Mechanics, Materials Science & Engineering, July 2017 – ISSN 2412-5954
Characterization, Design and Optimization of Industrial Phosphoric Acid Production Processes by Artificial Neural Network 1 Gholamhosseion Grivani1,a, Shahriyar Ghammamy2,b, Farzane Yousefi1,c, Mehdi Ghammamy3,d 1- Department of Chemistry, Faculty of Science, Damghan University, Semnan, Damghan, Iran 2- Department of Chemistry, Faculty of Science, Imam Khomeini International University, Qazvin, Iran 3- Department of Mechanical Engineering, Faculty of Engineering, Tehran University, Tehran, Iran a- grivani@du.ac.ir b- shghamami@yahoo.com c- f4380110672@gmail.com d- Mghamazi@alumni.ut.ac.ir DOI 10.2412/mmse.00.114.96 provided by Seo4U.link
Keywords: phosphoric acid, optimisation, modelling and simulation, genetic algorithm.
ABSTRACT. In this paper we optimized industrial scale phosphoric acid production processes by genetic algorithm and Artificial Neural Network. In this work an efficient method is suggested to design an optimized cast in order to increase the rate of phosphoric acid and purity percent in manufacturing process. The predicted results are in very good agreement with the experimental data with an error of less than 4.88%. The simulation model has been examined with real experimental data obtained from the Phosphate Mines. A parametric study has been made to find the optimum operating conditions of the pilot plant for a given phosphate rock. The effect of varying reactor(s) time, sulphuric acid rate, water rate, soil rate agitator–impeller speed has been investigated. More than 160 samples were made in laboratory and the results are derived by changing the parameters and using the limited component software. We also hope to gain feedback that will improve the modeling to better meet the needs of the phosphate industry. Finally, using the Pareto front extracted from optimization algorithms, optimum template selection and remodeling results have been examined with finite element software. Studies have confirmed that using the developed method is an effective tool to achieve the proper format in order to restructure and reduce the strain based on improving the convergence rate and the applied force.
Introduction. Phosphoric acid is an important intermediate chemical product. It is added to foods as a preservative, acidifying agent, flavour enhancer, and clarifying agent. Phosphoric acid is also used in processes such as the coagulation of rubber latex, electro polishing, soil stabilization, and as a catalyst in the production of propylene and butene polymers, ethylbenzene, and cumene. Eightly percent of the acid is used in the production of agricultural fertilizers. Production capacity for phosphoric acid yielded about 33 million tons of P2O5. It is mainly obtained through the attack of phosphate rock with sulphuric acid. Its quality depends greatly on the P2O5 rock content and nature of the present impurities, among which some can be recovered like uranium, rare earth, etc. . Almost all phosphoric acid needed for the fertilizer industry is produced by wet processes. In many of these processes, the raw phosphate ore is converted into phosphoric acid and calcium sulphate di-hydrate (gypsum) by adding a mixed solution of sulphuric and phosphoric acids to the reactor. genetic algorithm is one of the random optimizing methods invented in 1995 by Kennedy & Eberhart . multi objective particle swarm algorithm (MOPSO) simultaneously searches different spaces of design and optimizing points for the complex problems such as non-convex and discrete problems. In MOPSO method, selecting the best local guide (the global best particle) for each particle of the population from a set of Pareto-optimal solutions has a great impact on the convergence and diversity of 1
© 2017 The Authors. Published by Magnolithe GmbH. This is an open access article under the CC BY-NC-ND license http://creativecommons.org/licenses/by-nc-nd/4.0/
MMSE Journal. Open Access www.mmse.xyz
Mechanics, Materials Science & Engineering, July 2017 – ISSN 2412-5954
solutions, especially when optimizing problems with high number of objectives. In this paper an efficient method is suggested to design an optimized cast in order to increase the homogeneity of the microstructure material and to reduce the applied force required in the manufacturing process. In this work, we develop a design framework that is able to transparently capture the process phenomena involved regardless of the modelled process task. Reactions of phosphoric acid production display below [1-8 ]: 3Ca3(PO4)2.CaF2+ 14H2SO4 →10 Ca( H2PO4)2 +2HF↑
(1)
Ca( H2PO4)2+ H2SO4 + nH2O → 2H3PO4+ 10 CaSO4. nH2O
(2)
Chemicals and reagents In laboratory, we mixed different rate of water, soil and Sulphuric acid by different time. After several minutes, reaction completed. Table 1 shows the values of input parameters and their output objective functions. Table 1. Values of input parameters and their output objective functions. 1
400
40
210
540
425
0.6938
2
350
60
150
480
287
0.64134
3
300
60
180
540
330
0.7087
4
300
60
210
540
359
0.7584
5
450
60
90
360
174
0.6991
6
500
60
60
300
371
0.3372
7
500
60
30
300
297
0.2621
8
450
50
120
360
125
0.6964
9
500
40
120
300
407
0.4697
10
350
70
120
480
275
0.5387
11
300
80
120
540
159
0.5773
12
350
70
90
480
198
0.4653
13
300
80
60
540
88
0.3873
14
300
80
30
540
153
0.309
15
450
50
150
360
317
0.5806
16
500
40
180
300
494
0.5977
17
500
40
210
300
428
0.6583
18
450
50
150
480
370
0.54029
19
500
40
180
540
443
0.608
20
500
40
210
540
480
0.6409
21
350
70
90
360
165
0.461
22
300
80
60
300
101
0.50008
MMSE Journal. Open Access www.mmse.xyz
Mechanics, Materials Science & Engineering, July 2017 – ISSN 2412-5954
Modelling with artificial neural network Artificial neural network (ANNs) are non-linear mapping structures based on the function of the human brain. They are powerful tools for modelling, especially when the underlying data relationship is unknown. ANNs can identify and learn correlated patterns between input data sets and corresponding target values. An artificial neuron is a computational model inspired in the natural neurons. Natural neurons receive signals through synapses located on the dendrites or membrane of the neuron. When the signals received are strong enough (surpass a certain threshold), the neuron is activated and emits a signal though the axon. This signal might be sent to another synapse, and might activate other neurons. Each neuron in our brain accepts input from many other neurons and then provides a resulting output. This is precisely what we will be replicating in code. Each neuron class will have a structure where there is a body of attributes and one output [9-10].
Fig. 1. Neural network training. Considering four effective parameters in cast design includes water rate, acid rate, soil rate, and time, which have more effects on sulphuric acid rate and purity percent, in order to recognize the first objective function (sulphuric acid rate), artificial neural network is designed with four entrance neuron layers, two middle layer with 12 and 1 neurons and external layer with one neuron. Furthermore, in order to recognize the second objective function (purity percent), another neural network is designed with four entering neuron layers, middle layers, two middle layer with 14 and 1 neurons and external layer with one neuron. Figures 2 and 3 shows the Modelling of artificial neural network for first and second objective functions
MMSE Journal. Open Access www.mmse.xyz
Mechanics, Materials Science & Engineering, July 2017 – ISSN 2412-5954
Fig. 2. Modeling of artificial neural network for first objective function (sulphuric acid rate)
Fig. 3. Modeling of artificial neural network for second objective function (purity percent). The physical topology of a network refers to the configuration of cables, computers, and other peripherals. Physical topology should not be confused with logical topology which is the method used to pass information between workstations. Logical topology was discussed in the Protocol chapter. In order to train and test the neural network in order to forecast the homogeneity coefficient and amount of applied force in the cast, precise data are needed, so some models are produced by finite element software and changing the cast design parameters to diagnose the search space related to the neural network and calculate and save the objective functions. Produced data in pervious stage include corner and curving angles, radius ration, friction coefficient, strain homogeneity coefficient and applied force. Around 85 percent of the data are considered for training and others for testing the predictability of the network for new data. Squared error of the network for training is 0.0079 and in testing is 0.0096. Squared errors of the testing shows that the network is not affected by over training and maximum error of the neural network is 4.88 %. Neural network estimated the homogeneity coefficient with 95.12precision. Neural network output is compared to laboratory data in figure 5. MSEtr =0.0198 MSEts =0.0198 Ans =7
MMSE Journal. Open Access www.mmse.xyz
Mechanics, Materials Science & Engineering, July 2017 – ISSN 2412-5954
Fig. 5. (a) Squared error of the network for testing, (b) The output of the artificial neural network (ANN) model for data test. Training process of the second neural network is done same as the previous network, in order to predict the purity percent needed for the process based on the effective casting parameters. Using the extracted functions from the above neural network, we can begins the optimization process. Created functions by this neural network acts for cost functions in optimization process. Optimizing the multi-objective problems Multi-objective optimization is a class of problems with solutions that can be evaluated along two or more incomparable or conflicting objectives. Most optimization problems naturally have several objectives to be achieved (normally conflicting with each other), but in order to simplify their solution, they are treated as if they had only one (the remaining objectives are normally handled as constraints). The Multiobjective Optimization Problem (MOP) (also called multicriteria optimization, multiperformance or vector optimization problem) can be defined (in words) as the problem of finding: a vector of decision variables which satisfies constraints and optimizes a vector function whose elements represent the objective functions. These functions form a mathematical description of performance criteria which are usually in conflict with each other. Hence, the term “optimize” means finding such a solution which would give the values of all the objective functions acceptable to the decision maker. These types of problems differ from standard optimization problems in that the end result is not a single “best solution” but rather a set of alternatives, where for each member of the set, no other solution is completely better (the Pareto set). Multi-objective opti mization problems occur in many different real-world domains, such as architecture (stability vs. cost), and automobile design and as such are a very important problem domain. In solving the multi-objective problem, objective functions usually contradict with each other. It means that improving one function results in decline of the other function. so, all the functions cannot be seen in the best mode. In order to optimize all the objective functions at the same time optimized points are used. Improper points are the points that no other point is dominant over them. Proximate algorithms can find good answers (near optimization) in short time for hard optimized problems. Table 2 shows the design variables and their changes domain. This limit is considered as the limit of the problem. The homogeneity coefficient and maximum applied force are considered as object function [11-15].
MMSE Journal. Open Access www.mmse.xyz
Mechanics, Materials Science & Engineering, July 2017 – ISSN 2412-5954
Table 2. Design variables and their changes limit. Time
Sulphuric acid rate
Soil rate
Water rate
}300,540{
}30,210{
}30,80{
}300,500{
Pareto front shows the objective function toward each other. Optimized points can be considered where two objective functions are satisfied. As it was said, horizontal axis is phosphoric acid rate and vertical axis is purity percent. This chart is composed of discrete points each of which shows the optimized points for both objective functions. After repeating the optimizing algorithms, the optimal points (optimized area) are shown as the figure 6. The parts shown in figure 6 indicate a proper limit for choosing the proper optimizing point since both functions in this limit are satisfied and so several proper points are chosen form this limit. Table 3 shows the values of input parameters and their output objective functions.
Fig. 6. Desired range for the optimal point in Pareto. Table 3. Optimal point extracted from the Pareto frontunctions. Objective 1
Optimal 413 cast
Objective 2
Time(min)
sulphuric acid rate(ml)
Soil rate(g) Water rate
0.52876
360
68.2
67.5
486
Summary. Genetic Algorithms are a family of computational models inspired by evolution. These algorithms encode a potential solution to a specific problem on a simple chromosome like the one data structure information of Genetic algorithms are often viewed as function optimizers. Although the range and apply recombination operators to these structures so as to preserve critical of problems to which genetic algorithms have been applied is quite broad. During each successive generation, a portion of the existing population is selected to breed a new generation. Individual solutions are MMSE Journal. Open Access www.mmse.xyz
Mechanics, Materials Science & Engineering, July 2017 – ISSN 2412-5954
selected through a fitness-based process, where fitter solutions (as measured by a fitness function) are typically more likely to be selected. Certain selection methods rate the fitness of each solution and preferentially select the best solutions. This work proposes a way that facilitates the modelling, design and optimization of phosphoric acid production processes. As we have demonstrated modern process simulators such as genetic algorithm can be of considerable value for the design, operation and management of phosphoric acid production speed. Optimizing methods taken form nature, which are normally expressed by random qualities and start the research performance from several points. As this is ongoing work, we intend to further investigate the insights generated through the proposed designs from an industrial user perspective as well as to explore more design cases with respect to phosphogypsum utilization. Acknowledgment We gratefully acknowledge the financial support from the Research Council of Imam Khoemieni International University by Grant No, 751387-91. References [1] Papadopoulos, A.I., Seferlis, P., & Theodosiadis, K. (2007). Modeling, Design and Optimisation of Industrial Phosphoric Acid Production Processes. [2] Mohamed Azaroual, Christophe Kervevan, Arnault Lassin, Laurent André, Mohamed Amalhay, Lhachmi Khamar, Mohamed EL Guendouzi, Thermo-kinetic and Physico-Chemical Modeling of Processes Generating Scaling Problems in Phosphoric Acid and Fertilizers Production Industries, Procedia Engineering, 2012, Vol. 46, 68-75, DOI 10.1016/j.proeng.2012.09.447 [3] N. Boulkroune, A.H. Meniai, Modeling purification of phosphoric acid contaminated with cadmium by liquid-liquid extraction, Energy Procedia, 2012, Vol. 18, 1189-1198, DOI 10.1016/j.egypro.2012.05.134 [4] Ms.G. Bharathi kannamma, Dr.D. Prabhakaran, Dr.T. Kannadasan, Analysis and Simulation of Dihydrate Process for the Production of Phosphoric Acid (Reactor Section), American Journal of Engineering Research (AJER), 2013, Vol. 02, Iss. 07, 01-08 [5] John E. Cameron, Phosphoric Acid by Wet Process: Pond Water Management. [6] Samir I. Abu-Eishah,Nizar M. Abu-Jabal, Parametric study on the production of phosphoric acid by the dihydrate process, Chemical Engineering Journal 81 (2001) 231–250 [7] Gan, C., V. Limsombunchai, M. Clemes, A. Weng (2007), Consumer choice prediction : artificial neural networks versus logistic models, Lincoln University. Commerce Division. [8] Carlos A. Coello Coello, Nareli Cruz Cortes, Solving Multiobjective Optimization Problems Using an Artificial Immune System, Genet Program Evolvable Mach (2005) 6: 163. DOI 10.1007/s10710-005-6164-x [9] Mahdi Ghamami, Masoud Shariat Panahi, Maryam Rezaei, Optimization of locomotive body structures by using imperialist competitive algorithm, Journal of Computational and Applied Research in mechanical engineering (JCARME), 2014, 3(2): 105-113. [10] Rao, S.S, Engineering Optimization: Theory and Practice, Fourth Edition, 2009 by John Wiley & Sons, Inc. [11] Kennedy, J, Eberhart, R, Particle Swarm Optimization, Proceedings of IEEE International Conference on Neural Networks, Perth, Australia, 1942-1948. 1995 [12] Zitzler E., Laumanns M., Bleuler S. (2004) A Tutorial on Evolutionary Multiobjective Optimization. In: Gandibleux X., Sevaux M., Sörensen K., T’kindt V. (eds) Metaheuristics for Multiobjective Optimisation. Lecture Notes in Economics and Mathematical Systems, vol 535. Springer, Berlin, Heidelberg. MMSE Journal. Open Access www.mmse.xyz