Modelling of micro-electrodischarge machining during machining of titanium alloy Ti—6Al—4V using response surface methodology and artificial neural network algorithm

Abstract

Micro-electrodischarge machining (EDM) can produce microhole and other complex three-dimensional features on a wide range of conductive engineering materials such as titanium super alloy, inconel, etc. The micromachining of titanium super alloy (Ti—6Al—4V) is in very high demand because of its various applications in aerospace, automotive, biomedical, and electronics industries, owing to its good strength-to-weight ratio and excellent corrosion-resistant properties. The present research study deals with the response surface methodology (RSM) and artificial neural network (ANN) with back-propagation-algorithm-based mathematical modelling. Furthermore, optimization of the machining characteristics of micro-EDM during the microhole machining operation on Ti—6Al—4V has been carried out. The matrix experiments have been designed based on rotatable central composite design. Peak-current (Ip), pulse-on time (Ton), and dielectric flushing pressure have been considered as process parameters during the microhole machining operation and these parameters were utilized for developing the ANN predicting model. The performance measures for optimization were material removal rate (MRR), tool wear rate (TWR), and overcut (OC). The ANN model was developed using a back-propagation neural network algorithm, which was trained with response values obtained from the experimental results. The Levenberg—Marquardt training algorithm has been used for a multilayer feed-forward network. The developed model was validated using data obtained by conducting a set of test experiments. The optimal combination of process parametric settings obtained are pulse-on-time of 14.2093 μs, peak current of 0.8363 A, and flushing pressure of 0.10 kg/cm2 for achieving the desired MRR, TWR, and OC. The output of RSM optimal data was validated through experimentation and the ANN predicted model. A close agreement was observed among the actual experimental, RSM, and ANN predictive results.