Hybrid Taguchi - ANN based multi-input multi-output modelling of kerf characteristics in AWJM of Ti6Al4V with experimental validation

This study presents a comprehensive multi-input multi-output modelling framework for predicting kerf taper at entry and exit during abrasive water jet machining of titanium alloy Ti6Al4V (Grade 5) material. Five process parameters—water pressure ( W p ), traverse speed ( T s ), nozzle to orifice diameter ( N/O dia ), abrasive mass flow rate ( A mf ), and abrasive orifice size ( A os ) were varied systematically using a Taguchi based design of experiments, generating 27 experimental runs. Kerf taper angles at entry and exit were experimentally measured and a regression equation was developed to subsequently generate 300 input–output combinations for neural network training. Three intelligent modelling approaches, namely, adaptive neuro-fuzzy inference system, back propagation neural network, and genetic algorithm neural network (GANN) were employed to determine the optimal process parameters through forward and reverse modelling techniques. Among all the developed models, the GANN model exhibited the lowest mean-squared error, achieving values as low as 0.00041 and 0.00317 for the forward and reverse modelling techniques, respectively. The model predictions were validated against ten randomly generated experimental test cases. The GANN consistently demonstrated superior prediction accuracy with average deviations of 2.76% and 4.84% for forward and reverse modelling, respectively. Contour plots and sensitivity analysis identified W p as the most dominant parameter, contributing to approximately 70% reduction in kerf. This behavior is attributed to a more coherent and focused jet, characterized by high particle velocity and reduced jet deflection within the cutting zone. SEM analysis corroborated the computational results, revealing consistently greater kerf at entry when compared to the exit, distinctly identifying cutting, transition, and deformation zones along the kerf wall.