Feature Enhancement And Intelligent Recognition For Composite Faults Of Rolling Bearings

Large rotating machinery is in a harsh environment with complex working conditions,and equipment is easily damaged.As the core component of rotating machinery,rolling bearings directly affect the safety of the equipment.Not only that,the collected vibration signal often has the characteristics of non-linear,non-steady,and mixed with a large amount of external noise.Therefore,how to effectively monitor and diagnose the rolling bearing is a technical problem to be solved.In this paper,rolling bearing is regarded as the main research object.Through processing the collected bearing vibration signal,the feature enhancement and intelligent recognition methods for composite faults of rolling bearings were studied.The main content was presented as follows:(1)In the aspect of signal processing,variational mode decomposition(VMD)improved by the Grey Wolves Optimization Algorithm was used to enhance the signal characteristics.To a certain extent,VMD solved the problem of modal aliasing.But there are still problems in the in the selection of preset scales and balance constraint factors.Improper selection of parameters will cause signal under decomposition or over decomposition.In order to solve the above problem,the swarm intelligence optimization algorithm was introduced.Based on the research of several swarm intelligence algorithms,the improved VMD algorithms based on GWO was proposed.Combined with the average instantaneous frequency,the proposed optimization algorithm can effectively seek the best combination of coefficients.After the sparse decomposition and reconstruction of the processed signal by improved VMD,the noise was effectively removed.The rolling bearing faults simulation test bench was used to verify and analyze the improved algorithm.The experimental results show that:the improved VMD algorithm can effectively reduce the impact of noise,and enhance the signal characteristics.So the improved VMD method based on GWO is suitable for the rolling bearing faults signal processing.(2)In the construction of fault feature parameters,time domain feature parameters,frequency domain feature parameters,kurtosis spectral entropy and MFCC were used.In order to describe the bearing fault information in a comprehensive and accurate way,this paper first constructs a fusion feature parameter set containing multiple feature parameter types,followed by a supervised package-based feature reduction method for multidimensional fusion features.So that redundant and irrelevant feature parameters in the feature set are removed,and an optimal feature parameter set with sensitive features and outstanding discriminative ability is constructed.(3)In the aspect of fault pattern recognition,the fault classification algorithm based on improved ELM classifier was studied.In this paper,the ELM algorithm with simple model structure and strong generalization performance was applied to the bearing fault diagnosis model.However,ELM has shortcomings.In general,ELM only receives indirect information from hidden layer,but ignores the direct information from input layer.To solve the problem above,this paper proposed a parallel ELM(DP-ELM)model.Furthermore,in order to further reduce the number of hidden layer nodes and improve the classification accuracy,a double hidden layer structure was introduced.Combining the above two improved forms,DPT-ELM was proposed.Through experimental verification and analysis,it is proved that the DPT-ELM algorithm is superior to the traditional method(BP algorithm,etc.)in computing speed and classification accuracy.Meanwhile,the effectiveness of the DPT-ELM algorithm is verified by the 10 times of 10-fold cross-validation algorithm.The experimental results demonstrate the effectiveness of the improved algorithm.So it is suitable for intelligent fault diagnosis of rolling bearings.