Automatic detection and diagnosis systems have always attracted considerable interest in control engineering due to their positive effects of increasing safety and product quality in machinery condition monitoring and maintenance applications. Implementing automated detection and diagnosis has always been a challenge in rotating machines. In this article, we present the development of a strategy to detect and diagnose faulty bearings in a heavy-duty milling machine tool’s spindle head and its implementation in a real machine.
First, a comparison study of advanced methods for ball bearing fault detection in machine tool spindle heads is presented. Then, two automatic diagnosis procedures are compared: a fuzzy classifier and a neural network, which deal with different implementation questions involving the use of a priori knowledge, the computation cost, and the decision making process. The challenge is not only to be capable of diagnosing automatically but also to generalize the process regardless of the measured signals. Two actions are taken to achieve some kind of generalization of the application target: the use of normalized signals and the study of the Basis Pursuit feature extraction procedure. Finally, automatic monitoring system implementation on a real milling machine tool is presented.
