Mechanical condition monitoring of impulsively loaded equipment using neutral networks

The monitoring of the mechanical condition of electro-mechanical circuit breakers as reported by Demjanenko et. al. (see IEE Trans. on Power Delivery, vol. PD-7, no. 2, 1992), Park et. al. (1990), and Lai et. al. (1988) reflects the necessity of a noninvasive method for predictive maintenance. By far the most common source of malfunction of large circuit breakers is due to mechanical faults that are dependant on the number of operations of the breaker. In attempting to provide an alternative method for predicting the condition of a circuit breaker we have postulated that instead of using the spectral information we would prefer to simply make use of the original time domain signal. For the specific pattern recognition process a backpropagation trained perceptron type neural network was proposed. A variety of time domain preprocessing was applied to the signal to investigate the effect on classification. In conclusion it appears that a very accurate classification of the vibration signature of an impulsively loaded mechanical component can be achieved using a very simple neural network classifier after the application of appropriate preprocessing