Abstract In the present study, an Adaptive Neuro-Fuzzy Inference System (ANFIS) is used to model the friction stir welding (FSW) process for automation. ANFIS combines the benefits of both fuzzy logic and neural network, which enables the tool as one of the accurate and online tools for input-output modeling of any complex process. The four primary input variables of friction stir welding are considered as input, and ultimate tensile strength of the joint is considered as the output in the current work. The involvement of many variables of the fuzzy inference system and neural network in ANFIS architecture creates uncertainty during iteration. Therefore, few metaheuristic algorithms like genetic algorithm (GA) and particle swarm optimization (PSO) have been applied to the ANFIS system to fine-tune the internal parameters of ANFIS and to make prediction more precise. GA-ANFIS and PSO-ANFIS results are also compared with the established nonlinear regression model. The proposed approaches are compared based on mean square deviation, squared coefficient of correlation and mean percentage error. PSO-ANFIS had been outperformed the other two approaches based on its accuracy in prediction. The developed model of PSO-ANFIS and GA-ANFIS has been found to have a close agreement with the experimental results.
Paper
Full text
Metaheuristic Tuned ANFIS Model for Input-Output Modeling of Friction Stir Welding
Semantic Scholar · Engineering · 2019
Abstract
Abstract In the present study, an Adaptive Neuro-Fuzzy Inference System (ANFIS) is used to model the friction stir welding (FSW) process for automation. ANFIS combines the benefits of both fuzzy logic and neural network, which enables the tool as one of the accurate and online tools for input-output modeling of any complex process. The four primary input variables of friction stir welding are considered as input, and ultimate tensile strength of the joint is considered as the output in the current work. The involvement of many variables of the fuzzy inference system and neural network in ANFIS architecture creates uncertainty during iteration. Therefore, few metaheuristic algorithms like genetic algorithm (GA) and particle swarm optimization (PSO) have been applied to the ANFIS system to fine-tune the internal parameters of ANFIS and to make prediction more precise. GA-ANFIS and PSO-ANFIS results are also compared with the established nonlinear regression model. The proposed approaches are compared based on mean square deviation, squared coefficient of correlation and mean percentage error. PSO-ANFIS had been outperformed the other two approaches based on its accuracy in prediction. The developed model of PSO-ANFIS and GA-ANFIS has been found to have a close agreement with the experimental results.