The Research On Fault Diagnosis Methods Based On Improved SVM Algorithm

Nowadays,industrial processes are more complicated and expanding in scale.Meanwhile,the types of faults in industrial process are various.Therefore,it is important to diagnose faults accurately and effectively.In order to improve the performance of the fault diagnosis algorithm,an industrial process fault diagnosis algorithm called empirical mode decomposition(EMD)– just in time learning(JITL)-improved gravitational search algorithm(IGSA)-recursive least squares support vector machine(RLSSVM)is proposed.At present,fault diagnosis methods of industrial process are mainly adopted based on data driven.What's more,the data from industrial processes often contains noise because of the influence of the external condition or the device.Therefore,it is necessary to deal with the noise.But traditional methods are not good at handling noise in non-stationary and nonlinear data.This thesis applies EMD algorithm which has strong ability to decompose non-stationary signal.EMD decomposes signal into a series of independent component.Then the data are earned by the process of denoising signal.When dealing with abundant data,the traditional methods can't model timely and effectively.Furthermore,the model can't update with the change of working condition.As a result,JITL algorithm is adopted to solve the problem.JITL algorithm takes original data as the reference data set.By using the distance method,the related data in the reference data set can be found,and a mass of data will change to be small sample data.In this way,model can be updated quickly and the performance of fault diagnosis is improved.In fault diagnosis,the parameters of model have a significant impact to the diagnosis performance.This thesis adopts IGSA algorithm to optimize the parameters of model.Standard GSA algorithm has shortcomings such as slows convergence speed and falls into local optimal easily.Based on chaos algorithm and adaptive weighting,the EMD-JITL-IGSA-RLSSVM fault diagnosis algorithm is proposed to optimize the parameters,applies to the penicillin chemical process,and compares the diagnosis results with the traditional method.Theoretical research and experimental results show that the proposed fault diagnosis algorithm has better performance in fault diagnosis and increases thediagnosis quickness and accuracy.In the meantime,the research and results prove that the algorithm is better than the traditional fault diagnosis methods in diagnosis performance.