Method using L-kurtosis and enhanced clustering-based segmentation to detect faults in axial piston pumps

Abstract

An axial piston pump is a key component in hydraulic systems and has been widely used in industries. Failure of this component will result in costly downtime and serious accidents. Therefore, fault detection of axial piston pumps is paramount. The presence of periodical impulses in vibration signals indicates the occurrence of faults in axial piston pumps. However, in the operating condition, the vibration signals of faulty axial piston pumps are often contaminated by background noise and natural periodic impulses caused by the reciprocating movement of pistons. Therefore, extracting fault features from the background noise and natural periodic impulses of pistons is an important task in the fault detection of axial piston pumps. To solve this problem, a novel hybrid method of L-kurtosis and enhanced clustering-based segmentation is proposed. Both L-kurtosis and kurtosis can recognize the impulse easily, but L-kurtosis is not as sensitive to outliers compared to kurtosis. The traditional L-kurtosis-based denoising method can only suppress background noise. The enhanced clustering-based segmentation is an innovative two-cycle clustering-based segmentation method that can extract the relatively weak fault features from the background noise and natural periodic impulses synchronously. Therefore, it is more suitable for the fault detection of practical mechanical systems (i.e., piston pump, piston engine, and vane compressor) that have strong background noise and natural periodic impulses during their operations. To illustrate the feasibility of using the L-kurtosis and enhanced clustering-based segmentation method to process faulty signals of axial piston pumps, benchmark data simulation and experimental investigations are performed. The diagnosis results show that the proposed method enables the efficient recognition of faults in axial piston pumps.