DENMARK F. L.Smidth & Co. A/S Vigerslev All 77 DK-2500 Valby Copenhagen UNITED KINGDOM F. L.Smidth & Co.

, Ltd. 17, Lansdowne Road Croydon, GB-CR9 2JT FRANCE F. L.Smidth & Cie-France 55, Rue Ampre F-75017 Paris FEDERAL REPUBLIC OF GERMANY F. L.Smidth & Co. GmbH Geniner Strasse 133-135 D-2400 Lbeck ITALY F. L.Smidth & C. Italians, S.r.l. Via Lucilio, 25 1-00136 Rome SPAIN F. L.Smidth & Ca. Espaola, S.A. Calle de Jorge Juan, 46 Madrid 1 CANADA F. L. Smidth & Co. of Canada, Ltd. 1500 De Maisonneuve Blvd. West Montreal, Quebec, H3G 1N1 U.S.A. F. L.Smidth & Co. 300 Knickerbocker Road Cresskill, N. J. 07626 MEXICO F. L. Smidth & Ca., S.A. de C.V. Ave. Campos Eliseos, 385, 5 piso Torre B, Polanco Mexico City 5, D.F. BRAZIL F. L.Smidth, S.A., Comrcio e Indstrie Rua Nebraska, 459, Brooklin Paulista - CEP-04560 Caixa Postal 3506 01000 So Paulo -SP JAPAN F. L.Smidth & Co. (Japan) Ltd. Togeki Building 1-1, Tsukiji 4-chome, Chuo-ku, Kyobashi P.O. Box 78, Tokyo 104-91

DESCRIPTION OF RAW MEAL BLENDING IN THE CONTINUOUSLY OPERATING CF-SILO

by ib Finn Petersen and Annette Worre Jrgensen

Description of raw meal blending in the continuously operating CF-silo

Ib Finn Petersen and Annette Worre Jrgensen
F L Smidth & Co. A/S, Copenhagen
Summary The blending process of a continuously operating silo is fundamentally a transformation of compositional variations in the feed to the silo into reduced compositional variations in the discharge from the silo. The residence time distribution within the silo is shown to be the cause of this transformation. The operational principle of the so called CF-silo (Controlled Flow) is to extract raw meal at different rates from several outlets in the silo bottom and mix these streams together. An estimation procedure for the residence time distribution in the CF-silo is evaluated on the basis of flow pattern observations in small-scale experiments together with general physical principles. This procedure forms the basis of a simulation programme for the CFsilo blending process. Satisfactory agreement is found when results from this simulation programme are compared with observations corresponding to actual operational periods for CF-silos. Introduction A kiln feed with uniform chemical composition and fineness is important in order to obtain good operational conditions for the kiln and to achieve optimum cement quality on the basis of existing raw materials. In any case, an intermediate stock of raw meal is necessary in order to ensure continuous operation of the kiln and to allow for the routine maintenance stops of the raw mill. Today the trend is clearly towards the application of continuously operating silo systems for blending and storage of raw meal in cement plants. Continuous blending of raw meal Raw meal from the raw mill can be supplied to a continuously operating silo simultaneously with the extraction of kiln feed. In such through-flow systems, homogenization implies that compositional variations in the inlet flow to the silo are transformed into reduced compositional variations in the exit flow. It is common to evaluate the blending thus achieved in terms of homogenizing effect H. This evaluation involves sampling under defined conditions a number of samples from the silo inlet and exit, respectively. Based on these samples the homogenizing effect H is calculated as the ratio between the standard deviation in one chemical parameter (for instance % CaO, % CaCO3 or % LSF) in the feed to the silo and the standard deviation in the same chemical parameter in the discharge from the silo. [1] Sin,c Sin2 -- San 2 H= = Sout,c Sout2 -- San 2 The index c indicates that H is calculated from the measured standard deviations San and Sout respectively corrected for the analytical accuracy, San - This standard deviation S corresponding to a single deteran mination, can be evaluated from the results (X , xi2) of N double determinations including sample preparation for analysis:
M

San =

(Xi1 - Xi2)2

[2]

Obviously, it is only meaningful to evaluate the homogenizing effect H by expression [1] in cases where the measured standard deviation Sout is greater than San . It is generally acknowledged that when the true inhomogeneity S2out -- S2an of the kiln feed is less than 1 % LSF corresponding to 0.2% CaCO3 or 0.1% CaO, no further improvements with respect to kiln operational stability, run factor or cement quality can be achieved through additional homogenization. Evaluation of the homogenizing effect is a fairly simple method of characterizing the blending achieved in a continuously operating raw meal silo. However, the homogenizing effect is not only influenced by the silo design but also by the silo contents, the throughput rate and the character of the compositional variations in the feed to the silo. Consequently, it is of interest to consider in greater detail how blending is fundamentally affected in such a silo under stationary conditions. From a chemical engineering point of view the general treatment by Levenspiel1 concerning the non ideal flow through a vessel is relevant in this connection. Obviously no blending is achieved in a silo of the through-flow type if all material leaving the silo has spent the same time within the silo (plugflow). Blending requires that different parts of the input flow take different routes through the silo and thereby spend different times passing through the silo. The distribution of these times t is called the exit age distribution E(t) or simply the residence time distribution. It is convenient to define the residence time distribution E(t) as the normalized distribution:

E(t)d t=1

[3]

With this representation E(t) d t is the fraction of the exit flow which has spent a residence time between t and t+ d t within the silo. The residence time distribution corresponds to the concentration of tracer as a function of time which could be measured in the exit 3

Inlet concentration

to be all values smaller than t, i.e., 0 < < t. Consequently Cout(t), the outlet concentration at time t, becomes the average of all these contributions: Cout ( t ) =

C ( t _ t) . E(t)dt
0 in

[4]

Cin (t- T)

(t- T)

Time

Fraction of flow

Any kind of compositional variations in the inlet flow to a continuously operating silo under stationary conditions can, by means of the so-called convolution intergral [4], be transformed into corresponding compositional variations in the outlet flow provided the residence time distribution within the silo is known. Because residence time distribution and blending are fundamentally interrelated for a continuously operating silo, the major challenge in the design of such a silo is to establish a suitable residence time distribution for the flow of material through the silo. A large part of preferably the whole silo contents must be kept moving and different elements of the feed must pass through the silo at different rates in order to obtain a high degree of dispersion for the residence time distribution. In the design of the CF-silo (Controlled Flow) these principles have been taken into account. Description of the CF-silo The CF-silo is a continuously operating blending and storage silo for raw meal. The operating principle of this silo is to extract raw meal at different flow rates from several outlets in the silo bottom and mix these streams together. The silo is laid out as a cylindrical building with a raised bottom as shown in Figure 2. Raw meal is continuously supplied to the inlet in the centre of the silo top. Positioned in the silo top deck is the standard equipment including manhole, over- and underpressure valves and level indicators.

E (T) x dT

Residence time

Outlet concentration

Cout (t)

Cin (t- T) x E (T) x dT

Time Figure 1. Illustration of the transformation of inlet composition into outlet composition.

flow after a momentary addition of one unit tracer to the inlet flow at time zero. The residence time distribution within a silo is fundamentally responsible for the transformation of compositional variations in the inlet flow to reduced compositional variations in the exit flow. In the following, this transformation is considered in greater detail under stationary conditions. Let us focus attention on the outlet at time t (see Figure 1). All material which entered the silo at time (t - and spent the residence time T within the silo will ) appear in the outlet at time t. Let Cin (t represent ) the inlet concentration at time (t - and E( is the ) )dt fraction of the outlet flow which has spent the residence time within the silo. Accordingly the material which entered the silo at time ( - T) will give the contribution Cin(t - E( drto the outlet concentration at ) ) time t. In principle the residence time can be assumed 4

Figure 2. The CF-silo.

Figure 3. Flow patterns in model experiment.

The silo bottom is divided into seven identical hexagonal sectors each of which has an extraction outlet in the centre covered by a large pressure relief cone made of steel. Each of the hexagonal sectors is subdivided into six triangular segments. Accordingly the silo bottom consists, in total, of 42 such segments all equipped with porous aeration boxes. Three segments can be aerated simultaneously and independently. The aeration air is supplied from three rotary blowers and distributed by means of 42 solenoid valves, one for each segment. Each of the seven extraction outlets is equipped with a flap valve by which extraction from a given outlet can be started and stopped. From these outlets the material is conveyed by air slides to a centrally placed mixing tank below the silo. Within the mixing tank raw meal extracted from the different outlets is mixed together by strong aeration. The mixing tank is placed on load cells and the weighing signal from these cells starts and stops the total extraction process from the outlets in the silo bottom, so the material level inside the tank is kept within predetermined limits. This weighing signal can also be applied in the loss-ofweight system, where the mixing tank is used as kiln feeding installation2. Typically, seven different extraction rates from the silo bottom are realized in practice by positioning throttle plates with different openings in the outlets. Extraction from the outlets is carried out according to a sequence where three segments positioned at three different outlets are aerated at the same time and the flap valves of these outlets opened. The extraction sequence is cyclic in such a way that typically within 12 minutes, corresponding to the contents of kiln feed in the mixing tank every one of the 42 segments in the silo bottom is aerated once, together with opening of the connecting flap valves. The CF-silo is equipped with a PC unit (Programmable Controller) which controls the extraction sequence. This sequence is interrupted when the contents in the mixing tank reaches maximum level and the programme continues from the point of interruption when the contents in the mixing tank reaches minimum level.

results are given in Table 1 in terms of standard deviations with respect to CaO concentration. In addition the standard deviation at the silo inlet prior to this test period is included together with the fraction (Qt) of kiln feed extracted from the different outlets. Considerable blending occurs during the flow of material through the main silo volume. All standard deviations measured at the outlets in the silo bottom are substantially lower than at the silo inlet. The function of the mixing tank can also be elucidated from the results in Table 1. The contents of kiln feed in the mixing tank corresponding to 12 minutes of consumption is small in comparison with the variation period (4-8 hours) of typical compositional variations in the feed to the silo. It is therefore to be expected that the function of the mixing tank is simply to mix the material from the different outlets together without additional blending. Provided the compositional variations are random in nature the standard deviation of the composite Scomp when seven individual streams having the individual standard deviations Si (i = 1,2,....7) are mixed together in the ratios Qi (i = 1,2,....7) can be calculated as: Scomp = (Qi .Si )
i=1 7
2

[5]

Scomp = 0.07% CaO is obtained when this formula is applied to the data in Table 1. This result is in close agreement with the measured standard deviation after the mixing tank Sout = 0.08% CaO and in agreement with the concept that no additional blending takes place in the mixing tank. The mixing tank simply mixes the different streams together. The blending occurring as material flows through the main CF-silo volume can be illustrated by means of a simple model experiment. Here the flow pattern is formed in layers of sand of contrasting colour in a transparent half-cylinder equipped with outlets covered by cones. Figure 3 shows the flow pattern in such an
Table 1. Test results.
Standard deviation at silo inlet Sin = 1.97% CaO Outlet no. i 1
2 3 4 5 6 7

Blending in the CF-silo

In an attempt to understand how blending occurs in the CF-silo it is relevant to consider the results obtained in a special test. During this test samples were taken every 30 minutes from all seven outlets in the silo bottom and after the mixing tank. The whole sampling consisted of 25 samples from each sampling spot. The

Fraction of extraction Q i 0.210 0.178 0.154 0.133 0.119 0.108 0.098 Standard deviation Si% CaO 0.24 0.09 0.12 0.18 0.06 0.21 0.10 Standard deviation after mixing tank Sout = 0.08% CaO

experiment. The size of the three outlets close to the model front increase from left to right. During the experiment the degree of filling in the model silo is kept within narrow limits through the addition of extra layers of sand. The dominating features in the flow patterns shown in Figure 3 are the formation of rather broad flow zones above the cones. Within these flow zones predominating material movement is vertical. The rate of this vertical movement has its maximum in the axis of such a flow zone and decreases with the radial distance from this axis. In consequence of this velocity profile within a single flow zone, a residence time distribution will exist for the material leaving a single outlet. This observation from the model experiment can at least qualitatively account for the considerable reduction in standard deviation between the CF-silo inlet and the outlets in the silo bottom which is documented in Table 1. The key question now is, of course, whether this relationship can be quantified. Modelling the blending process in the CF-silo In an attempt to model the blending process in the CFsilo on the basis of the observations concerning flow pattern in the model experiment two problems must be solved: 1. A procedure for the calculation of the magnitude of the flow zones formed above the cones must be established 2. A functional relationship concerning the relative velocity profile of the vertical movement within a flow zone must be established. If an approximate solution can be found to these problems, it is possible to estimate the residence time distribution within a CF-silo. The first problem was solved primarily on the basis of the work of Giunta3 who dealt with the boundaries of flow zones formed above an uncovered outlet in a flat-bottomed bin. These boundaries are influenced by outlet diameter, material level within the bin and the flowability of the material characterized by the effective angle of friction of the material. However Giunta3 did not consider the case where the outlet was covered by a cone. Consequently, a correction was needed in order to account for the cones covering the outlets in the CF-silo bottom. This correction was made on the basis of the fundamental work of Jenike4 and Johanson5 with respect to gravity flow of bulk solids and with special reference to the work of Johanson6 concerning the use of flow-corrective inserts in bins. For the second problem concerning the relative velocity profile within a flow zone a solution analogous with fluid flow was applied. The parabolic velocity profile in Poiseulle flow of a Newtonian fluid in a tube is well known: v(r) = 2 v

obtained from expression [6] by a consideration where the tube is divided into differential cylindrical shells. E() =

2 0 for < 0.5 2 3 for 0.5

[7]

Here T is the mean residence time in the tube. It appears from expression [7] that in Poiseuille flow no fluid will leave the tube with a residence time less than half the mean residence time. With the solutions described above it is possible to estimate the residence time distribution of a CF-silo. This estimation involves in principle the following steps: 1. Calculation of the boundaries of the flow zones above the seven outlet cones. Consideration is given to silo dimensions, dimensions of the cones, filling level and the effective angle of friction of the raw meal. 2. Calculation of the tonnage of raw meal within each of the seven flow zones. Consideration is given to the bulk density of the raw meal. 3. Calculation of the mean residence time within each of the seven flow zones. This is possible as the total extraction rate of kiln feed from the silo as well as the fractions of kiln feed extracted from each of the outlets in the silo bottom are known. 4. Calculation of the residence time distribution for each flow zone by means of expression [7]. 5. Calculation of the residence time distribution for the whole CF-silo system. As mentioned above no additional blending takes place in the mixing tank. The total residence time distribution is therefore simply obtained by weighing together the residence time distributions for all the seven flow zones in accordance with the fractions of kiln feed extracted from the seven outlets. Having estimated the residence time distribution it is possible to simulate the blending in the CF-silo on the basis of actual compositional variations in the feed to the silo. In the simulation the analytical values determined in samples taken at hourly intervals from a continuous automatic sampler at the silo inlet are used. The corresponding concentration in the material extracted from the silo is calculated hour by hour by a numerical evaluation of the convolution integral [4]. Because plant operational conditions of a CF-silo are seldom ideally stationary this calculation takes into account hour by hour the actual tonnage of feed and discharge together with variations in the silo filling level. The whole simulation procedure has been programmed on a HP-1000 computer. Simulations results Results from a 3-day operational period with a 22. 4 m diameter CF-silo having a capacity of 21 0001 of raw meal is shown in Figure 4. The average silo filling in this period was 70% and the average throughput rate 260 tph. Figure 4a represents the CaO contents in the raw mill product fed to the silo measured at hourly intervals in samples from a continuous automatic sampler. Figure 4b represents the CaO contents in hourly spot samples from the silo. A simulation of this operational period has been carried out. The results of this simulation are shown in Figure 4c as the calculated CaO contents in the discharged material as a function of time. Proper results cannot be obtained in this simulation over the first 24 hours simply because it

((r) - (R-r )
2

[6]

where: v(r) = velocity in the distance rfrom the tube axis R = tube radius v = mean velocity The velocity has its maximum value 2 v in the tube axis and decreases to zero at the tube wall. Obviously velocity profile and residence time distribution are interrelated. The following residence time distribution E() corresponding to Poiseulle flow in a tube is 6

takes a certain period of time before the variations introduced at the silo inlet appear in the discharge. The standard deviation of the calculated CaO contents in the discharged material from hour 25 to 72 is 0.1 % CaO, which is in exact agreement with the measured standard deviation in the kiln feed spot samples over the same time interval. However, the mean 42.88% CaO of the simulation results is slightly lower than the mean 43.02% CaO of the kiln feed spot samples. This difference could arise either due to material moving very slowly in the silo or in consequence of the filter dust return which is included in the kiln feed spot samples but not accounted for in the simulation. From visual inspection a resemblance between the time record of the curves in Figures 4b and 4c is to be noticed. It would of course be of interest to add a shot of tracer to the silo inlet and investigate how the silo modifies this signal in comparison with the simulation model. As an alternative to such an extensive tracer test programme one could focus attention on an operational period during which a major disturbance occurred in the composition fed to a CF-silo. A 4-day operational period with a 28 m diameter CF-silo having a capacity of 35 000 t of raw meal is represented in Figure 5. During this period the average filling was 60% and the average throughput rate 260 tph. The CaO contents of the raw mill product determined in hourly samples from a continuous automatic sampler are represented in Figure 5a. A major disturbance starting at hour 32 was introduced in the feed to the silo. Especially the first part of this disturbance with a surplus of CaO will be difficult to even out due to the lack of counterbalance when this material starts to appear in the discharge from the silo. The contents of CaO in the kiln feed measured in spot samples every second hour are recorded in Figure 5b. The disturbance in the kiln feed composition starts at hour 41. A simulation of this operational period has been performed. The results of this simulation in terms of the calculated CaO contents in the extracted material as a func-

Figure 5. Time record of 4-day period of operation with 28 m diameter CF-silo together with simulation results.

tion of time are shown in Figure 5c. Once again no proper results can be obtained in the simulation over the first 24 hours. It appears from a comparison between Figures 5b and 5c that close agreement exists between simulation results and measurements concerning both magnitude and position in the time record of the disturbance in the composition of the discharge from the silo. These results demonstrate that the estimation procedure for the residence time distribution used in the simulation programme describes the actual conditions within a CFsilo fairly well. Conclusions Blending in the CF-silo is mainly affected due to the formation of rather broad flow zones above the cones covering the outlets in the silo bottom. The rate profile of the material movement down such a flow zone appears to be parabolic in nature. This rate profile causes a considerable degree of blending as material moves through. The function of the mixing tank is simply to mix the material extracted from the different outlets. This concept of the CF-silo blending process has been quantified into a mathematical model in the form of a simulation programme. Results obtained with this simulation programme corresponding to actual operational periods of CF-silos have shown satisfactory agreement with the corresponding measurements. With the flexible CFsilo blending system this simulation programme forms a relevant tool in the optimization of the raw meal blending process both at the project stage and under plant operational conditions. REFERENCES
1. LEVENSPIEL, O. Chemical Reaction Engineering. Wiley, New York, 1972, Chapter 9, 523 ft. 2. HOLMBLAD, L P and SCHOU, K. Kiln feed system based on the loss-of-weight principle, incorporated in the CF-silo. Zement-KalkGips, 7, 1983. 3. GIUNTA, J S. Flow patterns of granular materials in flat-bottom bins. Journal of Engineering for Industry, ASME, May 1969, pp 406413. 4. JENIKE, A W. Steady gravity flow of frictional-cohesive solids in converging channels. J. Appl. Mech., March 19,64, pp 5-11. 5. JOHANSON, J R. Stress and velocity fields in the gravity flow of bulk solids. J. Appl. Mech., September 1964, pp 499-506. 6. JOHANSON, J R. The use of flow-corrective inserts in bins. Journal of Engineering for Industry, May 1966, pp 224-230.

Figure 4. Time record of 3-day period of operation with 22.4 m diameter CF-silo together with simulation results.