Research on the generalisation method of diesel engine exhaust valve leakage fault diagnosis based on acoustic emission

Abstract

As the core of power machinery, diesel engines often work in harsh environments. Considering the non-linearity variation and divergence in feature distribution among different operating conditions or types of engines, it is time-consuming and impracticable to collect adequate fault-labelled datasets, therefore a novel fault diagnosis approach is proposed for diesel engine valve leakage fault diagnosis to transfer from one operating condition or type (source domain) to another (target domain) to enhance the generalization of fault diagnosis. The acoustic emission signals for a specified time interval are used as the input of a one-dimensional convolutional neural network to extract discriminative features, three domain-adaptive algorithms are applied to form fault diagnosis models and compare them. The fault simulation experimental results illustrate that the proposed method achieves high accuracy in fault diagnosis across-operating condition and across-engine-type, and the generalisation performance of the proposed method outperformed that of traditional ones.