Research On Fault Warning And Diagnosisof Ventilator Bearing Based On Improved VMD And ELM

As the lifeline equipment of coal mine production,monitoring the running state of mine fan is the key technology to reduce the occurrence of coal mine accidents.Therefore,this paper takes mine fan as the research object,extracts the running state information contained in its bearing vibration signal,and realizes the fault early warning and classification diagnosis of mine fan by analyzing the fault characteristics of bearing signal.In the part of signal decomposition,based on the advantages of variational mode decomposition(VMD)in signal processing,the vibration signal is processed.In order to solve the problem of improper selection of VMD parameters and the influence of endpoint effect on decomposition effect,an energy difference grid search method is proposed to optimize the VMD parameters,and support vector regression(SVR)is used to extend the vibration signal and improve the endpoint effect.Based on the bearing signal,the improved VMD is compared with EMD and VMD,which proves that the improved VMD algorithm has stronger performance and the decomposed signal components are closer to the original signal frequency components.In the part of feature extraction,the complex factors such as air door opening and closing and dust flying in the mine causes the vibration signal to be affected by the random parameters.In order to give full play to the reconstruction ability of VMD,the signal features are captured by combining the advantages of permutation entropy(PE)in the randomness of detection signal,and on this basis,multi-scale weighted permutation entropy(MWPE)is introduced to make up for the shortcomings of PE ignoring the numerical changes of subsequence and single scale.The improved VMD is compared with PE,WPE and MWPE respectively.It is proved that MWPE has stronger class difference and robustness in the characterization of vibration signal,and can be used as the eigenvector of vibration signal in this study.In the part of fault early warning,the negative selection algorithm(NSA)with the ability of self identification and non self space is introduced.Based on the improved VMD decomposition proposed in this paper,the eigenvalues of PE,WPE and mwpe are extracted respectively.The bearing normal state signal is taken as self sample,and the early warning model is established by combining with fixed radius NSA and v-detector NSA algorithm.The experiment proves that the three feature extraction methods combined with v-detector NSA can realize the fault early warning of mine fan.In the part of diagnosis and classification,ELM is used as the basis of fault diagnosis and classification algorithm.Aiming at the problem that ELM randomly generates input weight and hidden layer threshold,GWO with strong search ability is introduced to optimize elm parameters.In order to make the optimization result closer to the global optimization,the IGWO algorithm including good point set strategy,nonlinear strategy and individual memory strategy is used to make up for the shortcomings of GWO in local development and global search,and the IGWO-ELM fault diagnosis model is constructed.Finally,based on the actual signal,IGWO-ELM is compared with SVM,IGWO-SVM,ELM,PSO-ELM and GWO-ELM.It is proved that the improved VMD-MWPE feature extraction method combined with IGWO-ELM diagnosis model has higher classification accuracy and has more advantages in the field of bearing diagnosis.This paper includes 73 figures,9 tables and 103 references.