Study On Particle Swarm Optimization Diagnosis Method Of Stator And Rotor Faults For Induction Motors

Induction motor is one of the most widely used drive-equipments in industrial andagricultural production, its regular operation is essential for guaranteeing the safety inproduction. With the rapid development of modern industrial system, the capacity of asingle motor keeps increasing and the load also becomes more complicated. Once amotor failure happens, it will lead to economic loss, affect the production line safetyand product quality, and some times cause catastrophic failure. Statistical studies haveshown that the stator and rotor faults, which account for nearly30%and10%of totalfailures respectively, are the most frequent faults of induction motors. Consequently, itis of significant social and economic benefits to do the research on the stator and rotorfault diagnosis. In this dissertation, on the basis of the particle swarmoptimization(PSO) theory, stator and rotor fault diagnosis methods for inductionmotors were deeply studied. Then a modified PSO algorithm and its diagnosismethods were proposed systematically. The main works and the contributions aresummarized as follows.1) To enhance global convergence capability of particle swarm optimization andmake it suitable to actual project application, a novel hybrid algorithm was proposed,called SM-MBBPSO, based on the Nelder-Mead simplex method(SM) and a modifiedbare-bones particle swarm optimization(MBBPSO). On the one hand, an adaptiveinitialization strategy on inactive particles was proposed to maintain diversity ofswarm and improve search efficiency of particles; on the other hand, a new hybridstrategy based on K-means clustering was proposed to combine the powerful globalsearch capability of MBBPSO and the high accurate local search capability of SM.These two strategys make the hybrid algorithm achieve a nice balance of exploitationand exploration capability. Finally, simulation results on benchmark functionsdemonstrate the effectiveness of the proposed algorithm.2) In the traditional motor current signature spectrum analysis, the characteristiccomponents of broken rotor bars fault are often submerged by the fundamentalcomponent. In order to overcome this shortcoming, a fundamental-componet filteringmethod was proposed based on the SM-MBBPSO. According to the characteristic ofcurrent signal, the problem of waveform parameters identification was converted intoa optimization one. Using the waveform parameters estimated by the SM-MBBPSO,the fundamental component can be eliminated in the time domain, which can highlight the fault characteristic components. Finally, the results of simulation andlaboratory tests demonstrate the effectiveness and superiority of the proposed method.3) The impact of stator fault on the stator current harmonics and negativesequence current was analyzed in detail; and two fault detection methods based on theSM-MBBPSO were proposed. In the first method, the fundamental component waseliminated by using the above fundamental-component filtering method firstly. Then,the harmonic components of residual current signal were decomposed into series offrequency bands by wavelet packet. The variation of subband energy was treated asthe fault feature to detect stator fault. According to the characteristic of current signals,the second method can estimate precisely the amplitude and initial phase of thefundamental component by using the powerful global search capability of theSM-MBBPSO; and then calculate the value of negative sequence current directly. In areal case, the unbalanced supply voltage sources, the inherent asymmetries of themotor and load variation will cause a change of the negative sequence current. Forthis reason, the portion of negative sequence current caused by the factors other thanstator fault was taken out by a negative sequence impedance and support vectormachine; and the residual negative sequence current was used to diagnose the statorfault of induction motor. Finally, the results of laboratory tests demonstrate theeffectiveness of these two methods.4) To accurately recognize the stator and rotor faults of induction motors, a novelmethod for fault indentification was proposed based on the SM-MBBPSO and supportvector machine(SVM); and feasible diagnostic steps and analysis were alsointroduced. Firstly, the above fundamental-componet filtering method and waveletpacket were used to eliminate the influence of fundamental component and strengthenfault characteristics. Then according to the spectral characteristics of current signal infault condition, essential features of motor faults were chosen from all the frequencybands, and were considered as the input vector of SVM. The Binary tree vectormachine was used to solve the multi-class classification problem; and theSM-MBBPSO and cross-validation were taken to optimize model parameters. Finally,the experiment shows that the proposed method is effective to recognize the stator androtor faults of induction motors.