Fuzzy C-means Clustering-Based ANFIS Regression Modeling of Hybrid Laser-TIG Fabrication

Abstract

Hybrid laser-TIG welding is performed on austenitic stainless steel under mutual action of both laser and TIG heat source on weld pool. The process parameters were considered as the TIG current, laser power, pulse frequency and duration whereas the weld bead penetration was taken as the output of the system. Experimentation was conducted based on central composite design matrix. To test the suggested model, five test cases experiments were also performed by creating some random input combinations by keeping the range of each input variables intact. Multiple regression was conducted to establish the correlation between process and output parameters of hybrid laser-TIG welding. An adaptive neuro-fuzzy inference system (ANFIS) architecture had also been developed based on fuzzy c-means clustering (FCM) to conduct the regression analysis. The performance of both regression models was evaluated based on root-mean-square error and mean absolute percentage error. FCM-ANFIS regression was found to be more accurate in conducting prediction of response variable based on the value of both RMSE and MAPE in training as well as test cases. FCM-ANFIS-based regression seemed to be an advanced version of multiple statistical regression models in terms of an accurate control module design and subsequently for automation of the process.