Diesel engine fault diagnosis using intrinsic time-scale decomposition and multistage Adaboost relevance vector machine

Diesel engine is the most widely used power source of machines. However, faults occur frequently and often cause terrible accidents and serious economic losses. Therefore, diesel engine fault diagnosis is very important. Commonly, a single unitary pattern recognition method is used to diagnose the faults of diesel engine, but its performance decreases sharply when there are many fault types. Targeting this problem, a novel diesel engine fault diagnosis approach is proposed in this study. The approach is composed of four stages. Firstly, the nonstationary and nonlinear vibration signal of diesel engine is decomposed into a series of proper rotation components (PRCs) and a residual signal by the intrinsic time-scale decomposition (ITD) method. Secondly, six typical time-domain and four typical frequency-domain characteristics of the first several PRCs are extracted as fault features. Then, the modular and ensemble concepts are introduced to construct the multistage Adaboost relevance vector machine (RVM) model, in which the kernel fuzzy c-means clustering (KFCM) algorithm is used to decompose a complex classification task into several simple modules, and the Adaboost algorithm is used to improve the performance of each RVM based module. Finally, the fault diagnosis results can be obtained by inputting the fault features into the multistage Adaboost RVM model. The analysis results show that the fault diagnosis approach based on ITD and multistage Adaboost RVM performs effectively for the fault diagnosis of diesel engine, and it is better than the traditional pattern recognition methods.