Enhancement of the performance of a neural network based motor fault detector using graphical data analysis techniques

Motor breakdown is a problem which affects many diverse areas with little in common other than some reliance on electric motors. Because of this widespread hindrance, the monitoring and fault detection of motors is a very important topic. Neural networks can often be trained to recognize motor faults by examining certain common motor measurements. Unfortunately, several weaknesses exist for neural networks when used in this application. Examples of these shortcomings are that they can take a considerable time to train, often have less than desirable accuracy, and generally are very dependent on the choice of training data. Although neural networks can recognize the non-linear relationships that exist between motor measurements and motor faults, all aspects of the neural network fault detector performance can be improved if appropriate heuristics can be used to preprocess the input-output training relationship. The paper presents a novel approach of knowledge-based graphical data analysis for data preprocessing. The incorporation of this technique results in significant improvement of the overall performance of the neural network based motor fault detector