A Comparative Study of Artificial Neural Network and Response Surface Methodology for Optimization of Friction Welding of Incoloy 800 H

This article deals with the optimization of process parameters for friction welding of Incoloy 800 H rod and compares the results obtained by response surface methodology (RSM) and artificial neural network (ANN). The experiments were carried out on the basis of a five-level, four-variable central composite design. The output parameters were the tensile strength and burn-off length (BOL). They were considered as a function of four independent input variables, namely heating pressure (HP), heating time, upsetting pressure (UP), and upsetting time. The RSM results showed that the quadratic polynomial model depicted the interconnection between individual element and response. For optimizing the process parameters, ANN analysis was used, and the optimal configuration of the ANN model was found to be 4-9-2. For modeling aspect, a requisite trained multilayer perceptron neural network was rooted, and a quick propagation training algorithm was used to train ANN. The purpose of optimization was to decide the maximum tensile strength and minimum burn-off length of the welded joint which was done by varying the friction welding process variables. The order of importance of input parameters for friction welding of Incoloy 800 H was HP > UP > N > BOL. After predicting the model using RSM and ANN, a comparison was made for predicting the effectiveness of two methodologies. By analyzing the results, it was observed that as compared to RSM, ANN model was more specific.