Neuro-fuzzy and neural network-based prediction of various responses in electrical discharge machining of AISI D2 steel

In the present research, two neuro-fuzzy models and a neural network model are presented for predictions of material removal rate (MRR), tool wear rate (TWR), and radial overcut (G) in die sinking electrical discharge machining (EDM) process for American Iron and Steel Institute D2 tool steel with copper electrode. The discharge current (I (p)), pulse duration (T (on)), duty cycle (tau), and voltage (V) are considered as inputs to the network. A full-factorial design was used to conduct the experiments with various levels of I (p), T (on), tau, and V. The analysis of variance results reveal that I (p) is the most influencing factor for MRR and G, having the highest degree of contributions of 87.61% and 81.90%, respectively. In case of TWR, T (on) has the highest degree of contribution of 46.05% and is the most significant factor. The half of the experimental data set was used to train the networks and was tested for convergence with a different set of data to obtain appropriate number of neurons, epoch, and the fuzzy rule base. The mean square error convergence criteria, both in training and testing, came out very well. The developed models are found to approximate the responses quite accurately. Moreover, the predicted results based on above models have been confirmed with unseen validation set of experiments and are found to be in good agreement with the experimental results. The comparison results reveal that the artificial neural network and the neuro-fuzzy models are comparable in terms of accuracy and speed, and further, the proposed models can be employed successfully in prediction of MRR, TWR, and G of the stochastic and complex EDM process.