Experimental investigation of different NN approaches for tool wear prediction based on vision system in turning of AISI 1045 steel

Manufacturing sector is always looking for higher level of automation in various operations. However, a few key challenges affects the whole machining process from being fully automated. One of these problems is the computerized tool wear monitoring. Automatic tool condition monitoring is becoming increasingly important in the metal cutting industry due to the wear on the tool impacts the efficiency of the manufactured component. Tool wear and life are major factors that influence part quality. To evaluate the useful life of the tool, most industries rely on historical data. Tool wear may be measured in two ways: directly and indirectly. The tool wear measured using the indirect approach uses parameter that impact tool life. Tool wear is traditionally assessed using microscope, which is time-consuming method. Direct method such as computer vision system is fast and reliable approach to monitor the condition of tool during machining. The goal of this research is to employ computer vision techniques to automate flank wear assessment, predict flank wear, and improve tool life prediction. The process of measuring and monitoring tool wear is automated using computer vision techniques using industrial camera with telecentric lens. Different algorithms related to feed forward back propagation neural network: Levenberg - Marquardt algorithm, Bayesian regularization and scaled conjugate gradient are utilized to predict the tool life. Performance evaluation of these algorithms is done to find most accurate algorithm for tool wear prediction system. The presented approach helps industries to detect the status of tool wear and can be a potential approach for estimating tool life in turning operations.