3 rd International Conference on Production and Industrial Engineering CPIE-2013 104 Abstract - In the present study, artificial neural network (ANN) model has been developed to predict material removal rate (MRR), tool wear rate (TWR) and surface roughness (R a ) in Electro-discharge machining (EDM). A feed-forward neural network based on Levenberg- Marquardt back propagation technique of sigmoid activation function has been used. For this purpose, several neural network models with varying number of neurons at hidden layer have been tried. Finally, it has been found that at 10 neurons, model gives best prediction result. The validity of the neural network model has been checked with experimental data, and it has been concluded that the artificial neural network can predict the process performance with reasonable accuracy. Keywords – EDM, ANN, Feed-forward network, MRR, TWR and R a . I. INTRODUCTION In comparison with conventional machining, the unconventional machining has been more technologically advanced in process control and automation. The material removal mechanisms in unconventional process are based on different form of energy viz. electrical, chemical, thermal energy. The accuracy and the productivity of the unconventional machining process largely depend upon the input parameters setting. The parameters involve to control the machining performance are complex. Electro Discharge machining (EDM) is one of the most successful, practical and profitable non-conventional machining process for machining newly developed high strength alloys with high degree of dimensional accuracy and economical cost of production. The performance of the Electro Discharge machining depends upon suitable selection of parameters but the stochastic and complex nature of the process makes it too difficult to establish the relationships between output performance and input parameters. The erosion by an electric discharge involves phenomena such as heat conduction, melting, evaporation, ionization, formation, and collapse of gas bubbles and energy distribution in the discharge channel [1]. These complicated phenomena, coupled with the surface irregularities of electrodes, interactions between two successive discharges, and the presence of debris particles, make the process too abstruse, so that complete and accurate physical modeling of the process has not been established yet [2]. Artificial Neural Network (ANN) has the potential to prediction, optimization, modeling diagnosis, estimation in complex system [3]. Artificial Neural Network (ANN) has the capability to accommodate non-linearities, interactions and multiple variables. Neural Networks are also tolerant of noisy data and can operate very quickly in software, and in real time in hardware. Unlike statical models which generally require assumptions about the parametric nature of the factors (which may or may not be true), Neural Networks do not require a prior assumption of the functional form of model. Literature reports, Artificial Neural Network (ANN) has the potential to handle problems such as modeling, estimation, prediction and diagnosis in complex non-linear system. Kao et al. [4] have employed the feed forward neural network with back propagation algorithm for online monitoring of EDM. Based on result, they found discharge pulses as the controlling factor. Pellicer et al. [5] have used feed forward network with various back propagation algorithm like gradient descent with momentum and adaptive learning method, resilient back propagation algorithm, and Levenberg–Marquardt algorithm. It has been found that Different training algorithms have been found suitable for different geometries. Spedding et al. [6], Sarkar et al. [7], Protillo et al. [8] have developed the back propagation neural network for modeling of WEDM. It has been found that the performance of WEDM can be enhanced using neural network model. Gao et al. [9] applied artificial neural network (ANN) and genetic algorithm (GA) are used together to establish the parameter optimization model. An ANN model was set up to represent the relationship between MRR and input parameters, which adapted Levenberg-Marquardt algorithm and its network architecture was 3-26-1. It shows that the net has better generalization performance, and convergence speed is faster. Mandal et al. [10] have used ANN with GA. Their results show that the model is suitable for predicting the response parameters, such as MRR and electrode wear rate, with reasonable accuracy. Panda and Bhoi [1] have used back propagation neural network (BPNN) with Levenberg–Marquardt (LM) algorithm for the prediction of MRR. The objective of this paper is to establish neural network models for predicting material removal rate (MRR), Tool wear rate (TWR) and surface roughness (R a ) at various input conditions like current, gap voltage, pulse on time, flushing pressure and tool geometry in Electro discharge machining (EDM). For the development of the network model, multi layer back- propagation network Application of Artificial Neural Network for predicting multiple responses during EDM process Harshit K Dave 1 , Sudhanshu Kumar 1 , Vishal J Mathai 1 , Keyur P Desai 1 , Harit K. Raval 1 1 Department of Mechanical Engineering, Sardar Vallabhbhai National Institute of Technology, Surat, India (harshitkumar@yahoo.com)