Research On Fault Diagnosis Of Rolling Bearing Based On Adaptive Resonance-based Sparse Signal Decomposition

In the field of modern industry,rotating machinery occupies an indispensable position.The monitoring and maintenance of rotating machinery can greatly avoid the sudden problems of operating units.As a common component of rotating machinery,rolling bearing is widely used in power system,transportation,petrochemical metallurgy and other fields.However,affected by machining and assembly errors,impact load,dust and foreign matters,rolling bearings are prone to failure,which seriously threatens the safety and reliability of industrial equipment and production process.And in the actual production,the individual difference of rolling bearing life is large.Under similar working conditions,the life of bearings of the same type and batch has a high degree of dispersion,that is,various faults may still occur in the design life.Therefore,it is of great significance to develop the fault diagnosis technology of rolling bearing to ensure the safety and stability of rotating machinery.Resonance based sparse signal decomposition(RSSD)can separate complex fault signals into high and low resonance components with different resonance attributes,that is,according to the waveform characteristics,it is more suitable for processing mechanical vibration signals with nonlinear and non-stationary characteristics.However,RSSD method is sensitive to decomposition parameters,and its reliability is poor.Further optimization of RSSD method is needed to improve performance.Based on the traditional RSSD theory,this paper proposes an improved method,which studies the evaluation index and decomposition parameter optimization of RSSD deeply,and proposes an adaptive optimal decomposition parameter RSSD method.Combined with entropy theory,the general fault rules in RSSD subband are explored,and the fault signal pattern recognition of rolling bearing is completed.The main contents of this paper are as follows:(1)In the field of mechanical fault diagnosis,the current situation of bearing related feature extraction,fault pattern recognition and other research directions at home and abroad are sorted out.The advantages and limitations of different signal processing methods are summarized,and the research content of this paper is clear.(2)The decomposition parameters of RSSD have a decisive influence on its fault separation effect,but there is not a more accurate quantitative description of its separation effect.Based on the principle of RSSD,this paper analyzes the influence of RSSD parameters on the components of high and low resonance components.At the same time,two kinds of RSSD evaluation methods based on kurtosis and residual are analyzed.In order to improve the accuracy of the evaluation index,a quantitative evaluation index suitable for RSSD is proposed by establishing the vector space of RSSD input parameters,comparing the sensitivity of various time-domain and frequency-domain features to fault information,and combining the principle of maximum correlation and minimum redundancy for feature selection.The index matches the trend of the change of the frequency density coefficient of the fault characteristics,and has good interpretability and accuracy.At the same time,it can solve the phenomenon of abnormal value caused by waveform distortion in the existing kurtosis evaluation index.(3)RSSD has excellent performance in fault diagnosis of rolling bearing,but the method can not adaptively give the optimal decomposition parameters,which limits the reliability of RSSD.Subjective selection of RSSD parameters may not find enough fault information or even produce misjudgment.The intelligent group optimization algorithm has good search performance and can avoid the invalid decomposition of RSSD.However,some natural heuristic algorithms have weak approximation ability to the spatial optimal solution,resulting in low efficiency.Moreover,using the intelligent group optimization algorithm to optimize the quality factor or weight coefficient alone can not give full play to its global optimization performance.Aiming at the above two problems,this paper takes quality factor and weight coefficient as the optimization object of squirrel search algorithm at the same time,and proposes an RSSD method based on multi parameter optimization.This method uses the RSSD decomposition parameters optimized by squirrel algorithm to get the optimal low resonance component,and extracts its features to obtain the fault characteristic frequency of the signal.The effectiveness of the method is verified by the simulation signal and the actual bearing rolling element early fault signal experiment.(4)In view of the large difference in the complexity of rolling bearing vibration signal under various fault conditions,based on the above-mentioned adaptive RSSD,a fault diagnosis method of rolling bearing based on adaptive RSSD and subband entropy is proposed.Entropy can effectively represent the unbalance disturbance of mechanical equipment and obtain relevant information.It has the advantages of high calculation efficiency and reliable performance.It is widely used in feature extraction of rolling bearing fault signals.The signal is preprocessed by adaptive RSSD,the low resonance component is taken as the research object,and then combined with sub-band reconstruction,the general fault rules in the sub-band signal of RSSD are fully extracted and mined by analyzing the energy distribution and entropy value of sub-band.Finally,we use the limit learning machine for pattern recognition to complete the classification of fault signals.Through the test of experimental data,the fault diagnosis method based on adaptive resonance sparse decomposition and entropy analysis has high recognition accuracy,and realizes the accurate distinction of different fault positions of rolling bearings.