Tool Fault Diagnosis Based On Wavelet Neural Networks

With the enhancement of manufacture automatization and, especially, the appearance of flexible manufacture system (FMS), on-line monitoring of production processes has been receiving increased attention. In-process tool wear has a profound influence on the precision and roughness of workpiece, and, even results in discarded product and interrupted machine. Tool failure, statistically, accounts for over 75% of facility faults. Hence, on-line supervision of tool condition has become an urgent requirement.Wavelet transform is adept at analysing unstationary signals with its time-frequency localization ability, and more and more attention has been paid on its applications in tool condition monitoring. Neural network can effectively realize nonlinear mapping from inputs to outputs, have very good performance of pattern recognition, and is viewed as a widely used and highly effective technique.Main contents in this paper are presented below. In chapter 2, features are extracted from acoustic emission signals (AE) using wavelet transform process. In chapter 3, features obtained above are delivered to neural networks as inputs (BP, RBF and ART2), and tool fault diagnosis is implemented with relax-type wavelet neural networks. In chapter 4, the close-type wavelet networks are used for tool condition recognition.Main research work are as follows.(1) The AE signals of tool conditions is processed with wavelet analysis process, and eigen values of the root mean square (rms) are picked up as input-vectors of NN, such as BP, RBF, PNN, ART2 and the close-type WNN by which intelligent fault diagnosis is accomplished.(2) ART2 is realized here. It is known that up to 13 parameters must be needed to modulate while learning. However, there is not an exclusive solution in 13-dimension space, and...