Application of Two-level Neural Networks in Diesel Engine's Fault Diagnosis

摘要

One-level neural network is widely used in fault diagnosis at present. If not only the patterns of faults are needed, the degrees of faults are also required, two-level neural networks, whose first level is used to acquire the faults' patterns, and the second level to classify the faults, can be used to diagnose faults more effectively than one-level neural networks (NN). The advantages of two-level NN lie in: when new knowledge of faults can be obtained, only the first level needs to be retrained: and when new knowledge of faults-classifying can be obtained, only the second level needs to be retrained. But to a one-level NN, no matter new knowledge of faults or faults-classifying is obtained respectively or simultaneously, the whole NN needs to be retrained. The paper has showed that two-level NN is more speeding and accurate than one-level NN in training and diagnosing. What's more, two-level NN can be used to early faults' diagnosing and forecast. Because of its much more accurate diagnosing result to early faults than that of one-level NN.