An Evaluation of Engine Faults Diagnostics Using Artificial Neural Networks

Abstract

Application of artificial neural network (ANN)-based method to perform engine condition monitoring and fault diagnosis is evaluated. Back-propagation, feedforward neural nets are employed for constructing engine diagnostic networks. Noise-contained training and testing data are generated using an influence coefficient matrix and the data scatters. The results indicate that under high-level noise conditions ANN fault diagnosis can only achieve a 50–60 percent success rate. For situations where sensor scatters are comparable to those of the normal engine operation, the success rates for both four-input and eight-input ANN diagnoses achieve high scores which satisfy the minimum 90 percent requirement. It is surprising to find that the success rate of the four-input diagnosis is almost as good as that of the eight-input. Although the ANN-based method possesses certain capability in resisting the influence of input noise, it is found that a preprocessor that can perform sensor data validation is of paramount importance. Autoassociative neural network (AANN) is introduced to reduce the noise level contained. It is shown that the noise can be greatly filtered to result in a higher success rate of diagnosis. This AANN data validation preprocessor can also serve as an instant trend detector which greatly improves the current smoothing methods in trend detection. It is concluded that ANN-based fault diagnostic method is of great potential for future use. However, further investigations using actual engine data have to be done to validate the present findings.