Research On Adaptive Time And Frequency Feature Enhancement Of Mechanical System Health Monitoring

The high-end equipment manufacturing industry is the core competitiveness of a modern state and also occupies the important position in the strategic emerging industry.During the service process,the high-end mechanical equipment usually experiences complex and harsh service environment,such as high speed,high temperature,high load,which make the mechanical equipment and its key components malfunction or even failure.The faults of the important mechanical equipment will cause the economic loss and casualties.It is essential to accurately and timely diagnose faults of the mechanical equipment during its service process in order to guarantee the safe and reliable operation.This dissertation aims at studying the advanced health monitoring and fault diagnosis,and focuses on time and frequency feature enhancement.The problems of poor time-frequency resolution for fast time-varying and nonstationary signal,low accuracy at time-frequency ridge detection,and weak fault feature detection are solved in this dissertation.The high-order synchrosqueezing wavelet transform is proposed to enhance the energy concentration of the time-frequency representation,the adaptive time-frequency ridge detection algorithm is proposed to extract multiple ridges,the crossing ridge,and the gap ridge,the adaptive spectral kurtosis algorithm based on parameter estimation is proposed to adaptively estimate the property of the signal and extract the weak fault feature,the guided wave time-frequency feature enhacement algorithm and the guided wave mode estimation algorithm are proposed to enhance the guided wave features and separate the mixing modes of the guided wave.The high-order synchrosqueezing wavelet transform(HSWT)is proposed to analyze the fast time-varying and nonstationary signal.The core of the HSWT method is constructing high-order instantaneous frequency estimation operators.The improvement of the instantaneous frequency estimation accuracy of each point in the time-frequency plane can improve the energy concentration of the time-frequency representation.The HSWT method is realized in the frequency domain based on the Plancherel theorem.The theoretical framework of the HSWT method is built.The mathematical expression of the energy concentration of the time-frequency,the estimation accuracy of the instantaneous frequency,and the reconstruction accuracy in the time domain are given.The detailed proof about the mathematical expression is also given.The effectiveness of the HSWT method is investigated by a multi-component fast time-varying and nonstationary simulated signal,an experimental signal acquired from a planetary gearbox test rig,and a practical signal collected from a wind turbine planetary gearbox.The adaptive time-frequency ridge detection(ATFRD)algorithm is proposed to estimate the important time-frequency feature---the time-frequency ridge.In the ATFRE algorithm,the problem of the time-frequency ridge detection is regarded as the problem of target tracking.The target of the time-frequency ridge can be defined as three states,i.e.,existence,birth,and death.The spectrum peaks will be pre-processed before tracking the targets.The order statistics filter is used to eliminate the noise interference and the false peaks.The penalty function with an adjustable weighted factor is constructed to tracking the targets of the time-frequency ridge.The ATFRE algorithm is able to effectively extract the multiple ridges,the crossing ridge,and the gap ridge.The ATFRE algorithm is insensitive to the initial values and is robust to noises.The simulated signal with multiple ridges,the simulated signal with the crossing ridge and the gap ridge,the bat echolocation signal are used to verify the effectiveness of the ATFRE algorithm.The adaptive spectral kurtosis(ASK)algorithm based on parameter estimation is proposed to study the feature extraction of the bearing weak fault.In the ASK algorithm,a series of morphological filters with different structural elements are designed to obtain the peak distribution plane which indicates the mode number for the processed signal,Otsu's method is applied to analyze the peak distribution plane to obtain the information about the decomposed band,the adaptive filter bank based on Meyer wavelet is constructed to decompose the signal,the decomposed bands are analyzed to extract the fault feature.The proposed method is free from parameter selection and obtains a flexible and adaptive decomposition scheme.Parameters involved in the proposed method,such as the number of decomposition modes and the decomposition boundaries,are determined based on the analyzed signal.The simulated signal and the experimental signal are used to investigate the effectiveness of the ASK algorithm.The guided wave mode separation is studied to solve the mode mixing problem of the guided wave.Based on the analysis of the Hessian matrix of the time-frequency representation of the guided wave and the diffusion tensor theory,the feature approximation function is built to enhance the time-frequency feature,which solves the problem of poor time-frequency resolution due to the small excitation energy.According the estimated dispersive curves of the guided wave,the guided wave mode separation problem is regarded as the Lasso regression optimization problem,which realizes multi-mode separation.The guided wave mode-mixing signal and the fatigue crack experimental signal verify the effectiveness of the the guided wave time-frequency feature enhacement algorithm and the guided wave mode estimation algorithm.