ABSTRACT
Convectional algorithms such as least-square and gradient descent for Adaptive Neuro-Fuzzy Inference System (ANFIS) prediction of engineering process system is deficient by local optimum trapping problem. Therefore, this study is aimed at developing novel hybrid Genetic Algorithm (GA)-ANFIS- Box- Behken (BB) model for Hevea Brasiliensis Seed Oil Epoxidation (HBSOE) prediction. Computer codes for traditional ANFIS and hybrid GA-ANFIS modelling were written in Matlab 2015 environment. Whereas BB numerical optimization technique of Response Surface Methodology (RSM) of design expert V10 software was used to select optimum GA-ANFIS tuning parameters (Population Size (PS), Crossover Percentage (COP) and (MR): Mutation Rate). Sensitivity and uncertainty analyses on GA-ANFIS-BB model output (MSE) were investigated using Monte Carlo simulation in Crystal Ball software. ANFIS optimum result with gbell membership function gave R2 (0.69651), MSE (0.0825). Optimum GA-ANFIS-BB parameters (PS =90, COP=0.162, and MR=0.305) gave minimised MSE = 0.0085 and R2 = 0.998. The results showed that GA-ANFIS-BB predictability degree is higher than ANFIS; thus, GA-ANFIS-BB predicted HBSOE satisfactorily. Mean Monte Carlo base case simulation gave 65.09% certainty of the MSE. Sensitivity analysis shows that COP and MR have 51.7.6% and 26.6% negative percentage contribution on MSE respectively; while PS shows a positive 21.7% contribution. Thus, GA-ANFIS-BB model in this study can be used as a precursor and predictive tool for HBSOE fuzzy-based controller system design
TABLE OF CONTENTS
Title Page i
Declaration ii
Dedication iii
Certification iv
Acknowledgement v
Table of Contents vi
List of Tables vii
List of Figures viii
Abstract ix
CHAPTER 1: INTRODUCTION 1
1.1 Background of the Study 1
1.2 Problem Statement 4
1.3 Aim and Objectives 4
1.4 Justification of the Research 5
1.5 Scope of this Research 5
CHAPTER 2: LITERATURE REVIEW
2.1 Vegetable Oils 6
2.2 The Rubber Seed Oil 6
2.3 Thermosetting Materials of Plant Origin 7
2.3.1 Preparation of plant oil-based thermoset resin 7
2.3.1.1 Epoxidation process of triglycerides 7
2.4 The Importance and Uses of Bio-Based Resins 9
2.4.1 Bio-lubricants 9
2.5 Thermosets 10
2.5.1 Epoxy resins 10
2.5.2 Plasticizer 12
2.6 Adaptive Neuro Fuzzy Inference System (ANFIS) 13
2.7 Genetic Algorithm 14
2.8 Sensitivity Analysis and Uncertainty Analysis 16
CHAPTER 3: MATERIALS AND METHODS
3.1 Materials 19
3.2 Epoxidation Procedure 19
3.3 Analytical Techniques 20
3.3.1 Iodine value 20
3.3.2 Saponification value 21
3.3.3. Acid value 21
3.3.4 Specific gravity 21
3.3.5 Oxiraine oxygen content 21
3.3.6 FT-IR analysis 22
3.4 ANFIS Model Development 22
3.5 Hybridization of GA and ANFIS (GA-ANFIS) 25
3.5.1 BBD experiment for GA-ANFIS and optimization 26
3.6 Monte Carlo Sensitivity and Uncertainty Analyses 27
CHAPTER 4: RESULTS AND DISCUSSION
4.1 Physico-chemical characteristics of Rubber seed oil 28
4.2 FTIR Analysis Results 29
4.3 Descriptive Statistics of RSO Epoxidation 31
4.4 ANFIS Modelling Results 33
4.4.1 ANFIS simulation results 33
4.5 GA-ANFIS Predictive Model Optimization Results 34
4.5.1 Box-Behnken design (BBD) GA-ANFIS model adequacy 34
4.5.2 Model output Monte Carlo simulation results 37
4.5.3 Optimization of BBD-GA-ANFIS variables 39
CHAPTER 5: CONCLUSION AND RECOMMENDATION
5.1 Conclusion 43
5.2 Recommendation 43
5.3 Contribution to Knowledge 44
References 43
LIST OF TABLES
3.1 Coded values for Box-Benhken Design GA-ANFIS (BBD-GA-ANFIS) 27
4.1 Physico-chemical characteristics of rubber seed oil 28
4.2 Descriptive statistics of experimental data 33
4.3 ANFIS model efficiency for oxirane prediction 34
4.4 Experimental run for BBD-GA-ANFIS 35
4.5 Analysis of variance (ANOVA) results for response
surface BBD-GA-ANFIS 36
4.6 comparison between ANFIS and GA-ANFIS teaching 42
LIST OF FIGURES
3.1 A basic structure of the ANFIS 23
3.2 Flow chart of GA-ANFIS methodology 26
4.1a The Fourier transform infrared spectroscopy of the pure sample of RSO 31
4.1b The Fourier transform infrared spectroscopy of the
epoxidized sample of RSO 31
4.2. Contribution of input variable variation on MSE 38
4.3. (a): 100% Certainty of Total data probability distribution (b): Uncertainty level of base case mean MSE of BBD-GA-ANFIS rubber oil epoxidation model 39
4.4 Optimum conditions ramp for BBD-GA-ANFIS with minimized error 40
4.5 Desirability value for individual input, output variable and combined factors 40
4.6 GA-ANFIS error and R2 estimation for (a) training error (b) testing error 41
CHAPTER 1
INTRODUCTION
1.1 BACKGROUND OF THE STUDY
Issues related to health, non-biodegradability, environmental pollution, climate change, propelled environmental protection policies has led to the search of alternatives for petroleum feedstock. Exploration for cost-effectiveness, quest for renewability, recyclability, environmental safety and sustainability have spurred researchers to search for natural materials that exploits the synthetic capabilities of nature to substitute most of our industrial feedstocks based on non-renewable origin due to their realizable potential (Seniha et al., 2006; Adekunle, 2015; Adhvaryu, 2002).
The utilization of vegetable oil as a precursor for manufacturing industrial chemicals and polymer has been of utmost importance due to its abundance and renewability (Jabar and Olagboye, 2017). This has provoked the interest of many researchers to utilize its potential as a substitute for petroleum-based feedstock in chemical industries. Vegetable oils such as groundnut, sunflower, melon, soybean, and okra seed oil are considered edible, but non-edible vegetable oil which does not vie with food is needed as a substitute for petroleum-based feedstock (Dinda et al., 2016).
Vegetable oils are made up of triglycerides that contain unsaturated fatty acids which are the reactive sites for chemical modification, hence the higher the unsaturation of fatty acids, the more reactive the oil (Saurabh et al., 2012). Rubber seed (Hevea Brasiliensis) oil is highly unsaturated and exhibits a semi-drying property, hence its utilization in paint, soap, alkyd resin, and wood polish production. (Okiemen et al., 2002; Ramadas et al., 2009; Obanla et al., 2019; Cheah et al., 2017).
Earlier researches reported on the extraction and characterization of rubber seed oil (Onoji et al., 2017; Jisieike and Betiku 2020; Mohammed et al., 2020), epoxidation and hydroxylation of rubber seed oil (Okiemen et al., 2005; Sukhawipat et al., 2020), kinetics of the epoxidation of rubber seed oil (Obanla et al., 2019; Okiemen et al., 2002) and optimization of the epoxidation of rubber seed oil (Nwosu-Obieogu et al., 2020a) were well documented, and their findings implied that rubber seed is suitable for epoxidation.
Epoxidation deals with the conversion of unsaturated fatty acids triglycerides to oxirane rings(Dinda et al., 2016; Turco et al., 2019; Farah et al., 2020). The industrial method majorly involves the reaction of the C=C double bond with a per acid; the acid is formed by the reaction of an ordinary carboxylic acid (e.g. acetic acid) with hydrogen peroxide. The epoxides formed are potential raw materials for the synthesis of cross-linkable bio-resins.
Sundry operating variables such as process time, stirring speed, catalyst concentration, and temperature influence the performance of the epoxidation (Silviana et al., 2019). However, the relationship between input epoxidation process parameters and measurable output (oxirane value) is imprecise, blur, non-linear and vague as reported by earlier researchers (Nwosu-Obieogu et al., 2019; Silva et al., 2016; Nwosu-Obieogu et al., 2020b; Matusiak and Milchert 2018).
Modelling epoxidation is of great importance to process engineer; it analyses the input-output relationship of the process variables and also gives better insight into the behavior of the process; furthermore, it is also used for process design, control, and performance monitoring.
Recent reports show the capability and efficiency of soft-computational models such as an artificial neural network(ANN), Adaptive neuro-fuzzy inference system (ANFIS), Support Vector Machine (SVM), Non-Linear Multilinear Regression (NLMLR), Genetic Algorithm (GA) and Particle Swarm Optimization (PSO) for modelling intricate and complex processes (Oke et al., 2018; Oke et al., 2019; Oke et al., 2020; Soto et al., 2015; Cervantes et al., 2018).
Earlier studies on estimation and prediction of epoxidation of oil have been based on linear regression techniques (Nwosu-Obieogu et al., 2020b; Okeimen et al., 2002). Nonetheless, the relationship between epoxidation independent and dependent process variables cannot be elucidated and explained linearly. Therefore, the use of soft-computing models provide accurate results for complicated system analysis than conventional mathematical models (Alam and Naik 2009; Onoji et al., 2017; Almeida, 2002; Ojediran et al., 2020; Soto et al., 2018; Soto et al., 2019; Castillo et al., 2018; Kaveh et al., 2018; Olajide et al., 2014; Melin et al., 2012; Castillo and Melin, 2003; Aguilar et al., 2003).
From the foregoing, ANFIS has been widely applied as an efficient predictive model for a highly non-linear relationship due to its learning capability. However, it is normal for conventional ANFIS learning algorithm such as least-square and gradient descent algorithms to be trapped or get stuck at one of the local minima and proceeds away from optimum global solution (Humaidi et al., 2019).
Bio-inspired or evolutionary algorithm such as Genetic Algorithm (GA) is capable of reaching global optimum solution without being trapped in local minimum region of the search. Thus, the proposed GA-ANFIS model in this study was further hybridized with Box-Benkhen (BB) desirability optimization algorithm of response surface methodology in order to optimally select the best GA-ANFIS tuning parameters for rubber seed oil epoxidation prediction.
The developed hybrid GA-ANFIS-BB model bridges the lacuna found in rubber seed oil epoxidation research. In this study, the performance evaluation of traditional ANFIS and integrated GA-ANFIS-BB for modelling rubber seed oil epoxidation with sensitivity and uncertainty analyses were investigated. The uncertainty analysis was used to measure the degree of associated risk and certainty on the response of the model.
1.2 PROBLEM STATEMENT
Over dependence on petrochemical derivatives for epoxide production has caused the price of synthetic plastics, coatings, resins to escalate leading to scarcity of the materials, processing of sub-standard products, release of greenhouse gas, adverse effect on health, climate change and disposal difficulties have further complicated the issues. The epoxidation process, if not controlled leads to high formation of by-products and the desired output may not be actualized from the epoxides, hence the need to estimate and predict the process, which has proven to be rigorous, imprecise, uncertain, and cumbersome based on linear regression techniques and multi objective optimization. Conventional ANFIS is imperfect by local optimum trapping problem, hence, the use of Genetic algorithm integrated with ANFIS to improve ANFIS performance as well as control the relationship among process variables.
1.3 AIM AND OBJECTIVES
The major motivation of this study is to develop robust soft-computing model (GA-ANFIS-BB) with high capability of overriding suboptimum solution in conventional gradient descent algorithm for rubber seed oil epoxidation process .
The specific objectives are to:
1. Epoxidise and characterize the rubber seed oil
2. Investigate the effect of process variables on the rubber seed epoxidation.
3. Develop neurofuzzy structure in matlab environment for the prediction of rubber seed oil epoxidation and validate statistically the best neuro-fuzzy structure for the prediction.
4. Improve the prediction efficiency of ANFIS by integrating Genetic Algorithm (GA-ANFIS)
5. Determine the optimum performance of the GA-ANFIS parameters using Box-behnken design implementing response surface methodology.
6. Investigate the sensitivity and uncertainty analyses on GA-ANFIS-BB model output using Monte Carlo simulation in crystal ball software.
1.4 JUSTIFICATION OF THE RESEARCH
Epoxidation of rubber seed oil has both economic and unique prospects in Nigeria, due to the production of epoxide form an underutilized material (rubber seed). Soft computing models controls and predicts processes for desired output in the chemical industries, hence, its adoption in this research to develop a model for rubber seed oil epoxidation.
1.5 SCOPE OF THIS RESEARCH
The scope of this research is restricted to the following areas; improvement of ANFIS prediction of rubber seed oil epoxidation using GA , optimization of the GA-ANFIS parameters using Box-behnken design implementing response surface methodology and the sensitivity and uncertainty analyses of the BBD-GA-ANFIS model output using Monte Carlo simulation.
Login To Comment