Study On Fault Diagnosis Of Induction Motors Based On Bacterial Foraging Optimization Algorithm And Multivariate Relevance Vector Machine

Induction machines are complex electromechanical devices that are utilized in most industrial and agricultural applications,with the increase in production capabilities of modern manufacturing system,they often run continuously for extended hours in complex environments,and the request for the capacity of a single motor is keeping coming.Once a motor failure happens,it would not only damage the motor itselef,but only lead to disastrous consequences in a serious case.Statistieal studies have shown that the rotor broken bar,the stator winding interturn short circuit and the bearing faults,as the most common three faults of induction motors,account for nearly 88% of total failures.Therefore,it is of great theoretic and economic benefits to reseach on the three diagnosis technologies of induction motors which can detect respective faults in the incipient stage and prevent the further deterioration.Under such background,this thesis has done the research on fault diagnosis of induction motors based on bacterial foraging optimization algorithm and multivariate relevance vector machine,and the main contents and contributions of the thesis are as follows:1.To improve the global convergence of bacterial foraging optimization(BFO)and make it more suitable for the practical requirements of parameters optimization of induction motor fault diagnosis models,a fast global convergence backbone bacterial foraging optimization(BBBFO)is proposed.First,a chemotactic strategy based on Gaussian distribution is incorporated into this method through making use of both the historical information of individuals and the share information of group,thus guaranteeing its strong global search and local mining abilities;and second,a reproduction strategy based on swarm diversity is integrated into this method using bacterical diversity to evaluate individual health during the searching process,thus avoiding premature and a trap into local extremum.This method provides a more effective way for optimizing parameters of induction motor fault diagnosis models.2.A multivariate relevance vector machine based on multiresolution wavelet kernel(MWMRVM)is proposed,this method anisotropically extends kernel functions of MRVM via multiresolution and local characteristics of wavelet;it can be used to solve complex mult-ilassification and multivariate fitting problems.Compared with the model with a single kernel parameter,the proposed MWMRVM has a better flexibility and can obtain a sparse solution with more accuracy;thereby it can lay a solid foundation for establshing fault diagnosis models of induction motors.3.In the analysis for the traditional stator current spectrum,the fault characteristics of broken rotor bars of induction motors are often submerged by the fundamental component and can not be detected accurately.Aiming at this problem,an intelligent fault diagnosis method based on LDA/GSVD and BBBFO-MWMRVM for broken rotor bars of induction motors is put forward.In this method,the stator current signal,used as the medium,is denoised and reduced in its dimensionality to obtain fault feature vectors with low dimensionality,and then,they are used to fault diagnosis and detection by BBBFOMWMRVM.4.Through establishing stator winding interturn short circuit fault simulation model,it can be found that the phase shift between the stator line current and the phase voltage is very sensitive to the stator interturn short circuit fault,thus an intelligent fault diagnosis method based on phase shift and BBBFO-MWMRVM for stator winding interturn short circuit of induction motors is raised.The method uses the phase shift of the induction motors as the input fault feature vectors of BBBFO-MWMRVM,and BBBFO-MWMRVM contains not only MWMRVM classifier used to judge the phase of interturn short circuit fault but also MWMRVM fitter used for predicting turn number of short circuit.5.Aiming at the vibration mechanism and fault analysis of induction motor bearing,an effective intelligent fault diagnosis method based on hybrid domain,locality preserving projection algorithm based on geodesic distance(GLPP)and BBBFO-MWMRVM for bearing fault diagnosis of induction motors is presented.The high-dimensional and hybrid domain feature vectors are constructed by using time and frequency domains and chaotic characteristics of vibration signals.Meanwhile,to reduce the computational complexity of fault pattern recognition,a locality preserving projection algorithm based on geodesic distance(GLPP)is proposed to reduce the dimension of the high-dimensional feature vectors,and then BBBFP-MWMRVM classifier is used for fault diagnosis for the dimension-reduced feature vectors.The simulations and experiments demonstrate the validity and superiority of the proposed method.