Adaptive Local Iterative Filtering Theory And Its Application In Mechanical Fault Diagnosis

Condition monitoring and fault diagnosis of mechanical equipment are of great theoretical and practical significance to ensure healthy operation of mechanical equipment,early warning of early faults,and correct location and diagnosis of faults occurring.Most of the mechanical equipment vibration signals are nonlinear and non-stationary signals,therefore,the key to mechanical equipment fault diagnosis is how to extract fault features from nonlinear and non-stationary signals and perform pattern recognition.Adaptive local iterative filtering(ALIF)is a new nonlinear adaptive decomposition algorithm,which adopts the EMD sieving framework and adaptively decomposes complex signals into a series of intrinsic mode functions(IMFs)based on iterative filtering and the Fokker-Planck differential equation.Therefore,based on an in-depth study of ALIF,this paper proposes an improved ALIF theory to address the parameter selection problem,the modal mixing problem and the endpoint effect problem of ALIF.The focus is aimed at optimizing the filter performance of ALIF in the process of processing the measured signals,making the decomposed IMF accurately characterize the system dynamics information,and using its decomposition characteristics and its dynamics characterization properties for mechanical fault diagnosis.The effectiveness and superiority of the improved ALIF theory are illustrated through theoretical analysis,equivalent filtering tests,equivalent shock response,simulation performance analysis and experiments.In this paper,the research of mechanical fault diagnosis method based on the improved ALIF theory is carried out with rolling bearings as the object,and the improved ALIF theory is successfully applied to mechanical fault feature extraction and fault classification.The main work of the paper includes the following aspects:(1)The basic theory of ALIF is studied in depth,and the effectiveness of ALIF is illustrated by equivalent filter analysis,equivalent impulse response test,simulation performance analysis and feature extraction based on rolling bearings,and a solid foundation is laid for the subsequent improvement of ALIF.On this basis,ALIF is combined with Teager energy operator for rolling bearing fault diagnosis,and the fault features are extracted effectively.(2)For the selection of steady-state coefficients and threshold parameters in ALIF has a large impact on the decomposition results,a parameter-optimized ALIF-based faint bearing fault diagnosis method is proposed based on the global optimizing performance of particle swarm algorithm.Based on the obvious sparsity of the envelope spectrum of a bearing after it generates a fault,the minimum envelope entropy in each evolution process is used as the fitness function in the PSO parameter search.For the inevitable retention of noise components in the original signal in the IMF components obtained after decomposition,a noise reduction method based on the singular value difference spectrum is proposed to further enhance the diagnosis results.Simulation and experimental results verify that the parameter-optimized ALIF has good decomposition performance,and the proposed diagnosis method can accurately identify the fault characteristics of bearings,which has practical application value.(3)The modal mixing problem suffered by ALIF in processing mechanical equipment vibration signals is analyzed in depth,and it is found that the high-frequency intermittent signals and high-frequency discontinuous signals that are widely present in mechanical equipment vibration signals can destroy the performance of the original ALIF filter,which is the root cause of its modal mixing.The filter characteristics of ALIF in the background of Gaussian white noise are studied in depth and the complementary ensemble adaptive local iterative filtering(CEALIF)method is proposed.The proposed CEALIF method is applied to the early bearing fault diagnosis in the whole life cycle,and the weighted kurtosis index is defined to select the more sensitive components of the shocks,and the comparative analysis results verify the effectiveness and application value of the proposed method in the early bearing fault diagnosis.(4)To address the endpoint effect problem that appears in ALIF,based on the in-depth study of the characteristics of ALIF filter banks,it is found that ALIF is missing the information at both ends of the signal to be processed when constructing the Fokker-Planck filter,which makes errors and repeated inward expansion in the computation of the convolutional sliding operator and eventually causes the endpoint effect,based on which an adaptive localized filtering method based on data delay topology is proposed.Based on this,an adaptive local iterative filtering method based on data extension is proposed.By extending the data to be processed,the polar scale information of the two ends of the original signal is improved to make the decomposition results more accurate.Two extension methods are proposed,one is to use mirror extension to improve the polar scale information at both ends of the signal,and the other is to use the optimized support vector regression data prediction method to predict the time series and obtain more accurate scale information.(5)Taking rolling bearings as the research object,the improved ALIF theory is applied to the field of mechanical fault diagnosis.For the fault feature extraction probl em of rolling bearings,the CEALIF-based rolling bearing composite fault feature extraction method and the fault feature extraction method for rolling bearings of wind turbines in industrial sites are proposed.The rolling bearing fault classification under different loads and different rotational speeds,on the one hand,adopts the rolling bearing fault classification method based on CEALIF and Laplace fraction multi-domain feature selection.The multi-domain information of different rolling bearing faults is characterized from various aspects by the feature indicators in time domain,frequency domain and time-frequency domain,and the failure state of rolling bearings is expressed comprehensively.On the other hand,the rolling bearing failure signal is decomposed by parameter optimization ALIF to extract the IMF component that best represents the current state.The IMPE is used to accurately characterize the kinetic information of different faulty bearings,and the IMPE value of the most sensitive IMF is selected as the input of the neural network.The IMPE describes the time-shifted averaging index of IMF at multiple scales of coarse granularity,which is insensitive to small fluctuations of the sequence and has strong robustness,so as to realize the intelligent classification of rolling bearings with different failure degrees.