Research On Early Fault Feature Extraction And Pattern Recognition Method Of Axial Piston Pump

With the rapid development of modern industry,hydraulic pumps,as the power components of hydraulic systems,play an irreplaceable role in engineering applications.The axial piston pumps are widely used for its advantages of compact structure,high working pressure,high volumetric efficiency,and being easy for variable displacement control.In the hydraulic system,the heavy load and high-speed running for a long time make axial plunger pumps prone to various failures,which will make vibration and noise intensify,even decrease working efficiency.If it is more serious,it will cause casualty and property losses.Therefore,it is of great significance to monitor the running state of the axial piston pump and realize its fault diagnosis to ensure the normal and efficient operation.The internal structure of the piston pump is complex,and the types of faults are diverse,so the same type of faults with different degrees have different degrees impact on the pump.Especially in the incipient stage of the fault,the fault signal is relatively weak.These fault signals are easily covered by strong background noise,which makes it difficult to extract feature and locate the fault type.To slove these problems,taking axial piston pump as the research object,this paper takes early typical multi-fault diagnosis as the starting point,and takes different degrees fault diagnosis as the research goal,then using appropriate signal processing methods to complete fault diagnosis.The fault feature vectors are extracted from the vibration signal of the axial piston pump.The feature extraction and pattern recognition methods of the typical multiple faults diagnosis and different degrees faults diagnosis of the axial piston pump are thoughtly researched.This paper mainly conducts research from the following aspects:(1)Understand the internal structure and basic working principle of the axial piston pump;analyze its typical fault type and fault vibration mechanism;then determine the collection method of fault parameters.Aiming at the common faults of the swash plate type axial piston pump,the test bench is designed and built to complete the acquisition of normal signals,loose slipper,piston abrasion,valve plate abrasion and slipper abrasion with different degrees.(2)To slove these problems that the axial piston pump has a complicated structure and early weak fault signals are susceptible to noise interference,this paper proposes a method of combining variational mode decomposition(VMD),quantile permutation entropy(QPE)and multi-class support vector machine to study several typical fault signals,including those failures of slipper abrasion,loose slipper,piston abrasion and valve plate abrasion.Firstly,vibration signals of various states are collected to perform variational modal decomposition to obtain several intrinsic mode functions(IMFs).The signal reconstruction of each modal component is performed according to the correlation coefficient method.Then the quantile permutation entropy of the reconstructed signal is calculated as the eigenvector.Finally,the feature vector is input to the multi-class support vector machine for pattern recognition.Compared with other methods,the proposed fault diagnosis method proves its effectiveness.(3)The fault diagnosis methods of early weak axial piston pumps with different degrees of wear were studied.Its contents includes the following two aspects:1)Aiming at these problems that the axial piston pump with different degrees failure has similar failure characteristics that are difficult to identify,the fault diagnosis method based on the local S transform and ELM is proposed.Firstly,a local S transform is performed on the vibration signals collected in normal states and the slipper abrasion with different degrees.Then qualitatively and quantitatively compare different feature vector groups;choose to extract the maximum singular value of S matrix,the fundamental frequency energy share of the rotating shaft vibration and the base frequency energy ratio of the piston vibration as the three-dimensional feature vector.Finally,the three-dimensional feature vector is input into the extreme learning machine to complete the pattern recognition.Compared with other methods,the result shows that the proposed method can achieve higher pattern recognition efficiency with fewer feature vectors.2)Aiming at the problem that it is difficult to extract the change law of early fault features with different degrees,the traditional signal processing methods are difficult to meet the demand,this paper proposes the fault diagnosis method based on VMDF multiscale dispersion entropy and ELM.Based on the traditional VMD decomposition,this paper proposes a feature energy ratio(FER)variational mode decomposition feature energy reconstruction method(VMDF)for signal reconstruction.Then calculate the multiscale dispersion entropy(MDE)of the reconstructed signal.After analyzing the dispersion entropy at each scale,the peak scale entropy dispersion is selected as the feature vector.Finally,the extracted feature vector is input into the ELM to complete the failure mode recognition of slipper abrasion with different degrees.Compared with other methods,the result shows that the proposed method can not only reflect the changing law of fault degree,but also guarantee higher pattern recognition efficiency.