SENSOR-FAULT TOLERANT CONDITION MONITORING OF AN INDUSTRIAL MACHINE
Abstract
In this paper, a multi-sensor condition monitoring scheme is developed to diagnose machine faults in the presence of sensor failure. The signals from the monitored machine are decomposed using the wavelet packet transform (WPT). Two feature reduction schemes, using genetic algorithms are developed for feature selection in condition monitoring. One scheme assumes no prior knowledge about system costs or failure characteristics, and the other scheme aims to minimize the operating costs over a period of time. Two classifiers, radial basis function networks and support vector machines, are developed and compared in their ability to classify machine faults under conditions of sensor failure. The developed methodology is implemented in an experimental system, an industrial fish processing machine. The machine is instrumented with multiple accelerometers and microphones to continuously acquire signals of machine vibration and sound. The performance of the implemented fault diagnosis methodology is evaluated though experimentation.
