Adaptive spatial filters for Diacoustic® analysis of mechanical systems

It has been shown that in many cases wavelet analysis of the sounds of machines will reveal whether they are normal, failing or failed. The human observer can see the differences in false color images of the time-frequency pattern. The objective of this research is to design spatial filters that can be used to recognize the state of a mechanical system from its wavelet sound pattern. Such filters should be adaptive for two reasons: first, each instance of a given type of machine will have subtle sound pattern differences due to age, employment and the particular installation. Secondly, and more importantly, one would like to maximize the generality of the application such that a given instance of Diacoustic^® analysis can analyze the health of the broadest possible range of similar machines. This paper addresses techniques of identifying classes of failure using two dimensional spatial filters. The filters used are characterized by two (orthogonal) measures for features in the wavelet patterns.