A CNN-ELM Compound Fault Diagnosis Method Based on Joint Distribution Modification

In the industrial field, compound faults often occur on rolling bearings and it's difficult to diagnose them correctly. To solve this problem, this article proposes a CNN-ELM compound fault diagnosis method based on joint distribution modification. Firstly, considering the complementarity and coupling of data from multiple sensors, a data input trick of multi-sensor data connected in parallel is designed. Secondly, due to the discrepancy of distribution between the compound fault data features and the single fault data features, the marginal distribution matrix and the posterior distribution matrix are used to modify the CNN-ELM network, so that the network can extract more reliable data features for fault diagnosis. Finally, referring to the categories and criteria of bearing damage proposed by Paderborn University, the label code is defined. The corresponding data set is used to verify the proposed algorithm. Experimental results show that the algorithm can accurately obtain detailed fault information such as fault location, fault type, and fault severity.