Research On Fault Diagnosis Of Rotating Machinery Based On Laplace Feature Mapping Algorithm

There is a large proportion of rotating machinery in modern mechanical equipment,rotating machinery's operational condition monitoring and fault diagnosis analysis are critical to ensure the safety of equipment and the production efficiency.For rotating machinery with complex structure and system,the monitored vibration signal has obviously non-linear and non-stationary characteristics,it is hard to extract effective features from complex signal by using traditional time-frequency method.In order to identify and diagnose the fault status timely and accurately,and to extract the effective information from the monitoring signal,this paper now introduce a widely recognized and used nonlinear high-dimensional data processing method into the fault diagnosis of rotating component of rotating machinery---rolling bearings,gears and rotor system.The paper analyze theoretically the current main manifold learning method to find out the manifold learning method suitable to the field of mechanical fault diagnosis---Laplace feature mapping LE algorithm,which is used in the simulation data and the experiment data of three rotating components of rotating machinery fault.The main research content of this paper is as follows:(1)The paper analyses and summarizes the various operation failure and failure mode of the rotor system,rolling bearings,gears;elaborates the signal analysis and fault feature extraction method in time domain,frequency domain and time-frequency domain based on the testing vibration signal;puts forward to construct a high dimensional feature space containing equipment running status information by using 30 status feature vectors extracted from vibration signal,especially uses the various nonlinear state recognition method to identify the running status of rotating equipment from high dimension space recognition,and finally succeeded in applying the nonlinear manifold learning method--Laplace feature mapping algorithm into rotating equipment fault recognition.(2)The available main manifold learning methods are analyzed theoretically in the dissertation.Those methods can be classified into two categories which keep local features and global properties of the manifold topological structure in the process of dimension reduction.In order to contrast the validity of the methods,three classic high-dimensional nonlinear data sets with different characteristics: Swiss Roll,Swiss Hole and Punctured Sphere are processed by reducing dimension.Results from theoretical analysis and experiment indicated that Laplace Feature Mapping LE algorithm showed a strong “birds of a feather flock together” in nature,which coincident with clustering feature analysis in the pattern recognition.So we propose to apply LE recognition algorithm into fault diagnosis of rotating machinery.(3)As for the problem of k-nearest neighbors and N sample in manifold learning algorithm,the two groups of high dimensional data sets were processed for dimension reduction analysis by using 6 kinds of classic manifold learning algorithm.Through the overall consideration by combing 2d results after dimension reduction with the time required,we determined the values scope of k-nearest neighbors: 8 to 12,especially for the Laplace feature mapping LE algorithm values,K values 8.(4)According to the feature of recognition pattern method and the dimension reduction method,we suggest that high dimensional feature space is constructed by using time domain feature vector,the ratio of each frequency band energy obtained by wavelet packet and the EMD decomposing the original signal in overall band of energy as a feature vector,which includes all kinds of running state information of the rotating equipment in the high dimensional data space,and then use a variety of study algorism of nonlinear dimension reduction to identify the status of the equipment operation.(5)The paper applies the Laplace feature mapping LE algorithm into the fault classification for rolling bearing,gear box,and rotor system respectively.Based on the principle of from simple to complex,we firstly explore the feasibility of the algorithm being used to simulation signals,and in turn classify the measured complex fault signal,and proved the validity and feasibility of LE algorithm in fault diagnosis of rolling bearing,gear box,and rotor system.(6)In order to comparing with the results of dimension reduction in three methods of traditional PCA,MDS(linear method)and KPCA(nonlinear method),the classification effect of fault samples are shown in the 3d image;by using four quantitative parameters: the sample average recognition rate,the distance between classes Sb,the distance within class Sw and the logarithmic ratio ln(Sb/Sw),we can prove the equipment running status information of former three dimensional characteristic extracted by LE algorithm occupies a large proportion.When method is applied into the fault sample classification recognition of the rolling bearing,gear box,and the rotor system,it has certain advantages.To sum up,LE method has better classification effect in rotating machinery fault identification than traditional linear method,and even than nuclear KPCA.To better solve engineering practical problems,it is significantly to apply the LE algorithm into condition identification and fault diagnosis of mechanical equipment.