A matching pursuit approach to small drill bit breakage prediction

Wavelet-based pattern classification methods, which have been used extensively in imaging applications, have barely been explored in machining monitoring and diagnostic systems. Although the engineering literature recognizes wavelet methodologies as a practical alternative to the FFT for signals containing nonstationary information, most of it concerns only the Wavelet Transform. The WT can detect time-localized pre-failure machining signals, but it leaves two practical questions unanswered. The first is how to post-process the wavelet coefficients (the WT output) to extract the needed information. The second is what method to use for 'scanning' the signal with overlapping windows, since the WT covers only a short segment of the input signal. The Matching Pursuit (MP) approach handles both of these problems in a straightforward manner. As a pattern recognition method, it is adaptable to specific machining diagnostic tasks and eliminates the post-processing requirement. The scanning technique is likewise b...