An integrated monitoring and diagnostic system for roller bearings

This paper discusses a machine fault diagnostic system which integrates three techniques: 1. An autoregressive model, which compresses digitised vibration signals and preserves the information carried in the original signal. 2. A supervised artificial neural network for fault classification. 3. A fuzzy logic-based ''hypothesis and test'' program, which, when the artificial neural network fails to provide any suggestion, is able to provide the human diagnostician with some initial ''educated'' guesses of machine conditions. This integrated machine diagnostic system was developed on a 486 personal computer. Throughout the course of this development, the program has been tested with three types of vibration signal: 1. Vibration signals created using bearing physical models. 2. Vibration signals collected from two laboratory experiments ruing accelerometers. 3. Vibration signals collected from real production machine tools. In this article, the authors discuss the underlying theory of those three techniques. Experimental apparatus is introduced. Performance statistics are provided. For those conditions it was designed and developed to diagnose, the program demonstrated remarkably dependable performance.