Multisensor Information Fusion for Fault Diagnosis of Axial Piston Pump Based on Improved WPD and SSA-KSTTM

Abstract

Axial piston pumps have been extensively applied in hydraulic systems, and their reliability takes on critical significance in the stable operation of the entire hydraulic system. How to extract fault feature parameters and identify faults in axial piston pumps turns out to be vital in ensuring the safety and reliability of the hydraulic system. In this study, a fault diagnosis method of multisensor information fusion for axial piston pumps is proposed based on improved wavelet packet decomposition (WPD) and kernelized support tensor train machine (KSTTM). Four-layer WPD is used to obtain different frequency band features of vibration, sound, and pressure signals. The intraclass and interclass distances are then introduced for quantitative evaluation of the decomposition results, which enables the selection of the optimal wavelet basis function. Subsequently, a new structured feature representation model, the structure tensor space model, is proposed to fuse the features of multiple signals. A multiclass classifier model of KSTTM optimized by multiobjective salp swarm algorithm (SSA) is developed to identify the fault states of the pumps. The experimental results show that multisensor information fusion can obtain better results compared with a single sensor. As indicated by the results of the in-depth experiments, the tensor-based method is capable of effectively increasing the classification accuracy, and it is robust to small sample size problems compared with the conventional vector-based method.