A Permutation Entropy-based Importance Measure for Condition Monitoring Data Fusion in Fault Diagnosis

In condition monitoring and fault diagnosis, how to measure the importance degree of different condition monitoring (CM) data before data fusion is a vital issue. We propose an importance measure that can be modeled using a weighted average function. The weight is measured with the relative scale of the permutation entropy from each fault feature variable. Compared with some other importance measures in data fusion, the proposed measure focuses on the degradation trend represented by the permutation entropy, instead of the information volume represented by the Shannon entropy. Then, a multiple fault feature variable fusion method based on the proposed importance measure is further proposed in the D-S evidence theory framework. Finally, a case study involving an oil analysis-based dataset from a power-shift steering transmission is carried out to investigate the superiority of the proposed method.