Neural Networks and Deep learning approach to predict the Tool life during turning of Nimonic C-263 Alloy

Abstract

In the manufacturing industry, tool wear prediction is critical for increased productivity and product quality. The focus of this study is to compare the results obtained via an artificial neural network (ANN) models for tool wear prediction on machining a nickel based super alloy: Nimonic C-263 alloy with other prediction method and the effects different parameters have on the efficiency of prediction. In this study, flank wear area (in micron2) is used as the wear indication variable during machining since it influences the precision, stability, and reliability of the machine, whereas cutting speed, feed, and depth of cut are used as input factors. Other variables include cutting force, temperature and MRR. The experimentally determined values will be utilised to train an artificial neural network (ANN) for tool wear prediction. To minimise extraneous variables, correlations between the parameters and wear are determined. The ANN model's error of forecasted response values is compared to other models such as linear regression, as well as prediction values without the elimination of variables. A GUI will also be created that lets the user input the necessary parameters to trained neural network model which then predicts the flank wear area. Results indicate that eliminating parameters resulted in reduction of prediction error (from 11–8%) and also it was found that ANN produced better results than linear regression, decision tree and random forest. Hence, more precise estimation of tool wear can be performed in a lesser time-frame. Number of experimental trials had to be limited due to unfavourable circumstances. The dataset size had been increased by assuming linearity between two results. However, this is justified as the objective is to compare prediction accuracies before and after elimination of variables after a correlation study.