Research On Fault Diagnosis Of Ventilator Bearing Based On Weighted Permutation Entropy And ELM

Monitoring running status of rolling bearing which is the vital component of ventilator is the key technology to ensure safe production of coal mine.In this paper,vibration signals of rolling bearing in different states are selected as study subject,and the bearing fault warning and classification diagnosis are based on signal characteristic parameters.In fault early warning step,mel frequency cepstrum coefficient of vibration signal is different in different operating states,and its first dimension parameter is multiplied with spectral entropy to obtain product of spectral entropy and MFCC₀?MFPH?,which improves noise resistance of signal characteristic and maximizes the difference between normal signal and fault signal.The number of classification clusters is decided by bayesian information criterion,moreover,fuzzy C-means algorithm is used to adaptively obtain the high and low thresholds about MFPH of signal.Finally,double threshold method is used to find the starting point of fault signal,which realizes fault early warning adaptively.In signal decomposition step,Empirical Mode Decomposition?EMD?is taken as basic decomposition method,and correlation coefficient is used to eliminate false components.Support Vector Regression?SVR?extension followed by decomposition is adopted to suppress end effect.Adaptively Ensemble Local Integral Mean Decomposition?AELIMD?is used to alleviate mode mixing of EMD through assistance of noise data analysis.,and it is proved that AELIMD can be used to decompose signals to obtain more realistic frequency components through comparison experiments with EMD and Complementary Ensemble Empirical Mode Decomposition.In feature extraction step,the composition methods of feature vector about energy,permutation entropy,and weighted permutation entropy are introduced.It is proved that energy vector is not suitable as representative feature because of its poor stability through analysis and experiments based on actual rolling bearing vibration signals.Compared with permutation entropy,weighted permutation entropy feature vector has intra-class uniformity and inter-class differences.Therefore,weighted permutation entropy feature vector is determined as eigenvector representing original signal.In classification and diagnosis step,Extreme Learning Machine?ELM?serves as the basic method for fault classification,and differential evolution?DE?algorithm is used to perform parameter optimization on input weight and hidden layer threshold of ELM.Simulated annealing?SA?algorithm is added for second optimization.ELM,DE-ELM,and SA-DE-ELM models are used as classifiers to classify experiments of actual bearing vibration signal states compared by SVM.It is proved that under the condition of weighted permutation entropy feature extraction,using SA-DE-ELM model to identify bearing fault types can obtain higher accuracy and shorter diagnosis time.This paper includes 73 figures,14 tables and 109 references.