

0.02

0.01

0
9:15 9:45 10:15 10:45 11:15 11:45 12:15 12:45
Time (hh:mm)

Figure A-7 – Measured product flow

The three main peaks, shown in Figure A-7 (9:30, 10:55 and 12:15), are progressively
higher, due to the rising water temperature, shown in Figure A-3.

The periods of virtually no product flow, shown in Figure A-7, from 10:00 until 10:30,
from 11:25 to 11:50 and after 12:40 are where the feed pressure, shown in Figure A-6, is
below the osmotic pressure ~23 bar. The random product flow recorded during these
periods is due primarily to osmotic suck-back: the product being sucked back through the
membranes by (forward) osmosis. This introduced air in to the product flow pipe and
meter, which then gave erratic data until the air was expelled by positive product flow
when the feed pressure returned above the osmotic pressure.

203
 0.03
Product flow (L/s)

0.02

0.01

0
0 10 20 30 40 50 60 70
Feed pressure (bar)

Figure A-8 – Measured product flow versus feed pressure

Figure A-8 illustrates, as expected, that the product flow is roughly proportional to the
net driving pressure (the feed pressure less the osmotic pressure). It also highlights the
aforementioned affects of water temperature and osmotic suck-back and is marred overall
by the errors introduced by the turbine product flow meter, discussed in section B.1.1.
Recognition of these problems led to the introduction of the temperature control system
discussed in section 4.13 and the oval-gear flow meters.

204
 0.03
Product flow (L/s)

0.02

0.01

0
0 200 400 600 800 1000 1200 1400 1600
Inverter input power (W)

Figure A-9 – Measured product flow versus inverter input power consumption

Figure A-9 again shows the spurious product flow due to suck-back between 200 and
300 W. Ignoring this, it can be seen that water production starts below 400 W and
increases to a maximum of ~0.029 L/s at around 1500 W.

205
 40

35

30
Specific energy (kWh/m3)

25

20

15

10

0
0 200 400 600 800 1000 1200 1400 1600
Inverter input power (W)

Figure A-10 – Specific energy versus inverter input power

input power ( W )
specific energy (kWh/m 3 ) =
product flow (L/s) × 3600

Figure A-10 shows the specific energy consumption during the third ramp-up/ramp-
down, in particular, between 11:50 and 12:43. This data was selected in order to
eliminate spurious data caused by the osmotic suck-back described earlier.

It is apparent that the specific energy is roughly constant between 600 W and 1500 W.
This was encouraging in the design of the system intended to operate at variable power
directly from a PV array.

The average specific energy across this band is around 13 kWh/m3, which is very high,
perhaps more typical of system without energy recovery, and is caused primarily by the
use of very old membranes, during this testing.

206
 1200
Start

1000
Product concentration (mg/L)

800

600

400 End

200

0
0 200 400 600 800 1000 1200 1400 1600
Inverter input power (W)

Figure A-11 – Measured product concentration versus inverter input power

Figure A-11 shows the concentration of the product during the same period: between
11:50 and 12:43. Again this data was selected in order to eliminate spurious data caused
by osmotic suck-back; in particular, the concentration sensor was affected by the air
introduced.

At the start (11:50), the pressure is increasing and the product starts to flow. Initially, it
has very high concentration, exceeding the range of the conductivity meter and the
recorded data is limited to ~1100 mg/L. The excessive concentration is due to the salt
that diffused through the membranes while there was no flow.

As the product flow increases, the concentration falls rapidly to about 400 mg/L at
700 W. Thereafter, the concentration falls to below 300 mg/L at full power.

As the power is reduced, the concentration starts to rise again, and is a little higher than
shown at the same power on the ramp-up; this is due to the water temperature being
higher, as shown in Figure A-5.

Finally, the concentration rises to around 500 mg/L at 400 W.

207
A.2.1 Results data summary table
In order to summarise the data presented in the preceding graphs, three nominal input
powers were chosen: 500 W, 1,000 W and 1,500 W, and the data was binned at these
powers +/– 50 W. The data in these three bins was then averaged to produce Table 3-1.

Electrical input power (W) 502 997 1515

Speed (rpm) 494 864 1130
Electric motor torque (Nm) 6.9 8.7 10.3
Feed pressure (bar) 38.7 58.9 68.9
Concentrate pressure (bar) 37.7 56.4 65.1
Feed concentration (mg/L) 29000 29000 29000
Product concentration (mg/L) 550 283 262
Feed flow (L/s) 0.134 0.233 0.300
Product flow (L/s) 0.0075 0.022 0.0291
Specific energy (kWh/m3) 18.6 12.6 14.4

Table 3-1 – Repeated here for convenience

A.3 Power flow and efficiency calculations for Sankey diagram

The power flows and efficiencies shown on the Sankey diagram of Figure 3-2, are
calculated from Table 3-1, as follows. The values shown below correspond to the middle
column of Table 3-1.

electric motor shaft power = speed × torque

2π rad s
= 864 rpm × × 8.7 Nm
60 rpm

= 787 W

inverter & motor losses = electrical input power − electric motor shaft power

= 997 W − 787 W

= 210 W

208
feed power = feed flow × feed pressure

m3 10 5 Pa
= 0.233 L s × × 58.9 bar ×
1000 L bar

= 1372 W

concentrate flow = feed flow − product flow

= 0.233 L s − 0.022 L s

= 0.211L s

crossflow loss = concentrate flow × delta pressure

= concentrate flow × ( feed pressure − concentrate pressure )

m3 10 5 Pa
= 0.211L s × × (58.9 bar − 56.4 bar ) ×
1000 L bar

= 53 W

concentrate power = concentrate flow × concentrate pressure

m3 10 5 Pa
= 0.211 L s × × 56.4 bar ×
1000 L bar

= 1190 W

From section 5.11.6, NaCl solution at 32,800 mg/L has an osmotic pressure of 25.8 bar.
The testing described here used NaCl solution at 29,000 mg/L:

29000 mg L
osmotic pressure = 25.8 bar ×
32800 mg L

= 23 bar

209
desalination power = product flow × osmotic pressure

m3 10 5 Pa
= 0.022 L s × × 23 bar ×
1000 L bar

= 51 W

throughflow loss = feed power − crossflow loss − concentrate power − desalination power

= 1372 W − 53 W − 1190 W − 51 W

= 78 W

Returning to the plunger pump, an efficiency of 85% is assumed, see section 5.3.

feed power
pump shaft power =
pump efficiency

1372 W
=
0.85

= 1614 W

pump loss = pump shaft power − feed power

= 1614 W − 1372 W

= 242 W

recoverd power = pump shaft power − electric motor shaft power

= 1614 W − 787 W

= 827 W

210
 recoverd power
saving given by energy recovery =
pump shaft power

827 W
=
1614 W

= 51 %

The toothed belt is assumed to have an efficiency of 95%.

recoverd power
hydraulic motor output power =
belt efficiency

827 W
=
0.95

= 871 W

hydraulic motor output power

hydraulic motor efficiency =
concentrate power

871 W
=
1190 W

= 73 %

water-to-water efficiency = hydraulic motor efficiency × belt efficiency × pump efficiency

= 73 % × 95 % × 85 %

= 59 %

The results of the above calculations are summarised in Table A-1, along with those for
the other columns of Table 3-1.

211
Electrical input power (W) 502 997 1515
Motor shaft power (W) 357 787 1219
Inverter & motor losses (W) 145 210 296
Inverter & motor efficiency 71% 79% 80%
Feed water power (W) 519 1372 2067
Crossflow loss (W) 13 53 103
Concentrate power (W) 477 1190 1764
Desalination power (W) 17 51 67
Throughflow loss (W) 12 78 133
Pump shaft power (W) 611 1614 2432
Pump loss (W) 92 242 365
Recovered power (W) 254 827 1213
Saving given by energy recovery 42% 51% 50%
Hydraulic motor output power 267 871 1277
Belt loss (W) 13 44 64
Hydraulic motor efficiency 56% 73% 72%
Water-to-water efficiency 45% 59% 58%
Table A-1 – Caluclated powers and efficeincies

212
 Appendix B Sensor details and calibration

B.1 Turbine flow meters

The initial test rig, described in Chapter 3, was fitted with turbine flow meters, which are
described in this section.

(Oval-gear flow meters, see sections 4.9.2 and B.2, were used for the testing of the Clark
pump and are fitted to the new test rig.)

B.1.1 Product-flow turbine flow meter

• RS part number 257-149

• Pulse output

0.07

0.06

0.05

0.04
Flow (L/s)

0.03

0.02

0.01

0
0 50 100 150 200 250 300 350
Sensor frequency (Hz)

Figure B-1 – Product-flow turbine flow meter calibration

213
The product-flow turbine flow meter was calibrated using a “bucket and stopwatch”. The
flow was adjusted across the range of interest by means of the variable-speed drive on the
pump and a total of 120 measurements were made. The stopwatch was implemented in
LabView and operated manually, and the time taken to fill bucket ranged from 25 to 695
seconds. The “bucket” was actually a flask with a narrow neck; its volume was 1.665 L.
The sensor frequency was measured in LabView. Figure B-1, shows the calibration
results and the fitted straight line:

flow = 0.000210 × frequency − 0.00117

This straight line was compared against the original data and the normalised error
calculated:

RMS (errors)
Normalised error = = 2.4%
RMS (original values)

where RMS is root-mean-square. A normalised error of 2.4% was considered excessive,

but worse than this, a subsequent spot check of the calibration revealed a drift of 13%,
which renders the data useless for the study of efficiencies etc. and instigated the
replacement of the turbine flow meters with oval-gear flow meters.

B.1.2 Feed flow

• RS part number 257-026

• Internal pulse-to-analogue converter

• Manufacturer’s claimed accuracy +/– 2%

214
 0.3

0.25

0.2
Flow (L/s)

0.15

0.1

0.05

0
0.8 0.9 1 1.1 1.2 1.3 1.4
Voltage (V)

Figure B-2 – Feed-flow turbine flow meter calibration

The feed-flow turbine flow meter was also was calibrated using a bucket and stopwatch.
Twenty measurements were made and a straight line fitted:

flow = 0.4862 × voltage − 0.3903

which gave a normalised error of less than 1%. However, a subsequent spot check
revealed a drift of 4%, which was not considered acceptable, and lead to this meter also
being replaced.

215
B.2 Oval-gear flow meter calibration sheets

Figure B-3 – Clark-pump-inlet oval-gear flow meter calibration

216
Figure B-4 – Plunger-pump-inlet oval-gear flow meter calibration

217
Figure B-5 – Product-flow oval-gear flow meter calibration

218
B.3 Medium-pressure sensor specification and calibration sheet

Figure B-6 – Medium-pressure sensor specification and calibration

219
B.4 Concentration from conductivity
µS/cm ppm µS/cm ppm µS/cm ppm µS/cm ppm µS/cm ppm µS/cm ppm µS/cm ppm
10 5 610 302 1400 702 3200 1644 8200 4430 18750 10708 44000 26357
20 9 620 307 1420 713 3250 1671 8300 4486 19000 10852 45000 27035
30 14 630 312 1440 723 3300 1699 8400 4542 19250 11015 46000 27713
40 19 640 317 1460 733 3350 1726 8500 4598 19500 11169 47000 28391
60 28 650 323 1480 743 3400 1753 8600 4654 19750 11323 48000 29069
70 33 660 328 1500 754 3450 1781 8700 4710 20000 11476 49000 29747
80 38 670 333 1525 766 3500 1808 8800 4767 20250 11630 50000 30425
90 42 680 338 1550 770 3550 1835 8900 4823 20500 11784 51000 31103
100 47 690 343 1575 792 3600 1863 9000 4879 20750 11937 52000 31781
110 52 700 348 1600 805 3650 1899 9100 4935 21000 12091 53000 32459
120 57 710 353 1625 817 3700 1917 9200 4991 21250 12245 54000 33137
130 61 720 358 1650 830 3750 1945 9216 5000 21500 12399 55000 33815
140 66 730 363 1675 843 3800 1972 9300 5047 21750 12552 56000 34493
150 71 740 368 1700 856 3850 1999 9400 5103 22000 12705 57000 35171
160 75 750 373 1725 868 3900 2027 9500 5159 22250 12860 58000 35849
170 80 760 378 1750 881 3950 2054 9600 5215 22500 13013 59000 36527
180 85 770 383 1775 894 4000 2081 9700 5271 22750 13167 60000 37205
190 90 780 388 1800 907 4100 2136 9800 5327 23000 13321 61000 37883
200 95 790 393 1825 920 4200 2191 9900 5383 23250 13474 62000 38561
210 100 800 399 1850 932 4300 2245 10000 5439 23500 13628 63000 39239
220 105 810 404 1875 945 4400 2300 10200 5551 23750 13782 64000 39917
230 110 820 409 1900 958 4500 2356 10400 5664 24000 13936 65000 40595
240 115 830 414 1925 971 4600 2412 10600 5776 24250 14089 66000 41273
250 120 840 419 1950 983 4700 2468 10800 5888 24500 14243 67000 41961
260 125 850 424 1975 996 4800 2524 11000 6000 24750 14397 68000 42629
270 130 860 429 2000 1000 4900 2580 11200 6122 25000 14550 69000 43307
280 135 870 434 2025 1022 5000 2636 11400 6243 25500 14858 70000 43985
290 140 880 439 2050 1034 5100 2692 11600 6364 26000 15165 71000 44663
300 145 890 444 2075 1047 5200 2748 11800 6485 26500 15473 72000 45341
310 150 900 449 2125 1073 5300 2805 12000 6607 27000 15780 73000 46091
320 155 910 454 2150 1085 5400 2861 12200 6728 27500 16087 74000 46697
330 160 920 459 2175 1098 5500 2917 12400 6843 28000 16395 76000 48053
340 165 930 464 2200 1111 5600 2973 12600 6970 28500 16702 77000 48731
350 171 940 469 2225 1124 5700 3029 12800 7091 29000 17010 78000 49409
360 176 950 474 2250 1137 5800 3085 13000 7213 29500 17317 79000 50087
370 181 960 480 2275 1140 5900 3141 13200 7334 30000 17624 80000 50765
380 186 970 485 2300 1162 6000 3197 13400 7455 30500 17932 81000 51443
390 191 980 490 2325 1175 6100 3253 13600 7576 31000 18239 82000 52121
400 196 990 495 2350 1188 6200 3309 13800 7898 31500 18547 83000 52799
410 201 1000 500 2375 1200 6300 3365 14000 7819 32000 18854 84000 53477
420 206 1020 510 2400 1213 6400 3421 14200 7940 32500 19161 85000 54155
430 211 1040 520 2425 1226 6500 3477 14400 8061 33000 19469 86000 54833
440 216 1080 540 2450 1239 6600 3533 14600 8182 34000 20084 87000 55511
450 221 1100 550 2475 1251 6700 3589 14800 8304 34500 20391 88000 56130
460 226 1120 561 2500 1264 6800 3645 15000 8425 35000 20698 89000 56867
470 231 1140 571 2550 1290 6900 3701 15250 8576 35500 21006 90000 57545
480 236 1160 581 2600 1315 7000 3758 15500 8728 36000 21313 91000 58223
490 241 1180 591 2650 1344 7100 3814 15750 8879 36500 21621 92000 58901
500 247 1200 601 2700 1371 7200 3870 16000 9031 37000 21928 93000 59579
510 252 1220 611 2750 1398 7300 3926 16250 9182 37500 22235 94000 60257
520 257 1240 621 2800 1426 7400 3982 16500 9334 38000 22543 95000 60935
530 262 1260 632 2850 1453 7500 4038 16750 9486 38500 22850 96000 61613
550 272 1280 642 2900 1480 7600 4094 17000 9637 39000 23158 97000 62291
560 277 1300 652 2950 1508 7700 4150 17500 9940 39500 23465 98000 62969
570 282 1320 662 3000 1535 7800 4206 17750 10092 40000 23773 99000 63647
580 287 1340 672 3050 1562 7900 4262 18000 10247 41000 24387 100000 64325
590 292 1360 682 3100 1589 8000 4318 18250 10400 42000 25002
600 297 1380 692 3150 1617 8100 4374 18500 10554 43000 25679

Table B-1 – Conductivity (µS/cm) and concentration (ppm) data for sodium
chloride solution (DOW 1995 section 10.2)

220
The conductivity of sodium chloride solution increases with the concentration but the
relationship is not linear. DOW (1995) provide the data shown in Table B-1 as a means
of converting from one to the other. In order to program the conversion in LabView (or
any programming language) a mathematical function is more convenient than a look up
table.

70000

60000
Poorly fitted trendline:
y = 8.25137E-07x 2 + 5.67947E-01x - 1.43062E+02
50000 R2 = 9.99927E-01
Concentration (ppm)

40000

30000

20000

10000

0
0 10000 20000 30000 40000 50000 60000 70000 80000 90000 100000
Conductivity (µS/cm)

Figure B-7 – Concentration versus conductivity – linear scales

Figure B-7 shows the data from Table B-1 and a second-order polynomial trendline. The
trendline appears to be a very good fit and this is supported by the R2 value. However,
close inspection below 1000 ppm reveals that the fit is actually very poor in this critical
region, as illustrated in Figure B-8.

221
 100000

10000
Concentration (ppm)

1000

100
Improved trendline

Poorly fitted trendline

10

1
1 10 100 1000 10000 100000
Conductivity (µS/cm)

Figure B-8 – Concentration versus conductivity – logarithmic scales

The improved trendline, shown in Figure B-8 was obtained by taking the logarithm of
both concentration and conductivity prior to fitting a second-order polynomial. The
equation of the improved trendline is:

(
y = exp 0.00565 × (ln x ) + 0.952 × ln x − 0.642
2
)
where y is the concentration in ppm, and x is conductivity in µS/cm.

222
 4

0
Concentration errors (%)

1 10 100 1000 10000 100000

-2

-4

-6

-8
Conductivity (µS/cm)

Figure B-9 – Concentration trendline errors

Figure B-9 shows the percentage error between the improved trendline and the original
data. It suggests that the original data is itself a collection of four or five segments, and
that no further refinement of the trendline would be profitable.

223
B.5 Torque load-cell and calibration

B.5.1 Load-cell datasheet

Figure B-10 – Torque load-cell manufacturer’s test datasheet

224
B.5.2 Torque measurement calibration

30

25

20
Torque (Nm)

15

10

0
0.000 0.001 0.002 0.003 0.004 0.005 0.006 0.007
Voltage (V)

Figure B-11 – Torque measurement calibration

Torque (Nm) Voltage (V)

1.10 0.00025
2.20 0.00051
3.31 0.00077
4.41 0.00103
5.51 0.00128
6.61 0.00154
7.72 0.00179
8.82 0.00203
9.92 0.00229
11.02 0.00256
12.13 0.00280
13.23 0.00305
14.93 0.00342
16.64 0.00378
18.34 0.00419
20.04 0.00457
21.74 0.00496
23.45 0.00536
27.99 0.00639

Table B-2 – Data for Figure B-11

225
Figure B-11 shows the torque measurement calibration results and the fitted straight line:

torque = 4396 × voltage − 0.1049

The linearity is good: the largest error from the straight line was 0.12 Nm, which was
considered acceptable.

226
B.6 Irradiance pyranometer calibration certificate

Figure B-12 – Kipp and Zonen CM11 pyranometer calibration certificate

227
 Appendix C Clark pump details, test results and
analysis

C.1 Specification
Spectra Watermakers Inc. supply complete desalination systems that incorporate Clark
pumps; they do not normally supply Clark pumps separately, and so do not normally
supply detailed specifications of the Clark pump in isolation. The data used in this thesis
is a follows.

Maximum flow 3.5 US gallons per minute 0.22 L/s

Maximum pressure 950 psi 65 bar
Piston diameter 2.75 inches 69.8 mm
Rod diameter 0.875 inches 22.2 mm
Table C-1 – Clark pump specification

From the rod and the piston diameters, the theoretical recovery ratio Rt = 0.10124.

C.2 Test procedure

As outlined in section 5.2.2, the Clark pump was tested at a total of 55 discrete test
points. These were the 45 permutations of:

• feed flow: 0.042, 0.127 and 0.211 L/s,

• high pressure: 20, 30, 40, 50 and 60 bar,

• delta pressure: 0, 2 and 4 bar,

plus the 10 permutations of:

• feed flow: 0.085and 0.17 L/s,

• high pressure: 20, 30, 40, 50 and 60 bar,

• delta pressure: 0 bar.

These test points were the nominal targets for the manual adjustment of the valves,
shown in Figure 5-6, and the speed of the feed-pump motor. (The target motor speeds
were round numbers, which accounts for the odd choice of nominal feed flows.) Once the
adjustments were made (for each of the 55 discrete test points) and steady operation had

228
been achieved, data logging (via the LabView data acquisition system) was enabled for a
period of at least a minute. The data was then averaged over this test period and each test
provides one row in Table C-2. Of course, the manual adjustment of the valves was not
perfect and motor speed of the feed pump reduced slightly with pressure, and so the
measured values do differ slightly form the nominal targets; the measured values were
used for the analysis and modelling.

229
C.3 Measured data
High Concentrate Delta Feed
Feed Flow Product Flow
Pressure Pressure Pressure Pressure
(L/s) (L/s)
(bar) (bar) (bar) (bar)
0.04318 20.065 20.056 0.009 0.004182 2.597
0.04233 29.840 29.834 0.006 0.004029 3.655
0.04238 39.802 39.795 0.008 0.003966 4.733
0.04239 49.930 49.922 0.008 0.003884 5.821
0.04240 60.075 60.066 0.009 0.003794 6.895
0.08491 19.842 19.795 0.046 0.008240 2.966
0.08487 30.011 29.964 0.047 0.008034 4.013
0.08467 39.787 39.728 0.059 0.007895 5.051
0.08488 49.905 49.852 0.053 0.007727 6.067
0.08489 60.171 60.114 0.057 0.007614 7.153
0.12742 19.812 19.784 0.027 0.012417 3.500
0.12742 29.907 29.875 0.031 0.012120 4.533
0.12741 40.094 40.061 0.033 0.011859 5.580
0.12746 49.997 49.971 0.026 0.011573 6.562
0.12738 59.917 59.895 0.022 0.011452 7.531
0.16930 20.093 19.997 0.096 0.016718 4.181
0.16932 30.105 30.002 0.103 0.016195 5.220
0.16925 39.818 39.708 0.111 0.015878 6.173
0.16934 50.142 50.027 0.115 0.015429 7.163
0.16930 59.897 59.781 0.117 0.015245 8.133
0.21101 19.775 19.704 0.071 0.021093 4.970
0.21093 30.174 30.096 0.078 0.020296 6.051
0.21082 40.024 39.949 0.074 0.019927 7.036
0.21082 49.845 49.758 0.087 0.019302 7.927
0.21078 59.763 59.670 0.093 0.019130 8.896
0.04245 19.773 17.729 2.043 0.004115 4.529
0.04244 29.782 27.765 2.017 0.004042 5.602
0.04239 39.716 37.585 2.131 0.003963 6.779
0.04248 49.882 47.951 1.931 0.003907 7.661
0.04247 59.882 57.833 2.049 0.003799 8.799
0.12734 19.913 17.874 2.039 0.012334 5.347
0.12737 29.965 27.993 1.972 0.012099 6.361
0.12733 40.012 38.052 1.961 0.011904 7.409
0.12743 49.973 48.009 1.965 0.011658 8.436
0.12727 60.029 58.059 1.970 0.011479 9.436
0.21086 19.859 17.803 2.056 0.020901 6.825
0.21128 29.607 27.536 2.071 0.020274 7.861
0.21072 40.171 38.125 2.046 0.019848 8.935
0.21079 49.968 47.910 2.057 0.019328 9.828
0.21071 59.436 57.508 1.927 0.019045 10.600
0.04241 20.014 15.921 4.094 0.004156 6.546
0.04247 29.794 25.884 3.910 0.004070 7.440
0.04234 39.737 35.640 4.097 0.003979 8.712
0.04251 49.537 45.668 3.869 0.003881 9.534
0.04249 59.831 55.952 3.879 0.003865 10.627
0.12740 19.877 15.872 4.005 0.012371 7.242
0.12739 29.871 25.832 4.039 0.012181 8.349
0.12739 40.032 36.006 4.026 0.011947 9.431
0.12736 49.948 45.919 4.029 0.011686 10.404
0.12727 59.863 55.832 4.031 0.011552 11.438
0.21056 19.919 15.808 4.111 0.020964 8.759
0.21083 29.773 25.635 4.138 0.020335 9.796
0.21075 39.699 35.543 4.157 0.019962 10.854
0.21064 49.868 45.705 4.163 0.019452 11.855
0.21066 60.343 56.195 4.148 0.019183 12.885
Table C-2 – Clark pump test results data

230
The first four columns of Table C-2 represent the test conditions, while the last two
columns are the results.

With the exception of the delta pressure, the data in each column is that from an
individual sensor as shown in Figure 5-6.

C.4 Analysis
The data shown in Table C-2 was analysed in order to quantify leakages, pressure losses
and efficiencies, according to the definitions of these quantities given in sections 5.2.1
and 5.2.4.

In particular:

Cross-references to
sections 5.2.1 and 5.2.4
Combining and rearranging
Leakages QL = QF Rt − QP
equations 17 and 19

Pressure losses PL = PF − PH − (1 − Rt )(PE − PC ) Rearranging equation 35

QL
Volumetric efficiency ηV = 1 − Equation 22
QF Rt

PL
Mechanical efficiency ηM = 1 − Equation 34
PF − PE (1 − Rt )

QH PH − QF PF
Overall efficiency η Clark = Equation 38
QC PC − QE PE

Table C-3 – Clark pump analysis equations

Where:

QF is feed flow
QH is high-pressure flow
QP is product flow
QC is concentrate flow
QE is exhaust flow (discharge)
QL is leakage flow
PF is feed pressure (medium pressure)
PH is high pressure

231
 PC is concentrate pressure
PE is exhaust pressure (discharge)
∆P is delta pressure = PH – PC
Rt is theoretical recovery ratio = 0.10124, see section C.1.
The exhaust pressure PE was taken as a constant 0.1 bar, corresponding to 1 metre head
of water between that Clark pump and the water recirculation tank.

The high-pressure flow QH and the concentrate flow QC could not be measured because
the oval-gear flow meters on the test rig are limited to 20 bar. Instead, it was assumed
that QH = QF, and therefore, by virtue of equation 25: QF + QC = QH + QE , that QC = QE.
Fortunately, this assumption only affects the last equation in Table C-3: the overall
QH PH − QF PF
efficiency: η Clark = , and it affects it very little:
QC PC − QE PE

In practice, QH would be slightly less than QF, but, again by equation 25, QC would also
be less than QE by exactly the same amount. And, because PH ≈ PC and PF ≈ PE, the
error in the numerator will be largely cancelled out by a very similar error in the
denominator.

232
Leakages Pressure Losses Volumetric Mechanical Overall
(L/s) (bar) Efficiency Efficiency Efficiency
0.000189 0.467 95.7% 81.4% 96.9%
0.000256 0.539 94.0% 84.9% 97.3%
0.000324 0.607 92.4% 86.9% 97.5%
0.000407 0.669 90.5% 88.3% 97.5%
0.000499 0.715 88.4% 89.5% 97.4%
0.000356 0.825 95.9% 71.3% 94.9%
0.000558 0.843 93.5% 78.5% 96.2%
0.000676 0.880 92.1% 82.3% 96.7%
0.000867 0.878 89.9% 85.3% 96.9%
0.000980 0.920 88.6% 87.0% 97.0%
0.000483 1.380 96.3% 59.5% 91.8%
0.000780 1.387 94.0% 68.8% 94.2%
0.001040 1.401 91.9% 74.5% 95.2%
0.001331 1.387 89.7% 78.6% 95.8%
0.001444 1.355 88.8% 81.8% 96.3%
0.000422 1.971 97.5% 51.8% 88.7%
0.000947 1.989 94.5% 61.2% 92.0%
0.001257 1.952 92.7% 67.9% 93.7%
0.001714 1.894 90.0% 73.2% 94.7%
0.001895 1.874 88.9% 76.7% 95.3%
0.000269 2.815 98.7% 42.3% 83.9%
0.001059 2.837 95.0% 52.4% 89.0%
0.001416 2.827 93.4% 59.3% 91.4%
0.002042 2.713 90.4% 65.4% 92.9%
0.002210 2.672 89.6% 69.7% 93.9%
0.000183 0.601 95.7% 86.5% 95.7%
0.000255 0.685 94.1% 87.6% 96.6%
0.000328 0.753 92.4% 88.7% 96.9%
0.000394 0.785 90.8% 89.6% 97.2%
0.000500 0.806 88.4% 90.7% 97.2%
0.000558 1.408 95.7% 73.2% 90.7%
0.000796 1.465 93.8% 76.6% 93.5%
0.000987 1.506 92.3% 79.4% 94.8%
0.001243 1.521 90.4% 81.8% 95.4%
0.001406 1.499 89.1% 84.0% 95.9%
0.000446 2.877 97.9% 57.3% 81.7%
0.001116 2.912 94.8% 62.5% 87.7%
0.001485 2.940 93.0% 66.8% 90.7%
0.002012 2.830 90.6% 70.9% 92.4%
0.002288 2.761 89.3% 73.7% 93.5%
0.000137 0.750 96.8% 88.4% 94.4%
0.000230 0.820 94.7% 88.8% 95.9%
0.000308 0.917 92.8% 89.4% 96.3%
0.000422 0.951 90.2% 89.9% 96.6%
0.000437 0.994 89.8% 90.6% 96.9%
0.000527 1.541 95.9% 78.5% 88.7%
0.000715 1.605 94.5% 80.6% 92.5%
0.000949 1.670 92.6% 82.1% 94.0%
0.001207 1.637 90.6% 84.1% 95.0%
0.001333 1.664 89.7% 85.3% 95.6%
0.000353 2.958 98.3% 65.9% 78.9%
0.001009 2.972 95.3% 69.4% 86.6%
0.001374 3.009 93.6% 72.0% 89.9%
0.001873 2.976 91.2% 74.7% 91.8%
0.002144 2.958 89.9% 76.9% 93.1%
Table C-4 – Clark pump analysis results data (ordered to match Table C-2)

233
C.5 Graphs and discussion
High Pressure (bar) Delta Pressure (bar)
60

40

20

0.25
Feed Flow (L/s)
0.2
0.15
0.1
0.05
0

0.003
Leakages (L/s)
0.002

0.001

4
Pressure Losses (bar)
3

Figure C-1 – Clark pump leakages and pressure losses

In Figure C-1, the top two charts (high pressure, delta pressure and feed flow) show the
input conditions of the tests, taken from Table C-2. The bottom two charts show the
resulting leakages and pressure losses, taken from Table C-4.

Inspection suggests that the leakages are dependent mainly on the product of the high-
pressure and the feed flow. The expectation had been that the leakages would increase
with the sum of pressure and flow components. However, regression calculations (see
later) confirm that the product is indeed the dominant term. The delta pressure (up to 4
bar) has little effect on leakages.

234
Inspection of the bottom chart, in Figure C-1, suggests the pressure losses are dependent
on the feed flow and the feed flow squared. This was anticipated, since this “pressure”
loss includes frictional and viscous effects. Also, a relationship to delta pressure is
apparent.

High Pressure (bar) Delta Pressure (bar)

60

40

20

0.25
Feed Flow (L/s)
0.2
0.15
0.1
0.05
0
100% Volumetric Efficiency

95%
90%
85%
100% Mechanical Efficiency
80%
60%
40%
100% Overall Efficiency

90%

80%

Figure C-2 – Clark pump efficiencies

In Figure C-2, the top two charts are the same as those in Figure C-1 and represent the
input conditions, taken from Table C-2. The bottom the charts show the resulting
efficiencies, taken from Table C-4.

NB: volumetric efficiency × mechanical efficiency ≠ overall efficiency

235
With rotary pumps (those converting mechanical shaft power to hydraulic power) and
with hydraulic motors (those converting power the other way), it is common to define the
volumetric efficiency and the mechanical efficiency such that the product of the two
equals the overall efficiency. This is natural and desirable in systems in which the
components are all in series (for example an inverter, driving a motor, driving a pump)
since it maintains the condition that the system efficiency equals the product of the
component efficiencies. However, in systems with branches and loops (such as a reverse
osmosis system with brine-stream energy recovery), this condition cannot be achieved. It
is therefore not essential that the product of volumetric and mechanical efficiencies
should equal the overall efficiency. Certainly, this is not achieved by the efficiency
definitions set out for the Clark pump in Table C-3, as is clearly illustrated in Figure C-2.

From Figure C-2, it is apparent that the volumetric efficiency is very high throughout and
that the overall efficiency is dominated by the mechanical efficiency (pressure and
frictional losses).

C.6 Modelling
In order to construct a mathematical model of the Clark pump, for implementation in
Simulink, expressions were sought for the leakages QL and the pressure losses PL, in
terms of the input variables: feed flow QF, high pressure PH and delta pressure ∆P. The
eight formulas shown in Table C-5 were considered.

236
 Normalised
Formula under consideration
Error
QL = 7.68 × 10 - 3 × QF + - 1.51× 10 - 5 37 %

QL = 6.80 × 10 - 3 × QF + 2.74 × 10 - 5 × PH + - 1.02 × 10 -3 19 %

QL = - 2.33 × 10 -3 × QF + - 2.69 × 10 - 6 × PH + 2.17 × 10 - 4 × QF PH + 2.18 × 10 - 4 8%

QL = 1.78 × 10 - 4 × QF PH + 1.56 × 10 - 5 10 %

PL = 12.7 × QF + 0.0399 11 %

PL = 1.83 × QF + 42.2 × QF2 + 0.570 8%

PL = 1.43 × QF + 43.9 × QF2 + 7.06 × 10 - 2 × ∆P + 0.454 3%

PL = 49.2 × QF2 + 7.09 × 10 - 2 × ∆P + 0.528 3%

Table C-5 – Alternative formula considered for modelling leakages QL and pressure
losses PL

For each formula in Table C-5, the coefficients were determined by multiple linear
regression (least squares). Each formula was then tested with the original data set by
comparing its predictions against the original measured values. The errors, for each
formula, were then averaged and normalised using:

RMS (errors)
Normalised error =
RMS (original values)

where RMS is root-mean-square. The errors (residuals) were also examined at each
stage, in order to identify what components should be added to the formula for
consideration.

The fourth and eighth formula in Table C-5:

QL = 1.78 × 10 - 4 × QF PH + 1.56 × 10 - 5

PL = 49.2 × QF2 + 7.09 × 10 - 2 × ∆P + 0.528

237
in which, the units are litres/second and bar, were considered to offer a good balance of
precision and simplicity, and were used in the Simulink model of the Clark pump,
presented in section 5.2.3.

C.7 Testing of the Simulink model

The competed Simulink model of the Clark pump was presented in section 5.2.3. It was
tested by using the feed flow, high pressure and concentrate pressure data collected
during the original hardware testing (Table C-2) as inputs to the model. The output data
from the model was then compared against the original test results: product flow and feed
pressure, as shown in Table C-6.

Comparing Table C-6 against Table C-2 illustrates, as hoped, that the model closely
matches the measured data. The normalised errors for the product flow and feed pressure
are 0.95% and 0.85% respectively.

238
Modelled Feed Pressure Error Modelled Product Flow Error
(bar) (bar) (L/s) (L/s)
0.0042 0.0000 2.750 0.153
0.0040 0.0000 3.733 0.078
0.0040 0.0000 4.743 0.010
0.0039 0.0000 5.769 -0.052
0.0038 0.0000 6.797 -0.098
0.0083 0.0001 3.026 0.060
0.0082 0.0002 4.056 0.042
0.0080 0.0001 5.055 0.005
0.0078 0.0001 6.075 0.008
0.0077 0.0001 7.119 -0.034
0.0124 0.0000 3.449 -0.051
0.0122 0.0001 4.475 -0.058
0.0120 0.0001 5.508 -0.072
0.0118 0.0002 6.504 -0.058
0.0115 0.0000 7.504 -0.028
0.0165 -0.0002 4.155 -0.026
0.0162 0.0000 5.176 -0.044
0.0160 0.0001 6.166 -0.007
0.0156 0.0002 7.216 0.053
0.0153 0.0001 8.205 0.073
0.0206 -0.0005 4.879 -0.091
0.0202 -0.0001 5.938 -0.114
0.0198 -0.0001 6.929 -0.107
0.0194 0.0001 7.936 0.009
0.0191 0.0000 8.944 0.049
0.0042 0.0001 4.689 0.160
0.0040 0.0000 5.677 0.075
0.0040 0.0000 6.793 0.014
0.0039 0.0000 7.629 -0.032
0.0039 0.0001 8.755 -0.044
0.0124 0.0001 5.409 0.062
0.0122 0.0001 6.362 0.001
0.0119 0.0000 7.368 -0.041
0.0117 0.0000 8.381 -0.055
0.0115 0.0000 9.402 -0.034
0.0206 -0.0003 6.810 -0.016
0.0203 0.0000 7.820 -0.041
0.0198 0.0000 8.854 -0.082
0.0195 0.0002 9.858 0.030
0.0191 0.0001 10.689 0.089
0.0041 -0.0001 6.702 0.156
0.0041 0.0000 7.514 0.074
0.0039 -0.0001 8.701 -0.011
0.0039 0.0000 9.473 -0.060
0.0038 -0.0001 10.525 -0.103
0.0124 0.0000 7.312 0.070
0.0122 0.0000 8.357 0.008
0.0120 0.0001 9.373 -0.058
0.0118 0.0001 10.379 -0.025
0.0115 -0.0001 11.384 -0.054
0.0206 -0.0004 8.802 0.043
0.0202 -0.0001 9.832 0.036
0.0198 -0.0002 10.853 -0.001
0.0194 -0.0001 11.886 0.031
0.0191 -0.0001 12.933 0.048
Table C-6 – Simulink model output data and errors (ordered to match Table C-2)

239
Appendix D Netzsch 021 Moineau pump datasheet

240
 Appendix E Excerpt from (Thomson et al. 2001)

E.1 Cost of water over the lifetime of the equipment

A maintenance programme was devised in order to estimate the cost per cubic metre of
product water over the expected lifetime of the equipment. A 20-year life was chosen to
reflect the expectancy of the PV array. This figure is a little pessimistic for a high-quality
installation, but reflects the possibility of harsh environmental conditions in a tropical
maritime climate. The major points of the maintenance schedule are shown in Table E-1.

Frequency Action
Every day • Inspect installation for signs of malfunction
Every month • Clean PV array
Every 3 months • Dismantle RO modules and clean membranes
• Perform general maintenance
Every year • Replace RO membranes
• Replace inlet filters
• Maintain pumps and couplings
Every 5 years • Replace Clark pump
Every 10 years • Replace Moineau pump
Table E-1 – Outline maintenance schedule

The dominant yearly cost is the replacement of the membranes. It is possible that they
will not need to be replaced so frequently, but as the effect of the proposed cyclic mode
of operation has not been characterised, caution has been applied.

The average annual expenditure on replacement parts and consumables is estimated at

£1643. The annual labour costs are much lower, £293 based on an hourly wage of £3.
This is probably higher than expected in developing countries but the difference in cost
between this and a more likely rate is small, and allows some contingency.

241
The Net Present Value (NPV) of the maintenance costs is calculated using an 8%
discount rate. The calculations of the cost per cubic metre of product are show below.

Life cycle costs based on 20 years of operation

Annual product volume (m3) 1,424
3
Total amount of water produced (m ) 28,480

Initial hardware cost £23,055

Estimated shipping and installation costs £5,500
NPV of maintenance costs £28,446

NPV of Total investment £57,001

Cost per cubic metre £2.00

Table E-2 – Calculation of cost per cubic metre

242