Research On Feature Extraction Methods For Typical Faults Of Axial Piston Pumps

The hydraulic system has the advantages of large power-to-mass ratio,being easy to implement automatic control,easy overload protection,etc.It has been widely used in the field of machinery.The hydraulic pump is the power source and fault source of hydraulic systems.It is of great significance for the fault detection and diagnosis of the hydraulic pump to improve the reliability of the high-pressure hydraulic system.Axial piston pumps possess the following advantages: compact structure,small leakage,high volumetric efficiency,high pressure,and high efficiency.Thus,they are often applied to hydraulic systems.The faults of axial piston pumps are very difficult to diagnose due to the properties such as higher concealment,diversity and complex.Therefore,it is of great significance to investigate the feature extraction method for typical faults of axial piston pumps to achieve an efficient and accurate diagnosis of axial piston pump faults.The main topics discussed in this dissertation are given as follows.(1)Typical faults of axial piston pumps includs sliding shoe wear,loose shoes,center spring failure and valve plate wear.According to the research of the fault mechanism of axial piston pumps,it can be concluded that the vibration signals of the pump casing can reflect the fault information of the typical faults of an axial piston pump.Therefore,an experimental scheme suitable for typical fault diagnosis of axial piston pumps is designed,and an axial piston pump typical fault diagnosis test bench is set up in this dissertation.(2)The vibration signal of an axial piston pump is non-stationary and nonlinear.The signal processing techniques of time domain or frequency domain cannot reflect the local characteristics of the signal.Therefore,in this dissertation,the time-frequency domain analysis methods are carried out.The time-frequency domain analysis used in this dissertation includes the feature extraction method based on empirical mode decomposition(EMD)and singular value decomposition(SVD),the feature extraction method based on local mean decomposition(LMD)and SVD,and the feature extraction method based on variational mode decomposition(VMD)and SVD.In addition,the typical faults of the axial piston pump are diagnosed by using the fault features obtained by the above three methods through extreme learning machine(ELM)models,and the effects of different fault features on the fault diagnosis accuracy are compared.By comparison analysis,it can be concluded that the diagnosis precision of the ELM model whose inputs are the fault features obtained based on the EMD and SVD feature extraction method is higher for the center spring failure;the diagnosis accuracy of the ELM model whose inputs are the fault features obtained by the LMD and SVD feature extraction method is higher for the valve plate wear;the ELM model whose inputs are the fault features calculated by VMD and SVD method is much more efficient when dealing with the typical faults of axial piston pumps,and has been successfully used to diagnose sliding shoe wear with different degrees,loose shoe failure and multi-class faults of axial piston pumps.In summary,the typical faults of axial piston pumps can be reflected more effectively by using the feature extraction method based on VMD and SVD.