Design of a Distribution Network Scale Model for

Monitoring Drinking Water Quality

Joost van Summeren, Ph.D. 1; Sidney Meijering 2; Hendrik Beverloo 3;
and Peter van Thienen, Ph.D. 4
Downloaded from ascelibrary.org by University of California, San Diego on 06/26/17. Copyright ASCE. For personal use only; all rights reserved.

Abstract: To investigate transport of drinking water in topologically complex distribution networks and associated monitoring problems,
experimental facilities can provide a relevant complement to numerical models and living laboratoriess or playgrounds. An experimental
facility was designed which is a scale model of an existing supply zone serving ∼100,000 customers. The facility was designed for
(1) investigating transport of water qualities and potential contaminations; (2) providing a fast and flexible testing ground for networks
of real-time water quality and quantity sensors prior to installment in live distribution networks; and (3) training and demonstration purposes.
A scale analysis was used to investigate the possibilities including in a single facility the processes turbulence, turbulent diffusion, particle
transport mode, and shear stress in a way that represents the real-life network environment, and to determine the facility’s spatial and temporal
dimensions. Using hydraulic modeling, it was verified that the main flow patterns in the facility are representative for the real-life network.
The final design includes a 4 × 8 m model, transparent PVC pipes, and a decentralized fast control system that includes 28 demand and
4 supply locations to simulate complex flow patterns. The setup is currently being prepared for experimentation. This paper describes the
theoretical background and the design and construction process. DOI: 10.1061/(ASCE)WR.1943-5452.0000799. © 2017 American Society
of Civil Engineers.

Introduction water to customers, the possibilities of controlling the environment

are limited because regulations and boundary conditions have to be
Even though numerical models have become an important tool in respected in order to prevent adverse effects on the quality, safety,
the prediction of processes in drinking water distribution networks or continuity of the distributed water. The Vitens drinking water
and the understanding of their functioning, either on their own or in company has therefore decided to build a scale model to facilitate
conjunction with an additional system such as a sensor network, research on the dynamics of different water qualities and water-
analogue models on the laboratory scale of (parts of) drinking water carried substances through a network and their detection by a net-
distribution systems (DWDSs) continue to be a valuable compli- work of sensors, and for demonstration purposes.
ment. Mostly, laboratory models are built to investigate individual This paper describes the approach to building a more generically
processes for which the understanding is incomplete or to test novel applicable laboratory model network, combining scaling laws to
technologies in a controlled environment. Some examples are leak- identify dimensional ranges in which several processes can take
age detection and management (Meniconi et al. 2011), microbiol- place in a representative way at the same time. This approach is
ogy (Douterelo et al. 2013), or water quality sensors. Their designs applied to the design and construction of a scale model of the dis-
generally make them very suitable for but also limit their applica- tribution network of the city of Leeuwarden, Netherlands, and
tion to the process for which they were conceived. surroundings.
In order to circumvent these problems and/or create lifelike
testing facilities, some water companies and universities have des-
ignated parts of their networks as playgrounds or living laborato- Method
ries, in which new technologies can be validated in a live system.
Examples of such playgrounds are the Sunrise site in Lille, France,
General Methodology
and the Vitens Innovation Playground (VIP) in the Netherlands (de
Graaf et al. 2012). However, because these systems supply drinking Design strategies of previously constructed experimental facilities
vary, depending on the purpose of the facility. Facilities aimed
1
Scientific Researcher, KWR Watercycle Research Institute, at microbiological regrowth should provide for sufficiently
Groningenhaven 7, 3433 PE, Nieuwegein, Netherlands (corresponding long experimental times, control of nutrients in the water, and ap-
author). E-mail: [email protected] propriate sensors to simulate and investigate regrowth processes
2
Technical Expert, KWR Watercycle Research Institute, Groningenhaven representative of real networks (e.g., Douterelo et al. 2013). An
7, 3433 PE, Nieuwegein, Netherlands. experimental setup to examine leakage localization and real-time
3
Research Assistant, KWR Watercycle Research Institute, Groningenhaven control for a smart water network solution in a scaled distribution
7, 3433 PE, Nieuwegein, Netherlands. network requires highly precise and accurate flow and pressure
4
Senior Scientific Researcher, KWR Watercycle Research Institute, measurements (Gúnther et al. 2014, 2015). A design for a scaled
Groningenhaven 7, 3433 PE, Nieuwegein, Netherlands.
Note. This manuscript was submitted on February 8, 2016; approved on
facility to test wireless control and sensor networks incorporates an
March 7, 2017; published online on June 26, 2017. Discussion period open advanced hydraulic control system deployed over a wireless sensor
until November 26, 2017; separate discussions must be submitted for or actuator network composed of hydraulic sensors, motorized
individual papers. This paper is part of the Journal of Water Resources valves, and ad hoc wireless communication nodes (Tejada et al.
Planning and Management, © ASCE, ISSN 0733-9496. 2015).

© ASCE 04017051-1 J. Water Resour. Plann. Manage.

J. Water Resour. Plann. Manage., 2017, 143(9): 04017051

 To design an experimental facility aimed at investigating ∂c ∂c ∂2c
network distribution and monitoring problems, special attention þv ¼ Ddiff 2 ð1Þ
∂t ∂x ∂x
was paid to achieve (1) hydraulic transport processes that are
representative of those in real-life pipes; (2) hydraulic flow patterns with c = species concentration; v = characteristic velocity; x =
that are representative of a complex network topology and demand characteristic length scale; and Ddiff = turbulent diffusion coeffi-
patterns; and (3) a transparent and reconfigurable design suitable cient. Pipe diameter was used as the characteristic length scale
for fast and flexible experimentation. These steps were followed: for diffusion because diffusion is governed by eddies, whose size
1. Identify hydraulic processes relevant to transport of drinking is constrained by the pipe diameter. The equation can thus be re-
water in a distribution network; written in nondimensional form by using the scaling relations in
2. Derive scaling relations to determine the sensitivity of these pro- Table 1 to get
cesses to pipe length, pipe diameter, and time;

∂c 0 0 ∂c
0
t0 ∂ 2 c 0
3. Determine which processes can be combined in a single facility þv ¼ Ddiff 2 ð2Þ
Downloaded from ascelibrary.org by University of California, San Diego on 06/26/17. Copyright ASCE. For personal use only; all rights reserved.

and determine the facility’s dimensions and experimental time ∂t 0 ∂x 0 d0 ∂x 02

scales;
where the term between parentheses is a dimensionless scaling
4. Design the experimental facility constrained by given practical
term. The turbulent diffusion coefficient Ddiff is empirically deter-
restrictions;
mined as (Taylor 1954)
5. Determine the degree of network topological complexity;
6. Calculate patterns of water demand in the facility; Ddiff ¼ 5.05dus ð3Þ
7. Develop and test a control system to simulate fluctuating water
demand patterns; with shear stress velocity
8. Compare flow patterns in the facility to those in the real-life qffiffiffiffiffiffiffiffiffiffiffi
network using EPANET hydraulic modeling; and us ¼ v 0 Cf =8 ð4Þ
9. Construct the facility.
with v0 = scale parameter for the cross-sectional average velocity;
Scaling Relations of Hydraulic Transport Processes and Cf = Darcy–Weisbach friction factor that can be determined
by iteratively solving the empirical Colebrook–White relation
To investigate how well processes in the scale model represent real- (Colebrook 1939). Because Ddiff is proportional to d0 l0 =t0 , the
life situations, relevant hydraulic processes were identified and a scaling term for diffusion becomes l0 =d0 (Table 2).
scale analysis was used to determine dependencies of these proc-
esses on pipe length, pipe diameter, and time. For this purpose, the Shear Stress
governing equations were nondimensionalized and described in Particle motion is initiated when the bed shear stress exceeds a
terms of the scale parameters, using the scaling relations in Table 1. critical value
This allows for theoretical assessment of the behavior of physical
τ b0 ≥ τ c0 ð5Þ
processes in the scale model, for comparison with real-life dimen-
sions. The influence of turbophoresis has been neglected because it
The dimensionless critical shear stress τ c0 depends on the size
is mostly limited to small particles in transport mains, the Saffman
and shape of grains and of the friction factor Cf , which has
lift force which may enhance turbophoresis, and the Magnus force
been empirically determined as 0.03–0.06 (Einstein 1950). The
which is not expected to be significant in drinking water distribu-
bed shear stress (using the approximation for wide channels) is
tion systems (van Thienen et al. 2011).
expressed as
Turbulence and Turbulent Diffusion Cf ρf v2
The Reynolds number R determines the flow regime, which τb ¼ ð6Þ
changes from laminar to increasing degrees of turbulence with in- 8
creasing R (Reynolds 1883). Higher degrees of turbulence promote Because the friction factor Cf is dimensionless and ρf is scale-
the mixing of suspended particles and resuspension of settled independent, it is assumed that the shear stress scales with v20 ¼
particles and decrease pipe wall shear stress (Colebrook 1939). ðl0 =t0 Þ2 (Table 2).
In turbulent flows, lateral and radial mixing by means of turbulent
diffusion and turbulent dispersion will decrease concentration gra- Particle Transport Mode
dients. Molecular diffusion is negligible in comparison. Assuming Particulate material in a fluid can either accumulate and remain
plug flow, radial velocity gradients were neglected and therefore the stagnant on the pipe wall, mobilize on the pipe wall by bed-load
influence of turbulent dispersion was also neglected and only the transport (through rolling, sliding, and saltation of particles), or re-
influence of turbulent diffusion was considered. The advection- main in suspension or solution. A particle’s mode of transport
diffusion equation then becomes is described by the Rouse number P. Transport mode transitions

Table 1. Scaling Relations and Dimensionless Quantities

Parameter Scale parameter Dimensionless parameter Unit Quantity
d d0 d0 m Pipe diameter
l l0 l0 m Pipe length
v v0 v0 ms−1 Characteristic fluid velocity
t t0 ¼ v0 =d0 t0 s Characteristic time scale
c c0 c0 mol m−3 Characteristic species concentration (cross-sectional average)
τb τ 0 ¼ ðρgr − ρf ÞgDgr τ b0 Pa Shear stress (Shields criterion)
Note: Dgr = characteristic grain diameter; g = gravitational acceleration; ρgr and ρf = densities of the grains and the fluid, respectively.

© ASCE 04017051-2 J. Water Resour. Plann. Manage.

J. Water Resour. Plann. Manage., 2017, 143(9): 04017051

 Table 2. Scale Values of Relevant Distribution Processes
Parameter Dimensionless formulation Scaling term Process Tolerance margin factor
dν l0 d0
Reynolds number R¼ Turbulence 10
v t0
t0 l0
Diffusion scale parameter Ddiff 2 Turbulent diffusion 10
d0 d0
l 2
Cf ρf0 v 02 0
Shear stress τ0 ¼ Particle accumulation and resuspension 5
8 t0
w t
Rouse number P¼ s p0ffiffiffiffiffi Mode of particle transport 5
κus l0 d0
Downloaded from ascelibrary.org by University of California, San Diego on 06/26/17. Copyright ASCE. For personal use only; all rights reserved.

from stable to bed load to partial suspension (50% of grains in sus- parameters of the real-life network Srli . Although l0 , d0 , and t0
pension) to full suspension (100%) occur at values of the dimen- of the experimental facility generally differ for that of the real-life
sionless Rouse number P ¼ 7.5, 2.5, and 1.2, respectively. The network, a process i will behave equivalent to the real-life environ-
Rouse number is the ratio of downward (gravitational) and upward ment if the scale values are equal (Sefi ¼ Srli ). Because it was not
(turbulent) particle motion possible to design a facility with Sef;i ¼ Srl;i for all processes si-
ws multaneously, tolerance margin factors fm;i were defined for each
P¼ ð7Þ process with the purpose to design a single facility with dimensions
κus
such that 1=f mi ≤ Sefi =Srli ≤ f mi for all processes.
with κ = von Kármán number; us = shear stress velocity [Eq. (4)]; The margin factors have been determined for each process indi-
and ws = settling velocity of grains (Ferguson and Church 2004) vidually on the basis of their influence on the process. For turbu-
rffiffiffiffiffi lence, a relatively large tolerance margin (factor of 10) was used
g0 R 0 g 0 dgr
02
because the character of flow does not change much above the tran-
ws ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð8Þ
d0 C ν 0 þ 0.75C R 0 g 0 d 03 sition from laminar to turbulent flow. For turbulent diffusion, a
1 2 gr
large tolerance margin (factor of 10) was also used because the in-
fluence of diffusion on concentration gradients is limited for typical
with R = specific gravity for a submerged particle (e.g., 1.65
flow conditions. Because the mode of transport is relatively sensi-
for quartz in water); dgr = characteristic grain size; and con-
tive to variations in the Rouse number, a relatively small margin
stants C1 ≈ 18 and C2 ≈ 1 (Ferguson and Church 2004).
factor of 5 was chosen. The influence of shear stress on particle
Combining Eqs. (8) and (4) and p assuming
ffiffiffiffiffi gravity is scale-
accumulation and removal is not readily determined, and therefore
independent, the scaling term t0 =ðl0 d0 Þ is used for the Rouse
a relatively small tolerance (factor 5) was used.
number (Table 2).
In the scale analysis, many combinations of d0 , l0 , and t0 were
Flow investigated. Based on the scale analysis results and practical con-
The volume of water flowing through a pipe is not taken into siderations, the following dimensions were chosen for the model
account in the scale analysis. Nevertheless, a scaling relation is use- network: d0ef ¼ 50 mm, l0ef ¼ 26 m, and t0ef ¼ 96 s. (96 s in-
ful for design purposes, such as choosing pumps, flow meters, or stead of 100 s was chosen for practical convenience: an experi-
the volume of tanks. Flow through a pipe of uniform properties is mental time of 1 s corresponds to 900 s or 15 min, and the
expressed as authors verified that this does not change the conclusions from
the scale analysis.) An important practical advantage of this model
Q ¼ Av ¼ Q0 d 02 v 0 ð9Þ is its feasible size of 18 × 7 m, which can be further reduced
to a floor area of less than 8 × 4 m with some readjustments of
with Q0 ¼ ðπv0 d20 Þ=4. For plug flow, v0 equals the (approximately
pipes. For each of the four processes, Fig. 1 shows tabulated
constant) pipe velocity.
Sef =Srl values at various d0 and t0 values. The pipe length was
also varied, but for reasons of conciseness only the results for a
Results fixed length of l0ef ¼ 26 m are shown. The shaded table entries
represent facilities in which scale factors are within the tolerance
ratios (1=f mi ≤ Sefi =Srli ≤ fmi , Fig. 1). For each combination of
Scale Analysis dimensions, the number of successful processes were added
Scale parameter values were calculated for the processes turbu- and the summed values are listed in Fig. 2. For the preferred
lence, particle transport mode, shear stress, and turbulent diffusion, model (boxed field in Fig. 2), a total of three processes are
which depend on the pipe length l0 , pipe diameter d0, and time t0 within the tolerance bounds: turbulence, transport mode, and
(as described in “Scaling Relations of Hydraulic Transport shear stress.
Processes” and Table 2). For each process i, scale parameter values To keep the spatial dimensions of the facility feasible, a conces-
Srli were calculated for the real-life distribution network, using sion was made to the process of turbulent diffusion, which will be
characteristic network dimensions l0rl ¼ 22 km, d0rl ¼ 500 mm, overestimated in the experimental facility relative to the real-life
and t0rl ¼ 1 day. These dimensions serve as a reference in the scale network. The consequences were quantified using the analytical
analysis and are based on the approximate maximum pipe length expression of van Genuchten and Alves (1982), and it was calcu-
from production to customer, typical pipe diameter, and travel time lated that the influence of diffusion in the scale model is overesti-
assuming a velocity of 0.25 m=s. Scale parameters Sefi were then mated by a factor of ∼7 when considering a fluid moving through a
calculated for (imaginary) experimental facilities by varying its 500-mm-diameter pipe at 0.25 m=s. In real life, an initially sharp
dimensions l0ef , d0ef , and t0ef for comparison with the scale concentration front becomes progressively more diffuse, such that

© ASCE 04017051-3 J. Water Resour. Plann. Manage.

J. Water Resour. Plann. Manage., 2017, 143(9): 04017051

Downloaded from ascelibrary.org by University of California, San Diego on 06/26/17. Copyright ASCE. For personal use only; all rights reserved.

(a) (b)

(c) (d)

Fig. 1. Comparison of scale parameter ratios of experimental facility to real-life distribution network (Sef =Srl ) for (a) turbulence; (b) particle transport
mode; (c) shear stress; (d) turbulent diffusion

Fig. 2. Sum of processes for which scale parameters of the experimental facility deviate from the real-life network by less than the tolerance margin
factor

after 1 h the front has traveled 4 km and spread over 40 m [meas- • The use of shorter pipes (l0 ≤ 26 m) is not recommended because
uring the distance between 10 and 90% of the concentration of the of the corresponding undesirable short experimental times (t0 ≤
incoming species, Fig. 3(a)]. For the scale model, the front smears 100 s to simulate 1 day), even larger pipe diameter-to-length
more than ∼30 cm when it has traveled a distance of 4 m in 4 s ratios, and even larger overestimation of turbulent diffusion.
[representative of 300 m over 4 km, Fig. 3(b)]. This overestimation • Making concessions to other processes than turbulent diffusion
of the turbulent diffusion is difficult to avoid in experimental facili- is problematic. For example, a design with a very low R number,
ties of the desired dimensions. In Fig. 3, relative length scales and a will cause abundant laminar flow, where in real life turbulence is
fixed moving coordinate of the concentration front center was used expected and this will consequently result in decreased resus-
for mutual comparison. pension and increased accumulation of particulate material,
Other design choices are possible, but are considered less different mixing behavior, and an increase of the effective wall
straightforward: resistance. Furthermore, pipe diameters become very small (less
• The analysis demonstrates that meeting all four conditions than 20 mm), which causes practical challenges such as the con-
(including turbulent diffusion) would require an impractical nection of sensors and precise measurement with magnetic
floor area (l0 ≥ 178 m). valves of flow rates <10−2 m3 =h.

© ASCE 04017051-4 J. Water Resour. Plann. Manage.

J. Water Resour. Plann. Manage., 2017, 143(9): 04017051

Downloaded from ascelibrary.org by University of California, San Diego on 06/26/17. Copyright ASCE. For personal use only; all rights reserved.

(a) (b)

Fig. 3. Analytical calculation of turbulent diffusion for (a) the real-life network; (b) the experimental facility; curves represent a widening con-
centration front at equidistant steps in time

Design of the Experimental Facility 2. Network pipes:

a. Transparent PVC pipes were used, which allow for visual
The Vitens Innovation Playground is a living laboratory environ-
inspection of transport of different types of (colored) water
ment in the Province of Friesland, The Netherlands, operated by the
during experimentation. An advantage of PVC is its small
Vitens water company. The 2,270-km distribution network serves
pipe roughness, and hence its weak dependency of the
∼100,000 customers in a 750-km2 area. The part that was modeled
friction factor and the degree of turbulence on pipe dia-
encompasses the province capital of Leeuwarden and includes im-
meter. A clear disadvantage of transparent pipes is algae
portant transport and distribution mains. The choice for the subset
formation, although short experimentation times, cleaning
of pipes was based on expert judgment on obtaining an optimum actions, and the possibility to replace pipes can limit the
between practical feasibility and a hydraulically representative consequences.
scale model. It is not practical (or necessary for this purpose) to b. Diameters are limited by available sizes that are
include all existing pipes and topological complexity in the scale manufactured.
model. The authors therefore started by including pipes of largest c. The construction allows for future reconfigurations and
diameter and then moved to subsequently smaller-diameter pipes. replacements of demand nodes, pipes, and sensors.
The smallest pipes represent a real-life diameter of 300 mm (but not 3. Control, demands, and production:
all 300-mm pipes are included). Because only the largest pipes are a. Twenty-eight time-dependent demand nodes were used in the
incorporated, the scale model represents only ∼15% of the total facility for control, which allows for realistic modeling of
pipe length, but as much as ∼85% of the total pipe volume. Those real-life hydraulic patterns and provides sufficient freedom
pipes with a transmission function are well represented in the scale in investigating various demand patterns.
model, but pipes with a distribution function are not. The hydraul- b. To assign realistic demands on the 28 demand nodes, the real-
ics of the scale model was tested after this initial design and con- life network was subdivided into 28 neighborhoods, and the
sidered successful in representing the real-life hydraulics, so that corresponding demands in each neighborhood were aggre-
further iteration on network complexity was not considered neces- gated, which were then scaled and assigned to each demand
sary. A computer-aided design (CAD) was constructed for the node.
experimental facility, shown in Fig. 4. The following choices c. Water delivery is controlled by four pumps attached to each
were made: tank. Each pump produces at a constant pressure, constant
1. Network layout: flow, or variable flow pattern.
a. Based on the scale analysis, pipe lengths and diameters are d. Reservoirs are up to 2 m3 to guarantee delivery of the equiva-
respectively scaled to 1.18 × 10−3 and 0.1 times the real-life lent of 1 week of maximum demand.
dimensions. e. The combination of short experimental times (1 day is simu-
b. The degree of network complexity (Fig. 4) is a compromise lated in 96 s) and the use of irregular time-dependent demand
between practical feasibility and maximum network patterns requires fast and reliable control of low-volume
complexity. water demands. To meet this requirement, a programmable
c. Most network junctions are at 45° because of practical con- logic controller (PLC) was designed to assign set-point values
siderations. to the demand nodes. On each demand node a combination of
d. Relative distances of pipes were kept as close as possible to a fast-control magnetic valve, flow meter, and controller with
the real-life situation because distances affect travel times. proportional-integral (PI) calculation allows fast convergence
e. To save space, some of the long transport mains in the east to the set-point values (Fig. 5).
and south of the area were twisted. This limits the construc- f. Sixty manual valves were included to allow modeling of
tion area to 7.3 × 3.4 m. different hydraulic scenarios.

© ASCE 04017051-5 J. Water Resour. Plann. Manage.

J. Water Resour. Plann. Manage., 2017, 143(9): 04017051

 plc

FT
plc plc

FT FT

plc plc
plc
FT FT

plc
FT FT Flow transmitter
FT
plc

plc

FT plc
FT
FIT Flow indicator transmitter
plc
FT
plc plc
FT
FT FT
plc

FT
plc
plc plc
FT
FT FT

plc

FT
Hand valve
plc

FT plc

FT
Demand unit
plc

FT

FIT
plc plc
Magnetic valve
Downloaded from ascelibrary.org by University of California, San Diego on 06/26/17. Copyright ASCE. For personal use only; all rights reserved.

drain

plc FT FT

32mm-50mm
plc

FT
plc

FT
plc 63mm-90mm
FT

plc

plc FT

plc
FIT
FT

plc

FT

Reservoir
FIT
Reservoir Reservoir 2m3
FIT
1m3 1m3 Reservoir plc

plc
0.25m3 Control unit

Fig. 4. CAD drawing of the experimental facility that includes 28 water demands, four tanks and pumps, and 60 manual valves

Fig. 5. Comparison of set-point (smaller squares) and measured (larger squares) flow at a demand node: vertical lines show 1-s intervals of experi-
mental time; the demand pattern shows typical daily variations for a period of 14.5 h; demand peaks in the morning and evening are between
experimental times 15:24:43 and 15:25:00 and 15:25:28 and 15:25:34, respectively

Hydraulic Modeling are representative of other model times. This suggests that the
degree of network complexity and number of demand nodes in-
To assess possible consequences of network topology simplifica-
tions on network hydraulics, a hydraulic model of the experimental corporated in the design is sufficient to correctly model complex
facility was made and the results were compared with those of distribution patterns. Nevertheless, because of the necessary sim-
the real-life VIP network model. Results for both models at plifications in topology and demand in the experimental facility,
8:00 a.m. (Fig. 6) demonstrate that the two models have very small local deviations in flow patterns from real-life network can-
similar global flow patterns and it was verified that these results not be avoided:

© ASCE 04017051-6 J. Water Resour. Plann. Manage.

J. Water Resour. Plann. Manage., 2017, 143(9): 04017051

Downloaded from ascelibrary.org by University of California, San Diego on 06/26/17. Copyright ASCE. For personal use only; all rights reserved.

Fig. 6. Flow patterns for the (a) real-life network; (b) scale model; arrows show locations of likely local flow deviations from the real-life model; ovals
indicate possible occurrences of deviations in flow regime (black) and mixing style (gray)

• In the scale model, the flow southwest of the tank is models of the network (Fig. 7). According to the scaling
slightly higher than in the real-life model [cf. black arrows in relation v0 ¼ l0 =t0 , velocities should be 1.06 times higher after
Figs. 6(a and b)]. This is probably due to overestimated demand downscaling of the model. Indeed, velocity magnitudes closely
at nearby nodes. follow this scaling relation. Furthermore, the models show similar
• The division of flows between the two transport mains from the temporal variations in velocity and relative velocity magnitude
eastern reservoir is more symmetric in the scale model than in ratios between the four mains. This demonstrates that flow
the real-life model [cf. lighter arrows in Figs. 6(a and b)]. This is patterns are only weakly sensitive to the scaling of network
probably due to the demand on the most northern of the two dimensions.
mains [black oval in Fig. 6(a)], which is absent in the model In the model for the experimental velocity, a uniform pipe
for the facility and/or a smaller variation in diameters in the roughness of 0.1 mm was used to simulate the use of PVC in the
scale model than in real life. entire facility. This differs from the variable roughnesses in
• Dashed black ovals indicate regions where laminar flow can the real-life model. A decrease of the uniform roughnesses by a
occur in the scaled model, where turbulent flow is expected factor of 10 results in only a small increase (up to ∼10%) in flow
in the real-life network. velocities for the model of the experimental facility. This dem-
• Gray ovals indicate regions where incomplete mixing can occur onstrates that the choice of roughness has only a minor influence
due to closely-spaced T-junction proximity or, in one case, an on the flow patterns and the pressure drop due to wall resistance
X-crossing. is small both in real life and in the experimental facility.
To compare the effect of scaling of a network, flow velocities Due to the scaling to smaller dimensions, the Reynolds numbers
were compared on four transport mains for unscaled and scaled in the experimental facility are smaller by a factor of ∼10 than in the

© ASCE 04017051-7 J. Water Resour. Plann. Manage.

J. Water Resour. Plann. Manage., 2017, 143(9): 04017051

Downloaded from ascelibrary.org by University of California, San Diego on 06/26/17. Copyright ASCE. For personal use only; all rights reserved.

Fig. 7. Flow velocities of the four transport mains of the distribution network for (a) the unscaled (but topologically simplified) network; (b) the
scaled model; for both models a pipe roughness of 0.1 mm is used

real-life network. It is possible that in some parts of the network, Discussion

the flow regime changes from turbulent to laminar. Regions where
this is most likely to occur have been identified and are indicated in Main Objective, Possibilities, and Limitations
Fig. 6 (black ovals).
In the real-life model, distribution pipes facilitate a (limited) The designed experimental setup will offer a research environ-
amount of water transport, in particular at times of low demand. ment to investigate transport processes under controlled drinking
Distribution pipes are mostly absent in the scale model, which pos- water distribution conditions. The use of fast demand actuators
sibly results in an overestimated flow though the transport mains. It allows for fast experimentation: a model time of 1 day can be
was verified that the absence of distribution mains has only a small simulated in only several minutes. Furthermore, the experiments
influence: at two locations a difference in flow rate through trans- are not constrained by customer-related regulations on safety and
port mains of only a few percent was calculated. water delivery. As such, the setup provides a flexible and fast
Transparent PVC pipes are manufactured for only a limited testing ground for novel (real-time) water quality sensor networks
range of diameters. This causes a discrepancy compared with and monitoring and numerical modeling techniques, prior to
the pipe-diameter layout of the real-life model. Flow patterns of application in living laboratories or real-life networks. Further-
scale models were compared using the true versus the available more, the facility will be useful for demonstration and training
set of pipe diameters. This comparison shows that using a lim- purposes.
ited range of diameters has only a small influence on flow The main objective of the facility is to monitor water quality
patterns. dynamics of drinking water in a topologically complex distribution
In the experimental facility, mixing at junctions is biased toward network. To enable realistic modeling of turbulence, transport
incomplete mixing because of the smaller than actual length- mode, and shear stress, the facility’s dimensions (length, diameter,
to-diameter scale of pipes. By assessing T-junction proximity, lo- and time) were carefully chosen based on a scale analysis. Conces-
cations in the facility have been identified where underestimated sions cannot be avoided, however, and a choice was made to over-
mixing rates most likely affect local species concentrations. These estimate turbulent diffusion and quantified consequences of this
are indicated in Fig. 6 (gray ovals). overestimation.

© ASCE 04017051-8 J. Water Resour. Plann. Manage.

J. Water Resour. Plann. Manage., 2017, 143(9): 04017051

 Some other objectives are not suitable to combine in this facility.
Microbiological regrowth on pipe walls and in the bulk water is a
potential threat to customers and an active field of research but the
design of the facility is not suitable for microbiological research.
The surface-to-volume ratio of pipes influences the regrowth on
pipe walls and in the bulk water and this ratio is overestimated in
the scaled network compared with real life. Also, the setup was not
designed to study possible interaction between biofilms and accu-
mulation and resuspension of particulate material because the
hydraulic time scale is faster in the scaled model than in real life,
but it is difficult to speed up biological processes in a controlled
manner. Existing test facilities designed with realistic pipe dimen-
Downloaded from ascelibrary.org by University of California, San Diego on 06/26/17. Copyright ASCE. For personal use only; all rights reserved.

sions and experimental times of at least several weeks are better fit
for microbiological research (e.g., Douterelo et al. 2013).
Networkwide transport processes are influenced by local
(incomplete) mixing of water flows at double T-junctions (N- or
U-junctions) and X-crossings of pipes (although X-crossings are
not common in the supply zone). Mixing depends on the connect-
ing pipe length (in the case of double Ts) and R numbers of the Fig. 8. Construction of the experimental facility in progress as of
October 2015
incoming flows in a complex way, and to describe the mixing
behavior several scaling laws have been proposed that are based
on empirical results (Choi et al. 2008), computational fluid dynam-
ics modeling results, or a combination of both (Shao et al. 2014). Hydraulic Problems on the Household Level
A targeted study of mixing at junctions cannot be combined in the Although the experimental setup was designed to investigate an
scaled experimental facility because the real-life pipe length-to- ∼100,000 customer supply zone, it is also possible to investigate
diameter ratio is not maintained. By assessing T-junction proximity, on the scale of a neighborhood. Assuming different characteristics
locations in the facility were identified where mixing approxima- of the real-life network (e.g., 300-m instead of 22-km maximum
tions will likely affect local species concentrations. Mixing at junc- path length, 100-mm instead of 500-mm pipe diameter, and 1-h
tions is a local phenomenon that does not require a complex instead of 24-h travel time), it is possible to model the transport
network topology and is better studied in a facility with true pipe processes within the tolerance levels (Table 2) using scaling factors
dimensions. l0ef ¼ 26 m, d0ef ¼ 50 mm, and t0ef ¼ 200 s. This way, it is pos-
sible to investigate the influence of stochastic demand patterns on,
e.g., drinking water residence times. It is also possible to investigate
Outlook on Experiments
the accumulation potential of particulate material or the self-
The experimental facility is currently being prepared (Fig. 8) for cleaning capacity for various hydraulic circumstances and network
conducting various types of experiments that are listed in the fol- topologies (e.g., looped versus branched) that are known to influ-
lowing subsections. ence the discoloration risk (Vreeburg and Boxall 2007). A limita-
tion is that the network topology and installed pipe diameters of the
Transport of Water from Multiple Sources experimental setup are representative of a supply zone instead of a
The experimental setup will allow for investigation of transport, neighborhood.
residence times, and mixing zone dynamics of water from different
production sources. The facility can be used to investigate the
influence of customers’ demand patterns, production failure, Conclusions
leakages, and section isolation due to, e.g., pipe bursts, cleaning
actions, or repairs. A scale model was designed for investigation of drinking water
quality in a topologically complex distribution system. The main
Transport of Potential Contaminations conclusions from the design process are the following:
The facility can be used to study the transport and consequences • A scale analysis serves as a useful start of the design process.
of possible chemical or microbiological contaminations. Using It enables identification of the processes that can be included in
marker species, it will be possible to vary contamination source a single facility and corresponding spatial and time dimensions
locations, origin times, contamination durations and concentra- of the facility. The analysis showed that it is possible to include
tions, water demand patterns, network layout, and valve statuses, the processes’ turbulence, particle transport mode, and shear
and study the influence on spreading of contaminations. stress in a way that represents the real-life network environ-
ment, but turbulent diffusion will be overestimated. Microbio-
Testing Sensor Networks for Event Detection logical processes are not representative for the real-life situation
The setup provides a testing ground for real-life water quality due to a different time scale and surface-to-volume ratio of
sensor networks that can warn against potentially harmful events. the pipes.
The performance of sensor networks of various configurations • Scale factors based on this analysis for pipe length (1∶846), pipe
(e.g., number of sensors, sensor location, sensitivity threshold) diameter (1∶10), and time (1∶900) result in an installation of
can be tested for one or more detection objectives (e.g., contamina- practical dimensions (3.4 × 7.3 m2 ), which allows for fast ex-
tion source identification, monitoring efficiency, shortest detection perimentation (simulation of 1 day in 96 s).
time). For simulated contamination or pipe burst event scenarios, • To avoid unnecessary topological complexity but to capture
the event detection performance can be studied for various the transmission function and main flow patterns of the real-
hydraulic patterns. life system, only mains of large diameter (larger than 300 mm)

© ASCE 04017051-9 J. Water Resour. Plann. Manage.

J. Water Resour. Plann. Manage., 2017, 143(9): 04017051

 were included in the experimental facility. Numerical modeling an experimental drinking water distribution system.” Water Res.,
results demonstrate that the main water flow patterns in the 47(2), 503–516.
facility will be similar in strength and direction to the real-life Einstein, H. A. (1950). “The bed-load function for sediment transportation
distribution network. in open channel flows.” Technical Bulletin No. 1026, U.S. Dept. of
Agriculture Soil Conservation Service, Washington, DC.
• Fast and precise, decentralized control of water demand allows
EPANET [Computer software]. U.S. Environmental Protection Agency,
for fast and flexible experimentation that enables dynamic trans- Washington, DC.
port and mixing of water from different sources and transport of Ferguson, R. I., and Church, M. (2004). “A simple universal equation for
possible water quality anomalies and can be used as a testing grain settling velocity.” J. Sediment. Res., 74(6), 933–937.
ground for networks of real-time water quality sensors. Gúnther, M., Camhy, D., Steffelbauer, D., Neumayer, M., and Fuchs-
The experimental facility is currently being prepared for Hanusch, D. (2014). “Showcasing a smart water network based
research, testing, and demonstration purposes. on an experimental water distribution system.” Procedia Eng., 89,
311–317.
Downloaded from ascelibrary.org by University of California, San Diego on 06/26/17. Copyright ASCE. For personal use only; all rights reserved.

Gúnther, M., Steffelbauer, D., Neumayer, M., and Fuchs-Hanusch, D.

Acknowledgments (2015). “Experimental setup to examine leakage outflow in a scaled
water distribution network.” Procedia Eng., 119, 450–457.
The authors thank Bendert de Graaf (Vitens), Eelco Trietsch Meniconi, S., Brunone, B., Ferrante, M., and Massari, C. (2011). “Small
(Vitens), Marije Yszenga (Vitens), Joby Boxall (University of amplitude sharp pressure waves to diagnose pipe systems.” Water
Sheffield), and Stewart Husband (University of Sheffield) for Resour. Manage., 25(1), 79–96.
Reynolds, O. (1883). “An experimental investigation of the circumstances
fruitful discussions and helpful advice on the design of the facility.
which determine whether the motion of water shall be direct or sinuous,
Wim Hijnen (KWR) and Paul van der Wielen (KWR) are thanked and of the law of resistance in parallel channels.” Phil. Trans. R. Soc.
for helpful discussions on microbiology. This work was funded Lond., 174, 935–982.
by the EU SP 7 SmartWater4Europe Grant. The authors greatly Shao, Y., Yang, Y. J., Jiang, L., Yu, T., and Shen, C. (2014). “Experimental
acknowledge thoughtful and constructive reviews from two anon- testing and modeling analysis of solute mixing at water distribution pipe
ymous reviewers. junctions.” Water Res., 56, 133–147.
Taylor, G. I. (1954). “The dispersion of matter in turbulent flow through a
pipe.” Proc. R. Soc. A, 223(1155), 446–468.
References Tejada, A., Horváth, K., Shiromoto, H. S., and Bosman, H. (2015).
“Towards WaterLab: A test facility for new cyber-physical technologies
Choi, C., Shen, J., and Austin, R. (2008). “Development of a comprehen- in water distribution networks.” Proc., 1st ACM Int. Workshop on
sive solute mixing model (AZRED) for double-tee, cross, and wye Cyber-Physical Systems for Smart Water Networks, Association for
junctions.” Water Distribution Systems Analysis 2008, ASCE, Reston, Computing Machinery, New York, 9:1–9:4.
VA, 1–10. van Genuchten, M. T., and Alves, W. J. (1982). “Analytical solutions of
Colebrook, C. F. (1939). “Turbulent flow in pipes, with particular reference the on-dimensional convective-dispersive solute transport equation.”
to the transition region between smooth and rough pipe laws.” J. Inst. Technical Bulletin No. 1661, U.S. Dept. of Agricultural Research
Civil Eng., 122(3), 218–221. Service, Washington, DC.
de Graaf, B. R., et al. (2012). “Implementation of an innovative sensor tech- van Thienen, P., Vreeburg, J. H. G., and Blokker, E. J. M. (2011). “Radial
nology for effective online water quality monitoring in the distribution transport processes as a precursor to particle deposition in drinking
network.” Water Pract. Technol., 7(4), 1–12. water distribution systems.” Water Res., 45(4), 1807–1817.
Douterelo, I., Sharpe, R. L., and Boxall, J. B. (2013). “Influence of hy- Vreeburg, J., and Boxall, J. (2007). “Discolouration in drinking water
draulic regimes on bacterial community structure and composition in systems: A review.” Water Res., 41(3), 519–529.

© ASCE 04017051-10 J. Water Resour. Plann. Manage.

J. Water Resour. Plann. Manage., 2017, 143(9): 04017051