TOOL WEAR MONITORING OF TURNING OPERATIONS BY NEURAL-NETWORK AND EXPERT-SYSTEM CLASSIFICATION OF A FEATURE SET GENERATED FROM MULTIPLE SENSORS

Feature extraction and decision-making is a matter of considerable int erest for condition monitoring of complex phenomena with multiple sens ors. In tool wear monitoring, the extraction of subtle aspects of sign als from a range of transient and static events offers a special chall enge for diagnostic and control systems due to the broad range of info rmation in the signal. Features based on frequency spectra and statist ical transformations of a number of sensor signals were studied in an attempt to obtain a reliable indication of the evolution of tool wear. Two neural networks and an expert system using Taylor's tool life equ ation were used to classify the tool wear state. Despite the complexit y of the data and subsequent testing by the removal of two of the most clearly systematic features, a reproducible diagnosis of tool wear wa s obtained. (C) 1998 Academic Press Limited.