Prediction of Machining Characteristics of Wire Electrical Discharge Machined Hastelloy-X using Artificial Neural Network

Abstract

Due to extensive mechanical load bearing capability under high temperature and pressure, Nickel based super alloys are widely incorporated in aerospace and aviation industries in various sections like chemical, fuselage, engine, combustor components, etc. Hastelloy-X is a Ni-based super alloy consisting mainly Ni, Cr, Fe, Mo and Co, which has good corrosion and heat resistance capacity. Since Hastelloy-X is a difficult-to-machine material, a non-conventional Wire Electric Discharge Machining is used. This work aims at machining characteristics study of WEDM of Hastelloy-X and prediction of major machining performances using Artificial Neural Network (ANN). At first, full factorial design of experiments was set using Minitab which includes four input machining parameters namely pulse-on time (T-on), pulse-off time (T-off), wire feed (WF) and servo voltage (SV); kept at three levels; high, medium and low. Total 81 experimental runs were performed. After machining on WEDM, machining performances MRR (material removal rate) and SR (surface roughness) were measured. There after the neural network is trained in nntool in MATLAB to predict the MRR and SR. The predicted model has mean absolute percentage error (MAPE) of 6.371% for MRR prediction and 5.92% for SR prediction while the MSE (Mean Square Error) was found to be 0.389 and 0.129 for MRR and SR respectively. The trained network has training, validation and testing regression coefficient (R) values of 0.9756, 0.9916 and 0.9662 respectively. And the overall R value was 0.97746. After prediction, the samples with extreme values of actual and predicted outputs were studied for other machining responses like recast layer, surface cracks and kerf width. Out-turn of this research can be utilized for machining hard to machine materials in a high precision WEDM for different applications.