View
3
Download
0
Category
Preview:
Citation preview
Draft
Optimizing Chute-Flip Bucket System Based on Meta-Modelling Approach
Journal: Canadian Journal of Civil Engineering
Manuscript ID cjce-2018-0534.R2
Manuscript Type: Article
Date Submitted by the Author: 27-Jun-2019
Complete List of Authors: Bananmah, Mohammad ; Shiraz University, Department of Civil and Environmental EngineeringNikoo, Mohammad Reza; Shiraz University, Civil and Environmental EngineeringNematollahi, Banafsheh; Shiraz University, Department of Civil and Environmental EngineeringSadegh, Mojtaba ; Boise State University
Keyword: Chute-Flip Bucket system, Genetic Algorithm, Artificial Neural Network, Flow-3D numerical model
Is the invited manuscript for consideration in a Special
Issue? :Not applicable (regular submission)
https://mc06.manuscriptcentral.com/cjce-pubs
Canadian Journal of Civil Engineering
Draft
1
1 Optimizing Chute-Flip Bucket System Based on Meta-Modelling Approach
2 Mohammad Bananmah1, Mohammad Reza Nikoo2*, Banafsheh Nematollahi3, Mojtaba Sadegh4
3 1 Mohammad Bananmah
4 Research Assistant, Department of Civil and Environmental Engineering, Shiraz University, Shiraz, Iran.
5 Email Address: mohammadbananmah@gmail.com
6
7 2 Mohammad Reza Nikoo (Corresponding Author)
8 Associate Professor, Department of Civil and Environmental Engineering, Shiraz University, Shiraz, Iran.
9 Email Address: nikoo@shirazu.ac.ir
10 Phone: +98-713-613-3497
11 Fax: +98-713-647-3161
12
13 3 Banafsheh Nematollahi
14 Research Assistant, Department of Civil and Environmental Engineering, Shiraz University, Shiraz, Iran.
15 Email Address: b.Nematollahi@shirazu.ac.ir
16
17 4 Mojtaba Sadegh
18 Assistant Professor, Department of Civil Engineering, Boise State University.
19 Email Address: mojtabasadegh@boisestate.edu
20
21 Word count: 5634
22
23
24
25
26
27
Page 1 of 36
https://mc06.manuscriptcentral.com/cjce-pubs
Canadian Journal of Civil Engineering
Draft
2
28 Abstract: Optimal design of Chute-Flip Bucket (CFB) system depends on various
29 parameters, among which energy dissipation and cavitation prevention are the most
30 important. This study develops a simulation-optimization model based on a calibrated Flow-
31 3D numerical model, Multi-Layer Perceptron Artificial Neural Network (MLP-ANN), and
32 Genetic Algorithm (GA) optimization approach for determining the optimal geometry of the
33 CFB system. To alleviate the computational time burden of the Flow-3D numerical model, a
34 MLP-ANN meta-model is developed based on some limited simulations of Flow-3D. The
35 meta-model framework is then coupled with GA to provide an efficient design framework for
36 the CFB system. The proposed framework is employed to design optimal geometry of the
37 CFB system of the Jareh dam in Ahvaz, Iran. The results show that the obtained optimal
38 design increases the cavitation index up to 30% and energy dissipation up to 32% compared
39 to the old engineering design already in place.
40
41 Key words: artificial neural network, chute-flip bucket system, Flow-3D numerical model,
42 genetic algorithm
43
44 1. Introduction
45 Energy dissipation structures are installed to alleviate scouring and erosion in the downstream
46 of the dams (Shivashankara Rao 1982; Vischer and Hager 1995; Heller et al. 2005). Flip bucket
47 is an effective dissipation structure which is installed at the end of chute spillways to decrease
48 the kinematic energy of spill water. The chute spillway and the flip bucket can be considered a
49 unified energy dissipation structure, so-called Chute-Flip Bucket (CFB) system, due to their
50 interdependent performances. Avoiding cavitation on the chute spillway and maximizing
Page 2 of 36
https://mc06.manuscriptcentral.com/cjce-pubs
Canadian Journal of Civil Engineering
Draft
3
51 energy dissipation in the system are the main considerations when optimizing the performance
52 of the CFB system (Vischer and Hager 1995; Heller et al. 2005). Several studies have focused
53 on developing and designing a range of experimentation and simulation models to satisfy the
54 design criteria for dissipation structures and spillways, including: experimental approach
55 (Falvey 1990; Heller et al. 2005; Steiner et al. 2008; Ashiq and Sattar 2010; Pfister and Hager
56 2010a; Pfister and Hager 2010b; Kumar and Sreeja 2012), numerical simulation using Flow-
57 3D numerical model (Savage and Johnson 2001; Hoo et al. 2004; Nohani et al. 2013; Parsaie
58 et al. 2018; Morovati and Eghbalzadeh 2018), and numerical simulation using Smoothed
59 Particle Hydrodynamics (SPH) approach (Ferrari 2010; González-Cao et al. 2018). Despite the
60 progress made, less attention has been paid to optimizing the geometric design of the CFB
61 system based on coupling optimization approaches with numerical simulation models. This
62 highlights the need for an integrated simulation-optimization model to obtain optimal geometry
63 of the CFB.
64 Implementation of an integrated optimal design of the chute spillway and the flip bucket is
65 challenging. In this study, a simulation-optimization model for the CFB system of the Jareh
66 dam in Ahvaz, Iran is proposed based on calibrated Flow-3D numerical model, Multi-Layer
67 Perceptron Artificial Neural Network (MLP-ANN), and GA optimization approach. The main
68 decision variables for this simulation-optimization model are bottom slopes of the chute
69 spillway, curvature radius, and deflection angle of the flip bucket (Pfister et al. 2014).
70 This paper develops the proposed simulation-optimization model in three major steps. First,
71 the flow characteristics through the CFB system are simulated using a calibrated Flow-3D
72 numerical model, which is validated based on experimental data. The calibrated Flow-3D
73 numerical model is executed for different values of curvature radius and deflection angle of the
74 flip bucket as well as various slopes of the chute spillway to obtain a comprehensive scenario
75 database. Second, a meta-model based on MLP-ANN is trained and validated with the chute
Page 3 of 36
https://mc06.manuscriptcentral.com/cjce-pubs
Canadian Journal of Civil Engineering
Draft
4
76 spillway slope, curvature radius, and deflection angle of the flip bucket as its inputs, and the
77 cavitation index and energy dissipation derived from the calibrated Flow-3D numerical model
78 as its outputs. Finally, the optimal CFB geometry is delineated by coupling MLP-ANN model
79 and GA optimization approach. To show the efficiency and applicability of the proposed
80 framework, the developed simulation-optimization model is applied to the Jareh dam CFB
81 system, which is located in Khuzestan province, Southern part of Iran. In the following,
82 methodology, case study, results, and summary and conclusion sections of this paper are
83 presented.
84
85 2.Methodology
86 The proposed simulation-optimization framework for the Chute-Flip Bucket (CFB) system
87 comprises three main stages, which are summarized in Fig. 1.
88 In the first stage (steps 1 and 2), essential data including surface roughness, water depth, flow
89 velocity, and pressure, as well as CFB geometric parameters (i.e. curvature radius and
90 deflection angle of the flip bucket, and bottom slopes of the chute spillway) are gathered, which
91 are then used for developing the numerical Flow-3D numerical model. In the second step, the
92 geometric information and hydraulic properties of the flow are utilized to calibrate the Flow-
93 3D numerical model for efficient simulation of the CFB. It should be noted that this model is
94 calibrated based on the data gathered from Khuzestan Water and Power Authority (KWPA).
95 Then, the calibrated Flow-3D numerical model is executed for different geometric scenarios of
96 the CFB.
97 In the next stage (step 3), an MLP-ANN meta-model is trained and validated using the
98 input/output database generated by the calibrated Flow-3D numerical model.
99 Finally, in stage 3 (steps 4 and 5), this meta-model is coupled with the GA optimization
100 approach to obtain the optimal geometry of the CFB system with two objectives: 1. controlling
Page 4 of 36
https://mc06.manuscriptcentral.com/cjce-pubs
Canadian Journal of Civil Engineering
Draft
5
101 the cavitation problem, and 2. maximizing the energy dissipation of the system. In the
102 following sections, the main stages of the proposed framework are described in more details.
103
104 2. 1. Numerical simulation with the Flow-3D numerical model
105 Flow-3D is a superlative software for computational fluid dynamics problems with ability to
106 simulate 1-D, 2-D and 3-D fluid dynamics (Hirt and Nichols 1981; Hirt and Sicilian 1985).
107 One of the significant features of this software used in the analysis of the hydraulic structures
108 is its ability to model open-channel flow using Volume Of Fluid (VOF) model (Hirt and
109 Nichols 1981). This numerical model is developed based on the Navier-Stokes equations (Hirt
110 and Nichols 1981; Hirt 1990). In order to calculate the volume ratio of each fluid cell,
111 momentum equations for air and flow are solved by volume-fluid model, simultaneously. In
112 each cell, the summation of the volume ratio for the air and for the water is equal to one. The
113 volume ratio of the fluid q(αq) can be defined as (Hirt and Nichols 1981):
114 The cell is devoid of fluid q. αq=0
115 The cell is full of fluid q. αq=1
116 The cell is filled by fluid q and other fluid (s). 0<αq<1
117 The Flow-3D numerical model can be developed with the assumption that CFB is a rigid body
118 with known geometry. For more details about the Flow-3D software, refer to the Flow-3D user
119 manual (Flow Science Inc. 2011). In this study, a computer model of the CFB system for the
120 Jareh dam is constructed by the Rhinoceros software with scale of 1:1 (Fig. 2). This computer
121 model provides the CFB geometric with the information needed by the Flow-3D numerical
122 model. Calibration of the Flow-3D numerical model and mesh sensitivity analysis is then
123 performed using hydraulic information from the KWPA reports. This provides a suitable mesh
124 size.
Page 5 of 36
https://mc06.manuscriptcentral.com/cjce-pubs
Canadian Journal of Civil Engineering
Draft
6
125 In the next step, different scenarios of the geometric properties corresponding to the CFB
126 system were simulated using the Flow-3D numerical model. Estimated values of the cavitation
127 index in specific checkpoints and the energy dissipation of the entire system are calculated for
128 each scenario. Cavitation occurs when the cavitation index is less than or equal to the critical
129 cavitation index. Following the common guidelines for concrete-based hydraulic structures,
130 the critical cavitation index is considered in the range of 0.2 to 0.25. Cavitation index as a
131 hydrodynamic parameter that describes the cavitation process is defined as,
132
0
212
k VP P
V
for: k=1, 2, …, np (1)
133
134 Where,
Pv Vapor pressure (kPa),
P0 Reference pressure (kPa),
V Fluid velocity (m/s),
ρ Fluid density (kg/m3),
k Checkpoint's number,
σk The cavitation index at kth checkpoint,
Np Total number of checkpoints,
135 2. 2. Multi-Layer Perceptron-Artificial Neural Network (MLP-ANN)
Page 6 of 36
https://mc06.manuscriptcentral.com/cjce-pubs
Canadian Journal of Civil Engineering
Draft
7
136 Artificial Neural Networks are intelligent models based on the neural structure of the human
137 brain. This intelligent model can be coupled with other numerical and/or optimization models
138 in engineering problems that require high computational efficiency (Schalkoff 1997). In this
139 study, Multi-layer Perceptron-Artificial Neural Network (MLP-ANN) is coupled as a meta-
140 model with GA optimization approach for efficient computational speed. MLP-ANN contains
141 feedforward networks of simple processing units with at least one “hidden” layer, wherein
142 each processing unit is similar to a perceptron with non-linear functionality (Schalkoff 1997).
143 There is neither a specific rule to determine the optimum number of hidden layers and
144 neurons, nor a definite guidance to the effective choice of transfer functions in the MLP-
145 ANN models. Therefore, aforementioned parameters are defined through a trial and error
146 process. In this study, MLP-ANN model is utilized with 3 layers (one hidden) and 7 neurons
147 in the hidden layer. Four decision variables (i.e., the bottom slopes of the chute spillway,
148 curvature radius, and deflection angle of the flip bucket) are inputs to this meta-model, and
149 energy dissipation and the cavitation index are model outputs. In order to train the meta-
150 model, the available dataset (generated by Flow3D scenario simulation) is randomly divided
151 into two groups of training and testing validation categories. It is, however, important to
152 avoid extrapolation in any Neural Network modeling. Hence, it has been checked to ensure
153 the range of validation data is covered by the train data. To delineate the procedure in
154 selecting the training and testing data, upper and lower boundaries of data as well as average
155 and their standard deviations are presented in Table 3. The values of the average and standard
156 deviation for all training and testing data are close to each other
157 Finally, to evaluate performance of the MLP-ANN model, multiple statistical error indices
158 (i.e., Bias, Scatter Index (SI), Root Mean Square Relative Error (RMSRE), Mean Absolute
159 Relative Error (MARE), and Correlation Coefficient (CC)) are calculated for training and
160 validating data. These indices are defined as,
Page 7 of 36
https://mc06.manuscriptcentral.com/cjce-pubs
Canadian Journal of Civil Engineering
Draft
8
161
*
1 1
1 1n n
i ii i
Bias w wn n
(2)
2* *
1
1 ( )n
i ii
w W w WnSI
W
(3)
2*
*1
1 100%n
i i
i i
w wRMSREn w
(4)
*
*1
1 100%n
i i
i i
w wMAREn w
(5)
*
1
2 *2
1 1
n
i ii
n n
i ii i
w wCC
w w
(6)
162
163 Where,
w Values of the calculated cavitation and energy dissipation using the Flow-3D
numerical model,
w* Values of the calculated cavitation and energy dissipation from the ANN meta-
model,
W Average of ,w
W * Average of ,*w
n Number of total data points used in the training and validation processes of ANN
164
165 2. 3. Genetic Algorithm (GA) optimization
166 Genetic Algorithm (GA) is an efficient method for solving optimization problems (Khorshidi
167 et al. 2019). Once the fitness functions of the GA approach and optimization constraints are
Page 8 of 36
https://mc06.manuscriptcentral.com/cjce-pubs
Canadian Journal of Civil Engineering
Draft
9
168 defined, this algorithm makes an iterative effort to update a population of solutions using
169 selection, cross over, and mutation operators, until convergence to a global optimum is reached
170 (Kumar 2012). In each iteration (generation) of the GA optimization, individuals are randomly
171 selected from the current population with weights according to the fitness function. Superior
172 solutions according to the fitness function and optimization constraints are more likely to be
173 chosen. The mentioned process is repeated until some specific termination criteria are satisfied
174 and the optimized solutions are achieved (Davis 1991).
175 The objective function and constraints of the developed GA-based simulation-optimization
176 model are presented in the following. In the present framework, the decision variables are
177 slopes of the chute spillway (S1 and S2), deflection angle (θ), and curvature radius (R) of the
178 flip bucket. Furthermore, the objective functions are maximizing cavitation index and
179 maximizing energy dissipation, as defined by:
Maximize 1 21
npknorm norm
kf wt wt E
(7)
0 1 2( , , , , , , , )kVL V P P R S S for: k=1, 2, …, np (8)
1 2 1 2( , , , , , , , , , )E e V R S S y y g (9)
180 Subject to:
0.2k for: k=1, 2, …, np (10)
20 40 (11)
and = fixed at specified elevation1y 2y (12)
181 Where,
wt1 Importance weight of cavitation index objective,
wt2 Importance weight of energy dissipation objective,
Page 9 of 36
https://mc06.manuscriptcentral.com/cjce-pubs
Canadian Journal of Civil Engineering
Draft
10
σknorm Normalized cavitation index in the kth checkpoint,
ΔEnorm Normalized energy dissipation,
L MLP-ANN meta-model used to predict the cavitation index,
E MLP-ANN meta-model used to estimate energy dissipation,
R Curvature radius of the flip bucket,
Θ Deflection angle of the flip bucket,
S1 The bottom slopes of the first section of the chute spillway,
S2 The bottom slopes of the second section of the chute spillway,
y1 Upstream elevation of the chute spillway,
y2 Downstream elevation of the chute spillway,
G Gravitational acceleration,
µ Fluid viscosity,
182
183 3. Case Study
184 The Jareh dam is located at 49°43' East and 31°26' North on the Zard river, 35 kilometer
185 Northeast of Ramhormoz in Khuzestan province, southern part of Iran (Fig. 3). The main
186 purpose of this earthen dam is supplying agricultural water demand. Total reservoir volume of
187 the Jareh dam in the normal water level is 180 MCM (106 m3), and it's elevation from sea level
188 is 497 meters. CFB of the Jareh dam consists of a chute spillway with 325.65 m horizontal
189 length, which is located in the elevation range of 481.07 meters to 429.4 meters and the flip
190 bucket with a curvature radius of 25 meters and deflection angle of 45.5° at the end of the chute
191 spillway in the elevation of 429.4 meters (KWPA 2009).
192
193 3. 1. Experimental procedure
194 The main governing forces in some hydraulic structures such as spillways are gravity and
Page 10 of 36
https://mc06.manuscriptcentral.com/cjce-pubs
Canadian Journal of Civil Engineering
Draft
11
195 inertia forces. To this end, the Froude similarity law can be utilized to determine model scale
196 for such hydraulic structures in the laboratory (Novak and Cabelka 1981). The scale
197 considered for modeling hydraulic structures should consider several guidelines to minimize
198 scale effects. For example, the scaling factor for a model should not be less than a specific
199 value to avoid viscosity effects. Another important limitation is created by surface tension
200 and the corresponding Weber number. In more details, the water elevation in the model
201 should not be less than a specific value, usually determined through experience (about 3 cm)
202 (Kobus 1980). This approximate minimum value is 1.5 cm to 2 cm in some references. There
203 are some other limitations in selecting the similarity law and scale factor that can be studied
204 in source references (Novak and Cabelka 1981). As mentioned earlier, the Froude number
205 similarity was chosen in this study, and after careful consideration of the limiting factors, a
206 scale of 1:50 is chosen to construct the experimental model of the Jareh dam.
207 The experimental model is constructed with transparent Plexiglas with 3 main sections:
208 1. Upstream reservoir: as a sub-section of the lake, dam body, channel connected to the main
209 spillway, and the entrance of the emergency spillway,
210 2. The components of the CFB system including gated spillway, chute and flip bucket,
211 3. Outlet tank including outlet channel and a river section.
212 The details of constructed experimental model of the Jareh dam and its CFB in the laboratory of
213 Khuzestan Water and Power Authority is shown in Figs. 4 and 5.
214
215 3. 2. Measurement point selection
216 Twenty-three points (A to W) are selected along the experimental model of the Jareh dam to
217 calculate the hydraulic parameters, including velocity and water depth, in which 11 points are
218 placed on the CFB system (Fig. 6). In this study, 8 of the 11 points were selected as the
219 checkpoints to calibrate and analyze the mesh grid used in the Flow-3D numerical model (Fig.
Page 11 of 36
https://mc06.manuscriptcentral.com/cjce-pubs
Canadian Journal of Civil Engineering
Draft
12
220 7).
221 It is notable to mention that, one of the main reasons for utilizing deflector is avoiding cavitation at
222 the end of the chute using a technique that is so-called as constant cavitation number (Bingnan et al.,
223 1982; Falvey, 1990). With this in mind, the cavitation index increased in the deflector because of
224 surplus of the flow depth and pressure. The possibility of cavitation occurrence in the deflector is very
225 little, and hence, no point is chosen on the deflector as a potential point for cavitation.
226
227 3. 3. Experimental data uncertainties
228 As mentioned, experimental data for this study are collected from the experimental model
229 constructed by Khuzestan Water and Power Authority. In the available reports presented by this
230 organization, some uncertainties corresponding to the experimental data are mentioned, which
231 are summarized as:
232 Error in designing and constructing the experimental model:
233 This error is related to the level evaluation of the different sections corresponding to a dam such
234 as dam spillway, which is reported as ± 0.025 (m), approximately.
235 Error in water level measurement:
236 This error is considered due to measuring instruments and human errors, which is reported as ±
237 0.050 (m), approximately.
238 Error in water velocity measurement:
239 This error is considered approximately about ± 0.048 (m/s) for Pitot tube in high velocity
240 region and about ± 0.35 (m/s) for Current in low velocity region.
241 Error in flow discharge measurement:
242 This error is related to the errors in constructing the model, reading the water level, and model
243 leakage, which is about ± 2.79%.
244 Experimental data uncertainties are inevitable in any experimental study, which should be
245 considered in calibration of the numerical models. Minimizing uncertainties during laboratory
Page 12 of 36
https://mc06.manuscriptcentral.com/cjce-pubs
Canadian Journal of Civil Engineering
Draft
13
246 recording, such as error in reading flow characteristics and measurement instruments, is critical
247 for successful design.
248
249 4. Results
250 First step of the proposed methodology involves developing and calibrating the Flow-3D
251 numerical model of the CFB system. The proper mesh grid is acquired and Flow-3D
252 numerical model is calibrated by comparing the hydraulic parameters, including water
253 velocity and flow depth of the experimental and numerical simulations in 8 checkpoints along
254 the CFB system. Figs. 8 and 9 demonstrate this comparison for the values of water depth and
255 flow velocity with different mesh dimensions.
256 Figs. 8 and 9 show that 1-meter network dimension can be considered appropriate, because
257 further refining the mesh dimensions beyond 1-meter considerably increases the
258 computational time with no significant accuracy gain.
259 Furthermore, the MARE statistical error index in numerical model for 1-meter mesh
260 dimension is equivalent to 7 and 3 percent for depth and velocity compared to the
261 experimental data, respectively. Therefore, this mesh dimension is acceptable given the
262 efficient computational speed and appropriate accuracy.
263 Subsequently, the cavitation index for 6 critical checkpoints (Fig. 7) was calculated to check
264 the cavitation phenomenon and evaluate the risks associated with different geometric scenarios
265 of the CFB system (Ball 1976; Falvey 1990). The energy dissipation was also calculated for
266 different geometric scenarios using Bernoulli’s equation.
267 After estimating cavitation index and energy dissipation values, a meta-model based on 3 layers
268 perceptron artificial neural network is trained and validated using several statistical error
269 indices (Eq. 2- Eq. 6). This meta-model is subsequently used for predicting the cavitation index
270 and energy dissipation of the CFB. Table 1 shows the results of different statistical error indices
Page 13 of 36
https://mc06.manuscriptcentral.com/cjce-pubs
Canadian Journal of Civil Engineering
Draft
14
271 used in the validation stage of the MLP-ANN meta-model. It should be noted that all indices
272 are dimensionless.
273 As mentioned earlier, the optimal geometry of the CFB system was obtained through a MLP-
274 ANN meta-model coupled with a GA optimization model. This simulation-optimization model
275 optimizes the objective function while satisfying the design constraints presented in section 2
276 (Eq. 7- Eq. 12). Furthermore, this coupling leads to maximizing the cavitation index and energy
277 dissipation for the CFB system by determining the optimal slopes for the chute spillway and
278 optimal curvature radius and deflection angle for the flip bucket (Bananmah 2016). In the
279 developed GA optimization framework, the number of population and generation are set to 40
280 and 100, respectively, based on a trial and error procedure. Furthermore, the optimization
281 function weights are set to 0.7 for cavitation index (wt1) and 0.3 for energy dissipation (wt2). It
282 is notable that the aforementioned weights significantly influenced to the optimal design of the
283 CFB system. The results show that the optimal slopes of the first and second sections of the
284 chute spillway are 3.27% and 23%, respectively. Moreover, the optimal curvature radius and
285 deflection angle of the flip bucket are 25.46 meters and 56.57 degrees, respectively. In addition,
286 after obtaining optimal geometric parameters, the proposed CFB system was simulated by
287 Flow-3D numerical model to check the performance of the optimal CFB system in terms of
288 cavitation and energy dissipation. This procedure shows that the proposed optimization-
289 simulation model increases the cavitation index and energy dissipation up to 30% and 32%,
290 respectively, compared to the actual constructed CFB system of the Jareh dam. Fig. 10 presents
291 comparison results of the cavitation indices of the optimal CFB system proposed in this study
292 with the actual CFB system in 6 critical checkpoints. Performance of the proposed CFB system
293 has improved in all critical checkpoints compared to the actual CFB system.
294 Moreover, in this study, the objective function is optimized using different weighting scenarios
295 (Fig. 11). With this in mind, the utilized weighting scenarios are clarified in more details in
Page 14 of 36
https://mc06.manuscriptcentral.com/cjce-pubs
Canadian Journal of Civil Engineering
Draft
15
296 Table 2. In addition, the minimum, average, and maximum values of the cavitation indices at
297 critical checkpoints along the CFB system for different weighting scenarios are shown in Fig.
298 12. This is the result of a sensitivity analysis on different weighting scenarios, and includes
299 comparison of the calculated cavitation indices and energy dissipation through the proposed
300 framework using different weights for each objective function. In other words, different
301 weighting scenarios are utilized in the proposed model to investigate the impact of each
302 weighting scenarios on the optimal decision variables and obtained objective functions.
303 Finally, the geometric parameters corresponding to the CFB system is stated in Table 2 for
304 different weights obtained from sensitivity analysis (Table 2).
305
306 4. Summary and Conclusion
307 In this study, a simulation-optimization model based on a calibrated Flow-3D numerical
308 model, Multi-Layer Perceptron Artificial Neural Network (MLP-ANN), and GA optimization
309 approach is proposed to determine the optimal geometry of the Chute-Flip Bucket (CFB)
310 system of the Jareh dam. Given the acceptable performance of the proposed methodology in
311 simulating the flow characteristics of the CFB system, it can be concluded that the
312 simulation-optimization model proposed in this study is an acceptable alternative for
313 experimental method with suitable accuracy and computational speed. Furthermore, the
314 results show that using a MLP-ANN meta-model is appropriate for predicting cavitation
315 index and energy dissipation in the CFB systems. This robust framework leads to increasing
316 cavitation index up to 30% and energy dissipation up to 32%, compared to the actual
317 constructed system already in place. The proposed framework tackles computational issues
318 by training a meta-model that can closely emulate the response of the CFB system as well as
319 the Flow-3D numerical model. Although the computational burden is decreased using the
320 proposed framework containing 4 state variables, the computational speed for a larger set of
Page 15 of 36
https://mc06.manuscriptcentral.com/cjce-pubs
Canadian Journal of Civil Engineering
Draft
16
321 decision variables is still challenging. The computational burden for modeling the CFB
322 system due to a wide variety of decision variables can be addressed in the future studies to
323 find more efficient solutions. Furthermore, this research is underway to implement the
324 developed simulation-optimization solution considering flood uncertainty using Fuzzy Set
325 theory. Moreover, the proposed framework also can be applied to other dissipation structures.
326
327 Conflict of Interest: The authors declare no conflict of interest.
328
329 References
330 Ashiq, M., and Sattar, A. 2010. Optimization of energy dissipation works for Nai Gaj dam
331 project. In World Environmental and Water Resources Congress 2010: Challenges of
332 Change, pp. 1622-1633.
333 Ball, J.W. 1976. Cavitation from surface irregularities in high velocity. ASCE Journal of the
334 Hydraulics Division, 102.
335 Bingnan, L., ZHENYING, G., and Donghai, P. 1982. A rational profile for flip buckets of
336 high dams. Science in China Series A-Mathematics, Physics, Astronomy & Technological
337 Science, 25(12): 1343-1352.
338 Bananmah, M. 2016. Hydraulic multi-objective optimization of chute-flip-bucket system in
339 dams: case study: Jareh Dam, M.Sc. thesis, Department of civil and environmental
340 engineering, Shiraz university, Shiraz, Iran.
341 Davis, L. 1991. Handbook of genetic algorithms.
342 Falvey, H.T. 1990. Cavitation in chutes and spillways. Engineering Monograph 42, Water
343 Resources Technical Publication, US Printing Office, Bureau of Reclamation, Denver.
Page 16 of 36
https://mc06.manuscriptcentral.com/cjce-pubs
Canadian Journal of Civil Engineering
Draft
17
344 Ferrari, A. 2010. SPH simulation of free surface flow over a sharp-crested weir. Advances in
345 Water Resources, 33(3): 270-276.
346 Flow Science Inc. 2011. FLOW-3D user’s manual 10.1 edition. Santa Fe, N.M., Flow Science,
347 Inc.
348 González-Cao, J., García-Feal, O., Domínguez, J.M., Crespo, A.J.C., and Gómez-Gesteira,
349 M. 2018. Analysis of the hydrological safety of dams combining two numerical tools: Iber
350 and DualSPHysics. Journal of Hydrodynamics, 30(1): 87-94.
351 Google Earth 7.3.2.5776. 2019. Jareh dam 31° 26' 32'' N, 49° 42' 22'' E, elevation 1418ft. 2D
352 map, viewed 10 October 2016.
353 Heller, V., Hager, W.H., and Minor, H.E. 2005. Ski jump hydraulics. Journal of Hydraulic
354 Engineering, 131(5): 347-355.
355 Hirt, C.W., and Nichols, B.D. 1981. Volume of Fluid (VOF) method for the dynamics of free
356 boundaries. Journal of Computational Physics, 39(1): 201-225.
357 Hirt, C.W., and Sicilian, J.M. 1985. A porosity technique for the definition of obstacles in
358 rectangular cell meshes. In 4th International Conference on Numerical Ship
359 Hydrodynamics.
360 Hoo, D., Boyes, K., Donohoo, S., and Cooper, B. 2004. Numerical flow analysis for spillways.
361 In Ancold Bulletin, Hobart, Tasmania, 24-29 October 2003, pp. 127-138.
362 Khorshidi, M.S., Nikoo, M.R., Sadegh, M., and Nematollahi, B. 2019. A multi-objective risk-
363 based game theoretic approach to reservoir operation policy in potential future drought
364 condition. Water Resources Management, 33(6): 1999-2014.
365 Khuzestan Water and Power Authority (KWPA). 2009. Final report of the hydraulic model of
366 the dissipation structure of the Jareh dam. Water Research Institute, Khuzestan, Iran.
367 Kobus, H. 1980. Hydraulic modelling. In Hydraulic Modelling, German Association for
368 Water Resources and Land Improvement.
Page 17 of 36
https://mc06.manuscriptcentral.com/cjce-pubs
Canadian Journal of Civil Engineering
Draft
18
369 Kumar, R. 2012. Novel encoding scheme in genetic algorithms for better fitness. International
370 Journal of Engineering and Advanced Technology, 1(6): 214-219.
371 Kumar, C., and Sreeja, P. 2012. Evaluation of selected equations for predicting scour at
372 downstream of ski-jump spillway using laboratory and field data. Engineering
373 Geology, 129: 98-103.
374 Morovati, K., and Eghbalzadeh, A. 2018. Study of inception point, void fraction and pressure
375 over pooled stepped spillways using Flow-3D. International Journal of Numerical Methods
376 for Heat & Fluid Flow, 28(4): 982-998.
377 Novák, P., and Cabelka, J. 1981. Models in hydraulic engineering; physical principles and
378 design applications. Pitman Advanced Publishing Program.
379 Nohani, E., Sakhaei, S., and Afrous, A. 2013. The Study of cavitation phenomenon in the
380 chute of Jareh dam by applying Fow-3D model. Bull. Env. Pharmacol. Life Science 3(1):
381 83-90
382 Parsaie, A., Moradinejad, A., and Haghiabi, A.H. 2018. Numerical modeling of flow pattern
383 in spillway approach channel. Jordan Journal of Civil Engineering, 12(1).
384 Pfister, M., and Hager, W.H. 2010a. Chute aerators. II: Hydraulic design. Journal of Hydraulic
385 Engineering, 136(6): 360-367.
386 Pfister, M., and Hager, W.H. 2010b. Chute aerators. I: Air transport characteristics. Journal of
387 Hydraulic Engineering, 136(6): 352-359.
388 Pfister, M., Hager, W.H., and Boes, R.M. 2014. Trajectories and air flow features of ski jump-
389 generated jets. Journal of Hydraulic Research, 52(3): 336-346.
390 Savage, B.M., and Johnson, M.C. 2001. Flow over ogee spillway: Physical and numerical
391 model case study. Journal of Hydraulic Engineering, 127(8): 640-649.
392 Schalkoff, R.J. 1997. Artificial neural networks. McGraw-Hill Higher Education.
Page 18 of 36
https://mc06.manuscriptcentral.com/cjce-pubs
Canadian Journal of Civil Engineering
Draft
19
393 Shivashankara Rao, K.N. 1982. Design of energy dissipators for large capacity spillways.
394 In Proc., Brazilian Committee on Large Dams, pp. 311-328.
395 Steiner, R., Heller, V., Hager, W.H., and Minor, H.E. 2008. Deflector ski jump
396 hydraulics. Journal of Hydraulic Engineering, 134(5): 562-571.
397 Vischer, D.L., and Hager, W.H. 1995. Energy dissipators, hydraulic structures design
398 manual. IAHR, Balkema, Rotterdam, Netherlands.
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
Page 19 of 36
https://mc06.manuscriptcentral.com/cjce-pubs
Canadian Journal of Civil Engineering
Draft
20
414 Figure Captions:
415 Fig. 1. Flowchart of the proposed simulation-optimization framework for optimal design of the CFB
416 Fig. 2. Geometric model of the Jareh dam constructed by Rhinoceros software with a scale of 1:1
417 Fig. 3. The plan view of the Jareh dam (Google earth 7.3.2.5776 2019)
418 Fig. 4. The experimental model of the CFB system of the Jareh dam (KWPA 2009)
419 Fig. 5. The experimental model of the Jareh dam in details (KWPA 2009)
420 Fig. 6. Measurement points on the experimental model of the Jareh dam (KWPA 2009)
421 Fig. 7. Selected critical checkpoints from cavitation point of view and measurement checkpoints for calibration
422 of the Flow-3D model on the CFB system
423 Fig. 8. Comparison of the calculated velocity by Flow-3D numerical model and the experimental data for different
424 mesh dimensions
425 Fig. 9. Comparison of the calculated water depth by Flow-3D numerical model and the experimental data for
426 different mesh dimensions
427 Fig. 10. Comparison of the cavitation indices of the proposed model and the actual model of the Jareh dam
428 Fig. 11. Sensitivity analysis for the cavitation index and energy dissipation of different weighting scenarios
429 Fig. 12. Sensitivity analysis for cavitation index of different weighting scenarios
430
431
432
433
434
435
436
437
438
Page 20 of 36
https://mc06.manuscriptcentral.com/cjce-pubs
Canadian Journal of Civil Engineering
Draft
21
439 Symbols:
440 CFB Chute-Flip Bucket
441 GA Genetic Algorithm
442 MLP-ANN Multi-Layer Perceptron Artificial Neural Network
443 SPH Smoothed Particle Hydrodynamics
444 VOF Volume Of Fluid
445 αq the volume ratio of the fluid q
446 Pv vapor pressure (kPa)
447 P0 reference pressure (kPa)
448 V fluid velocity (m/s)
449 ρ fluid density (kg/m3)
450 k checkpoint's number
451 σk the cavitation index at kth checkpoint
452 np total number of checkpoints
453 SI Scatter Index
454 RMSRE Root Mean Square Relative Error
455 MARE Mean Absolute Relative Error
456 CC Correlation Coefficient
457 w values of the calculated cavitation and energy dissipation using the
458 Flow-3D numerical model
Page 21 of 36
https://mc06.manuscriptcentral.com/cjce-pubs
Canadian Journal of Civil Engineering
Draft
22
459 w* values of the calculated cavitation and energy dissipation from the
460 ANN meta-model
461 W average of w
462 W * average of *w
463 n number of total data points used in the training and validation
464 processes of ANN
465 S1 the bottom slopes of the first section of the chute spillway
466 S2 the bottom slopes of the second section of the chute spillway
467 θ deflection angle of the flip bucket
468 R curvature radius of the flip bucket
469 wt1 importance weight of cavitation index objective
470 wt2 importance weight of energy dissipation objective
471 σknorm normalized cavitation index in kth checkpoint
472 ΔEnorm normalized energy dissipation
473 L MLP-ANN meta-model used to predict the cavitation index
474 e MLP-ANN meta-model used to estimate energy dissipation
475 µ fluid viscosity
476 y1 upstream elevation of the chute spillway
Page 22 of 36
https://mc06.manuscriptcentral.com/cjce-pubs
Canadian Journal of Civil Engineering
Draft
23
477 y2 downstream elevation of the chute spillway
478 h flow depth
479 g gravitational acceleration.
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
Page 23 of 36
https://mc06.manuscriptcentral.com/cjce-pubs
Canadian Journal of Civil Engineering
Draft
24
499 Tables:
500
501 Table 1. Statistical error indices based on the results of the MLP-ANN meta-model
Statistical error index Bias SI RMSRE MARE CCTrain -0.0082 0.0793 8.0832 6.4968 0.9971Dissipation of
energy (%) Test -0.7034 0.0441 8.8375 6.9182 0.9961Train 0.0004 0.0603 8.1404 5.0833 0.9985Cavitation
index Test -0.0002 0.0597 7.9973 5.1459 0.9983502
503 Table 2. Sensitivity analysis results for different weight scenarios
Decision variablesWeight scenario's number
1wt 2wtS1 S2 R θ
1 0.1 0.9 7 23 25.72 56.572 0.3 0.7 6.99 23 25.72 56.573 0.5 0.5 4.83 23 25.51 56.574 0.7 0.3 3.27 23 25.46 56.575 0.9 0.1 3.01 23.01 25.45 56.57
504
505 Table 3. The values of average and standard deviation of training and testing procedure
Variables
Proc
edur
e
S1 S2 R θ σ ΔE
Data ranges [3* - 7**] [20 - 50] [20 - 35] [20 - 40] [5.44 - 4.67] [22 - 43]
Average 5 35 28 30 4.87 36
Tra
in
Standard deviation 1.63 10 5.61 7.91 0.25 10.25
Data ranges [3 - 7] [20 - 50] [20 - 35] [25 - 35] [5.55 – 4.62] [28 – 41]
Average 5 35 27.5 30 4.86 37.65Tes
t
Standard deviation 1.63 10 5.59 6.83 0.23 8.54
506 * Lower bound of data507 ** Upper bound of data508
Page 24 of 36
https://mc06.manuscriptcentral.com/cjce-pubs
Canadian Journal of Civil Engineering
Draft
1
Start
Geometry parameters of the Chute-Flip Bucket (CFB) system
including slopes of the chute spillway as well as deflection angle
and curvature radius of the flip bucket
Gathering hydraulic parameters of the Chute-Flip
Bucket (CFB) system including water depth, velocity
and pressure
Creating the main model of the CFB using Rhinoceros as an input for the Flow-3D model
Mesh sensitivity analysis to reach best mesh dimension Decreasing
the mesh
dimension
Are the results
suitable compared to
experimental data?
Flow-3D numerical model is calibrated and ready to use
No
Yes
No
Yes
Determining the requirement time to reach steady state condition Mesh assignment and imposing boundary conditions
Is the flow steady?
Running the calibrated Flow-3D model for various scenarios with different decision variables to obtain the
input/output database for training and validating MLP-ANN meta-model
Training and validating the MLP-ANN meta-model to predict the flow characteristics
Calculating the cavitation index and energy dissipation for different scenarios
Determining the objective function and constraints with respect to design criteria of the CFB
Coupling developed GA optimization model with validated MLP-ANN meta-model
Choosing the best geometry of the CFB using developed simulation-optimization model
End
Step 1: Collecting data
Step 2: Numerical simulation using calibrated Flow-3D numerical
model
Step 3: Developing MLP-ANN meta-Model based on input/output results of the calibrated Flow-3D numerical
model for different scenarios
Step 4: Developing simulation-optimization model by coupling MLP-ANN meta-model to Genetic Algorithm
(GA) optimization
Step 5: Determining the optimal dimensions of the CFB
Increasing the simulation time
Page 25 of 36
https://mc06.manuscriptcentral.com/cjce-pubs
Canadian Journal of Civil Engineering
Draft
Fig. 2. Geometric model of the Jareh dam constructed by Rhinoceros software with a scale of 1:1
209x115mm (300 x 300 DPI)
Page 26 of 36
https://mc06.manuscriptcentral.com/cjce-pubs
Canadian Journal of Civil Engineering
Draft
Fig. 3. The plan view of the Jareh dam (Google earth 7.3.2.5776 2019)
Page 27 of 36
https://mc06.manuscriptcentral.com/cjce-pubs
Canadian Journal of Civil Engineering
Draft
Fig. 4. The experimental model of the CFB system of the Jareh dam (KWPA 2009)
Page 28 of 36
https://mc06.manuscriptcentral.com/cjce-pubs
Canadian Journal of Civil Engineering
Draft
Fig. 5. The experimental model of the Jareh dam in details (KWPA 2009)
Page 29 of 36
https://mc06.manuscriptcentral.com/cjce-pubs
Canadian Journal of Civil Engineering
Draft
Fig. 6. Measurement points on the experimental model of the Jareh dam (KWPA 2009)
Page 30 of 36
https://mc06.manuscriptcentral.com/cjce-pubs
Canadian Journal of Civil Engineering
Draft
Fig. 7. Selected critical checkpoints from cavitation point of view and measurement checkpoints for calibration of the Flow-3D model on the CFB system
Page 31 of 36
https://mc06.manuscriptcentral.com/cjce-pubs
Canadian Journal of Civil Engineering
Draft
Fig. 8. Comparison of the calculated velocity by Flow-3D numerical model and the experimental data for different mesh dimensions
210x159mm (300 x 300 DPI)
Page 32 of 36
https://mc06.manuscriptcentral.com/cjce-pubs
Canadian Journal of Civil Engineering
Draft
Fig. 9. Comparison of the calculated water depth by Flow-3D numerical model and the experimental data for different mesh dimensions
210x159mm (300 x 300 DPI)
Page 33 of 36
https://mc06.manuscriptcentral.com/cjce-pubs
Canadian Journal of Civil Engineering
Draft
Fig. 10. Comparison of the cavitation indices of the proposed model and the actual model of the Jareh dam
217x159mm (300 x 300 DPI)
Page 34 of 36
https://mc06.manuscriptcentral.com/cjce-pubs
Canadian Journal of Civil Engineering
Draft
Fig. 11. Sensitivity analysis for the cavitation index and energy dissipation of different weighting scenarios
217x159mm (300 x 300 DPI)
Page 35 of 36
https://mc06.manuscriptcentral.com/cjce-pubs
Canadian Journal of Civil Engineering
Draft
Fig. 12. Sensitivity analysis for cavitation index of different weighting scenarios
217x159mm (300 x 300 DPI)
Page 36 of 36
https://mc06.manuscriptcentral.com/cjce-pubs
Canadian Journal of Civil Engineering
Recommended