Intelligent Multi-Sensor Process Condition Monitoring

Abstract

Loss of production and machine breakdown are critical challenges facing modern machining. The main objective of this work is to develop an intelligent multi-sensor process condition monitoring that is able to predict the wear propagation in the cutting tool using information obtained from the analysis of cutting force and acoustic emission (AE) signals generated during turning of steel. The time domain for cutting forces and AE signals are processed for relevant features about fresh and worn tools. Principal component analysis (PCA) is used to eliminate redundant and irrelevant features. The most relevant features are used as inputs for the two classifier used in this investigation, namely, back propagation neural network (BPNN) and polynomial classifier (PC). The classifiers parameters are optimized to achieve faster computations and better predictions. To improve accuracy, leave-one-out (LOO) method is used to train both classifiers. LOO uses all the data samples for training the system. Classifiers training is modeled by correlating the extracted features with the actual measured tool wear. Comparing to BPNN, PC shows a dramatic reduction in training and prediction time. The results show the effectiveness of PCA in selecting feature that retains as much as possible of the variation in the original data. Such a system is of vital importance to the automation of manufacturing facilities. Also the use of features enhances the accuracy of both method in comparison to the use of raw data.