Investigation Of Fault Feature Extraction And Early Fault Diagnosis For Rolling Bearings

Rolling bearings has been widely used and key components of rotating machinery,which plays an essential role in the performance of the whole equipment.Therefore,the condition monitoring techniques of rolling bearings are always a central topic for the maintenance of rotating machinery.Based on the analysis of the existing methods of fault diagnosis of rolling bearings,aiming at the key technologies in the rolling bearings,i.e.,fault feature extraction,early fault detection and fault diagnosis,the physical essence of the fault mechanism is investigated.The advanced signal processing technologies are employed to process the vibration signal of rolling bearings.The main research contents are as follows:(1)Focused on the difference of the complexity under different bearing fault types,a feature extraction method based on multi-scale symbol dynamic entropy(MSDE)for fault location of rolling bearings is proposed.The calculation of SDE can be divided into two steps: first,apply the symbolic dynamic filter(SDF)to convert the time series into the symbol time series;Then,construct the potential state patterns matrix and state transition matrix.Four simulated signals are applied to evaluate the sensitivity of the proposed SDE method.The analysis results demonstrate that SDE not only achieves better performances in complexity estimation but also saves the computation time greatly.Combined with the concept of multiple scale analysis,multi-scale symbolic dynamic entropy(MSDE)is proposed in this paper.MSDE can describe the complexity of the vibration signal in different scales.Then,the minimum redundancy maximum relevance(m RMR)approach is introduced to refine the fault features and the least square support vector machine(LSSVM)is used to complete the fault pattern identification.The proposed method is numerically and experimentally demonstrated to be able to recognize the different fault types of rolling bearings.(2)A feature extraction method based on hierarchical fuzzy entropy(HFE)is proposed to recognize the different levels of fault severity.Focused on the disadvantages of the multi-scale fuzzy entropy(MFE)method,the concept of HFE method is proposed in this paper.HFE method considers both the low and high frequency components of the signal through the hierarchical analysis.Meanwhile,HFE is compared with MFE by analyzing the white noise and 1/f noise,the comparison results demonstrate that HFE can describe the complexity of the white noise and 1/f noise accurately.Meanwhile,HFE has a stable performance of estimating the complexity.After obtained the features using HFE,laplacian score(LS)method is applied to select the feature vectors with most important information.Lastly,binary tree support vector machine(BT-SVM)classifier is employed for pattern recognition.Compared with MFE method,the extracted fault features using HFE present better divisibility than MFE,which can effectively recognize the different levels of fault severity.(3)In order to complete the early fault warning,this paper presents a novel early fault detection method based on symbolic dynamics filtering(SDF)called anomaly measure M.Only the normal condition data are required in the proposed method,which is easy to obtain.Compared with RMS and Kurtosis factor,anomaly measure M is both sensitive to initial fault and steadily increases with the damage growth,which is suitable to monitor the bearing performance degradation.Then,a fault alarm is triggered by using cumulative sum(CUSUM).The proposed method is experimentally demonstrated to be able to detect the early fault effectively using two run-to-failure experiments.(4)Focused on the difficulty of extracting fault signatures at early stage due to the weak fault symptoms and strong noise,a strategy of combination of intrinsic characteristic-scale decomposition(ICD)and resonance-based sparse signal decomposition(RSSD)is proposed.In order to remove the noise,ICD is proposed to decompose the vibration signal.The effectiveness of ICD is tested using the simulated vibration signals,the analysis results demonstrate that ICD method has certain advantages in alleviating the mode mixing and obtaining more accurate decomposition components in comparison with LMD method.Focused on the difficulty of selection of the suitable Q-factors in RSSD,a characteristic frequency ratio(CFR)is used to optimize Q-factors.Combined with CFR index,the optimized resonance-based sparse signal decomposition(ORSSD)is proposed to select most suitable QH-factor and QL-factor.Combined with ICD,this paper presents a novel early fault diagnosis approach based on ICD and ORSSD.The experimental vibration signals collected from two bearing accelerated life test rigs are employed to evaluate the effectiveness of the proposed method.Test results demonstrate that the proposed method can extract the weak fault characteristics and complete the early fault diagnosis of rolling bearings.