Fault diagnosis of rotating machinery using an intelligent order tracking system

This research focuses on the development of an intelligent diagnostic system for rotating machinery. The system is composed of a signal processing module and a state inference module. In the signal processing module, the recursive least square (RLS) algorithm and the Kalman filter are exploited to extract the order amplitudes of vibration signals, followed by fault classification using the fuzzy state inference module. The RLS algorithm and Kalman filter provide advantages in order tracking over conventional Fourier-based techniques in that they are insensitive to smearing problems arising from closely spaced orders or crossing orders. On the basis of thus obtained order features, the potential fault types are then deduced with the aid of a state inference engine. Human diagnostic rules are fuzzified for various common faults, including the single fault and double fault situations. This system is implemented on the platform of a floating point digital signal processor, where a photo switch and an accelerometer supply the shaft speed and acceleration signals, respectively. Experiments were carried out for a rotor kit and a practical four-cylinder engine to show the effectiveness of the proposed system in tracking the rotating order with precise inference.