A rule-based intelligent method for fault diagnosis of rotating machinery

Abstract

To better equip with a non-expert to carry out the diagnosis operations, a new method for intelligent fault identification of rotating machinery based on the empirical mode decomposition (EMD), dimensionless parameters, fault decision table (FDT), MLEM2 rule induction algorithm and improved rule matching strategy (IRMS) is proposed in this paper. EMD is used to preprocess the vibration signals for mining the fault characteristic information more accurately. Then, dimensionless parameters are extracted from both the decomposed signals in time domain and envelop spectrum in frequency domain respectively to form the conditional attributes of a FDT. Moreover, MLEM2 algorithm is run directly on the FDT to generate decision rules imbedded in the data. To make the following classification process more robust, the IRMS is adopted to resolve the conflicting and non-matching problems. Finally, data of rolling element bearings with four typical working conditions is used to evaluate the performance of the proposed method. The testing result demonstrates that the method has high accuracy and systematically good performance. It is proved to be a convenient, concise, interpretable and reliable way to diagnose bearings’ faults. The advantages are also confirmed by the comparisons with the other two approaches, i.e. the principal component analysis (PCA) and probabilistic neural network (PNN) based method as well as the wavelet transform (WT) and genetic algorithm (GA) based one. Furthermore, thank to the FDT working as a data interface, the method is more transplantable, therefore it may be applied to diagnose other types of rotating machines effectively.