Sparse Representation For Bearing Transients Feature Extraction And Fault Diagnosis Under Variable Speed Operation

The fault diagnosis and condition monitoring are of great significance for the safe and reliable operation of machinery system.Rolling element bearing is one of the most commonly used parts of rotating machine,and its health condition is hence critical for keeping the whole machine functioning properly.Localized faults in bearing tend to result in shocks and thus arouse transient impulse responses in the vibration signal.Such fault-induced transient detection and feature analysis have been one of the most crucial as well as challenging problems for bearing condition monitoring over the decades,by which many scholars have been attracted.From the literature,most of existent methodologies are tailored for bearing fault diagnosis under a constant speed condition.However,rolling element bearings are often subjected to variable speed and load in reality,which compared with stationary condition,is more destructive and fault-prone.Signals measured from the machine working under such variable speed operations are usually more complicated,leading to obstacles for bearing transient feature extraction and fault diagnosis.It is,hence,of great significance to explore approaches customized for bearing fault diagnosis under variable speed.This research is financially supported by the Natural Science Foundation of China(NSFC)“Signal Transients Extraction under the Frame of Sparsity and Its Application in Rotating Machine Fault Diagnosis”(No.51375322).With the aim of local fault features extraction for bearing under variable speed condition,novel fault diagnosis methods based on sparse representation are proposed in this thesis,which are detailed in the following.(1)The failure types and characteristics of vibration signals of the bearing with a local fault are analyzed.The thesis also reveals the key issues of the signal sparse representation.Furthermore,the feasibility and difficulty of using sparse representation for bearing transient feature extraction under variable speed are both stated.To address such difficulties,specific methods are then proposed.(2)One of the key steps of sparse representation is to construction the complete dictionary.The secret of a successful atom selection lies in the morphological similarity between the atom and the target signal.The dictionary constructed in this way can optimally represent bearing fault induced impulses.To achieve this,this thesis firstly research the fault-induced transient morphology by modeling the bearing fault impulse response,which paves the way for the complete dictionary construction.The signal can then be sparsely represented and fast reconstructed using Stage-wise Orthogonal Matching Pursuit algorithm.Via analyzing the fault characteristic order extracted from the reconstructed signal,the fault diagnosis can then be completed.The experimental signals validate the effectiveness of the proposed method.(3)The order analysis is dependent on the shaft rotating speed when performing bearing fault diagnosis under time-varying speed.However,the way of acquiring the speed information by installation of tachometer leads to increased hardware cost and also confines the application of proposed method.To solve this problem,a novel speed estimation method based on time-frequency enhancement and fusion of different frequency bands is proposed.The TF ridge candidates are enhanced firstly by amplitude sum-square method,and then search the TF ridge candidates in lower band and the resonance band.With a partial standard deviation criterion,the TF ridge candidates are fused to acquire the accurate estimation of shaft IF.Compared with conventional methods,the proposed method performs better in precision and less sensitivity to noise.The experiment results validate the effectiveness of the suggested technique for estimation of shaft IF.With this method,the bearing fault feature can be sparsely represented and bearing fault types can be diagnosed without tachometer,which reduce the hardware cost and enhance the applicability of the engineering application.In this thesis,the bearing transient feature can be sparsely extracted via analyzing the fault-induced impulse responses,and the instantaneous shaft rotating speed can be estimated by time-frequency enhancement and information fusion under variable speed condition.With such contributions,the bearing fault can be detected and fault types can be diagnosed without any prior information but the vibration signals.The proposed methods have been examined by experimental data and offer new perspectives for bearing fault diagnosis under time-varying speed condition.