Local Mean Decomposition And Its Application In Rolling Bearing Fault Diagnosis

Whether the running state of rolling bearing is good or not, that related to the service life and performance of the whole equipment, so the bearing fault diagnosis has the realistic meaning and theoretical significance. The rolling bearing vibration signals are of strong nonlinear characteristics, traditional signal processing and analysis methods have a lot of limitations from vibration signal, so that, the improvement LMD method combining with singular value decomposition, spectral kurtosis, approximate entropy, support vector machine can make correct judge of the fault type. The main content is as follows:Do discussion and simulation to the principle of local mean decomposition algorithm, through the analysis of the decomposition principle and process, three aspects of the methods were improved to enhance the decomposition of correctness, completeness and effectiveness. First, added to the high frequency harmonic to reduce effect of mode confusion to production function(PF) component; Then, introduce rational Hermite interpolation method for amplitude segment and envelope line interpolation, optimal curve can get more intuitive original information signal average function and envelope function; Further increase the signal energy index as local mean decomposition(LMD) iteration termination conditions, so as to increase the decomposition efficiency, reduce workload analysis and the running time. This improved measure will be applied to simulated signal and actual signal, compared with the original LMD, EMD(empirical mode decomposition) in the decomposition ability, timeliness, the number of iterations and other aspects of iterations, Verified the effectiveness of the improved method.The singular value decomposition(SVD) filter noise that existed in the signal effectively and be decomposed by the LMD method, then choose the PF components, and draw the spectral kurtosis figure, then construct filter. Finally, analyzed the filtered signal by envelope spectrum, extraction fault characteristic, detects the fault type, the effectiveness of the proposed method is verified by the simulation and the actual signal. The results indicate that the characteristic frequencies can be extracted effectively using the method and be used to make correct judge of the fault type.In order to improve the classification effect of fault types,the actual bearing fault signal do LMD decomposition, extracted approximate entropy that characterizes the different components of the vibration signal PF complexity as the input vector of classifier, use the cosine adaptive inertia weight method improved particle swarm optimization(PSO) algorithm, apply to support vector machine(SVM) parameter value in the iterative optimization, in order to improve the classification accuracy and reduce the operation. The feature vector inputs into the trained PSO SVM classifiers to identify faults, the results show that classification is consistent with the actual situation, and rapidly and accurately diagnose rolling bearing failure.