A New Neural Network Learning Algorithm And Its Application In The Classification And Recognition Of Rolling Bearing Signals

As an important part of modern machinery and equipment,whether rolling bearings are normal or not affects the smooth progress of industrial production,and rolling bearing signals contain rich bearing status information.Therefore,it is of great theoretical research value to accurately diagnose the state of the rolling bearing signal in time.To this end,this thesis takes the rolling bearing signal as the research object,combines ensemble empirical mode decomposition and the support vector machine learning network improved by surface-simplex swarm evolution,and proposes a new rolling bearing recognition diagnosis method.This method mainly focuses on the signal recognition and feature extraction.In terms of fault recognition,neural networks and support vector machines have unique advantages but also have their own shortcomings.Based on the study of many pattern recognition methods,this thesis uses the basic structure of support vector machines to build neural network models.In the process of parameter optimization of the neural network model of the vector machine structure,there are problems such as too many control parameters and be trapped in the local optimal solution easily.Therefore,this thesis uses a surface-simplex swarm evolution algorithm to optimize the structural parameters of the support vector machine.It is a new swarm intelligence optimization algorithm that uses a completely random way to establish a simplex neighborhood search operator of particles to reduce the control parameters of the algorithm and also establishes a particle multi-state search strategy to avoid the algorithm from falling into a local optimal solution.And it only has one control parameter,and strong global search capabilities.In terms of fault feature extraction,ensemble empirical mode decomposition and improved ensemble empirical modal decomposition methods are introduced to extract the characteristics of the non-stationary characteristics of the rolling bearing fault vibration signal.Specifically,the rolling bearing signal is decomposed to obtain a series of intrinsic mode function.,and then AR model parameters and autocorrelation spectral models are constructed for each mode,and AR model parameter characteristics and autocorrelation spectral characteristics are constructed.The fault features extracted by the two methods are sent to the SVM network optimized by surface-simplex swarm evolution to complete the classification and recognition of the rolling bearing signals.The experiments show that the features extracted in this paper can well reflect various fault signals,and the SVM network algorithm based on surface-simplex swarm evolution algorithm has a good convergence effect and classification performance.