Research On Feature Extraction And Improved Extreme Learning Machine Classification Method In Bearing Fault Diagnosis

With the industrial intellectualization and the increasingly complex structure of machinery and equipment,the health state monitoring of machinery and equipment has become an important research direction.As a key component of equipment,rolling bearing has become an important monitoring object.The complicated and changeable working conditions of mechanical equipment put forward higher requirements for fault state diagnosis methods.Aiming at the difficulty of state feature extraction,a feature extraction method based on vibration signal decomposition and entropy value is proposed.Aiming at the low accuracy of fault diagnosis,a fault diagnosis method based on the improved Extreme Learning Machine(ELM)was proposed.The main research contents of this paper are as follows:Firstly,the vibration signal Decomposition and entropy feature extraction method based on the improved Complete Ensemble Empirical Mode Decomposition(CEEMD)algorithm were studied.In order to solve the problem of noise and Mode mixing in the Decomposition Mode of CEEMD algorithm,the local mean value of the signals with special noise was calculated as residual,and the Permutation Entropy Improved Complete Ensemble Empirical Mode Decomposition(PEICEEMD)algorithm was proposed to detect abnormal signals in the components.This method can decompose the bearing vibration signal effectively and avoid the problem of mode mixing effectively.In order to extract complete vibration signal features,combining the advantages of fuzzy approximate Entropy and weighted permutation Entropy,the Hybrid Entropy(HE)value of the Intrinsic Mode Function(IMF)is proposed as the feature vector.The experimental results show that the bearing signal features extracted by this method are more complete and conducive to subsequent fault diagnosis.Secondly,the fault vibration signal decomposition was further studied.Since the parameters of Adaptive Local Iterative Filtering were set by human experience to affect the decomposition effect,Improved Adaptive Local Iterative Filtering(IALIF)was proposed to decompose the signal by using the Improved Sine Cosine Algorithm(ISCA)to find the optimal solution of the parameters.Experimental results show that this method can decompose bearing vibration signals more effectively.Because the permutation entropy ignored the amplitude information of the signal and could not represent the abrupt signal,Rayleigh entropy was introduced on the basis of the amplitude-aware permutation entropy,and the Rayleigh permutation entropy was proposed as the eigenvector.Experimental results show that this method can effectively detect the impact signal,and is more sensitive to the amplitude change.Thirdly,the weights and thresholds in the ELM model algorithm for fault classification have random effects on the classification accuracy,and the whale optimization algorithm is studied to optimize the weights and thresholds of ELM algorithm.Adaptive Whale Optimization Algorithm(AWOA)is proposed to improve its search Uber length by using Adaptive weight,because Whale Optimization Algorithm is easy to fall into local optimum.Experimental results show that AWOA-ELM method is effective and has advantages.Finally,the bearing experimental platform data of Case Western Reserve University in the United States,the bearing data of the Research Center of University of Paderborn in Germany,and the actual bearing data of Shanghai Baogang hot rolling mill are taken as the simulation and test objects.The method studied in this paper is used to extract the features and diagnose the data of different fault types and the same fault types with different damage sizes.The experimental results show that this method has a good diagnostic effect,practicability and superiority.