Bearing Fault Diagnosis Based On Multi Feature Extraction And PSO Ptimized Neural Network

In mechanical equipment, rolling bearing is one of the most widely used basic component, which operation condition good or not has directly related to safe production and function realization of the whole machine. With industrial technology developing rapidly, the stable operation of rolling bearing is receiving more and more attention. Based on the vibration signal of rolling bearing, this paper researches fault diagnosis methods and simulation experiments.The main works of this paper are as follows:First, this paper gives a general introduction to the rolling bearing structure,classification, principle and some common method of the rolling bearing fault diagnosis.Aiming at the limitations of past signal analysis methods, rolling bearing fault diagnoisis based on multi feature extraction and particle swarm optimized neural network is established.Secondly, this paper introduces local mean decomposition algorithm to decompose the vibration signal. The LMD analysis method can effectively deal with the complex signal and display the characteristics of local signal. However, due to the actual signal often contains a lot of noise, serious impact the accuracy and effect of local mean decomposition algorithm. Lifting wavelet structure is simple and the method is flexible,this paper introduced lifting wavelet to denoise signal, and then adopting LMD method to decompose signal. A series of simulation experiments verify the accuracy and efficiency of this method.Then the time domain parameters of fault signal, sample entropy of signal and the PF component energy after LMD decomposed and are used as fault feature vectors.Classification of bearing faults can be made by using time domain characteristic parameters, sample entropy can be used to reflect the complexity of the signal, the PF component energy after LMD decomposed can reflect the deeper level of information. The time domain parameters, sample entropy and the PF component energy can describe the internal information from different characteristics of the system. Put these characteristicparameters into neural network classification model, can make the information complementary, and make up for the lack of information in a single feature.Next, the problem of bearing fault condition recognition is researched, against the convergence problem of BP neural network, the particle swarm optimization is introduced to the BP neural network. Particle swarm optimization(PSO) is a new optimization algorithm based on swarm intelligence, has a powerful capability of global exploration.The inertia weight is an important parameter of PSO algorithm. In this paper using a hybrid particle swarm optimization algorithm based on random weighting combine with compression factor. Using this hybrid PSO algorithm to optimize BP neural network weights and thresholds can help BP neural network convergence to the global optimum rapidly.Last, the datasets of the rolling bearing fault from the Case Western Reserve University is taken as the experiment research object, and the experiment is conducted from different fault part of rolling bearing and different damage degree. The results demonstrate the method of bearing fault diagnosis based on multi feature extraction and particle swarm optimized BP neural network has achieved good results.