Multiscale improved Boltzmann–Shannon interaction entropy for pressure-compensated piston pump fault diagnosis

Abstract This paper presents a feature extraction method based on multiscale improved Boltzmann–Shannon interaction entropy (MIBSIE), combined with random forest (RF) for identifying the wear fault states of pressure-compensated piston pump (PcPP). The deterioration of the wear state in the key friction pairs will degenerate the performance of the piston pump. However, the external signal differences before and after wear are minimal, and it is difficult to obtain fault characteristics under different wear conditions. In addition, the distribution characteristics of the external signals are often drowned by environmental noise. Therefore, the stability and effectiveness of fault feature extraction cannot be guaranteed. To address these issues, a novel entropy method called MIBSIE is proposed. First, Boltzmann–Shannon interaction entropy (BSIE) quantifies the signal’s frequency complexity using an equal-frequency partition strategy, quantifies the spatial complexity using a geometric partition strategy, and combines the two. Then, the BSIE is extended to multiscale analysis through a coarse-graining process to obtain richer fault features. Third, the obtained features are input into the RF classifier for supervised training. Finally, the effectiveness of the proposed method is verified through simulated and experimental signals. The results show that, compared to regular information entropy methods, the proposed method has the best stability and the highest recognition accuracy in identifying the wear state of the PcPP.