Research On Fault Diagnosis Of Rolling Bearing Based On Optimized EMD Decomposition And GA-BP Neural Network

With the electromechanical systems turning more complicated,the original non-linear problem has become more and more prominent.Therefor,it's not satisfactory to adopt traditional theoretical modeling and a single theory or method.Therefor,the study of the intelligent diagnosis method which is suitable for nonlinear fault signal is of great theoretical and practical value.Based on the key components of electromechanical equipment-bearing,this thesis presents a new met hod of fault diagnosis of rolling bearing,and the validity of the method is validated by experimental simulation.The key of fault diagnosis method lies in two parts,the extraction of characteristic and failure mode of recognition.In the study of the modal aliasing problem of Empirical Mode Decomposition algorithm(Empirical Mode Decomposition,the EMD),an improved EMD algorithm based on wavelet packet is proposed.The core idea of this method is to preprocess the original signal by wavelet packet analysis.The crude signal is disposed by wavelet packet analysis.Namely,on the premise of keeping the nature of the IMF,It solves the problem of orthogonality of each IMF component and inhibits the modal aliasing,thus improving the ability to decomposite the EMD.In order to reduce the signal processing time,this thesis puts forward the method of energy criterion,which is to select the reconstructed signal contained the most fault characteristic information and the highest energy as an object of the next step of the EMD decomposition.In addition,this thesis presents a new fault diagnosis algorithm based on the optimization of GA-BP neural network.In this method,the genetic algorithm has been used to optimize the network structure of the BP neural network,initial weights and threshold,and to improve the convergence precision and convergence speed of BP neural network algorithm.Finally,this thesis designed a simulation experiment using the theory of this thesis and the experimental data of bearing vibration.The experimental results show that the method can effectively extract the fault feature quantity and quickly identify the fault type,which has high feasibility and accuracy.