On-line tool condition monitoring using artificial neural networks

In metal cutting processes, the condition of a cutting tool as it comes into contact with the workpiece greatly affects the quality of the machined part and hence the technical aspects and economics of the manufacturing process. On-line monitoring and assessment of the state of a cutting tool is therefore considered to be a significant factor in the cost-effectiveness of the whole process. Multiple sensors are used in this work to provide complementary information about the process and this helps to improve the confidence factor of the resulting diagnostics. The use of multiple sensors, however, entails integration and fusion of the sensory information to elicit the essential features from the data by removing the redundancy present. Artificial neural networks, which mimic the functional behaviour of the biological neural network system, are used to integrate and fuse information from the multiple-sensor source. The problem of on-line tool wear monitoring in turning operations is approached by applying a three-layered, errorback-propagation-based network for fusion of three machinery performance-indicating features. A demonstrator system has been developed from this research and is capable of classifying previously unseen data into five discrete levels (three levels of flank wear and two levels of chipping).