Reaserch Of Industrial Equipments’ Life Prediction Based On The Improved SVM

Modern industrial equipments are developing towards large-scale, high-precisiondirection. Convergence between the various components is much close, and if one ormore components fail, it will impact significantly on the entire equipment, which willlead immeasurable economic losses and even serious casualties. Thus, it is of greatsignificance to take effective maintenance measures to detect anomalies or evaluatesystems health status quantitatively. In order to solve the above problems, researchsof performance degradation analysis and remaining life prediction have been widelyconcerned of the experts and scholars at home and abroad. But, the theories are notcompleted, which need to be researched futher.Data-driven is applied widely to monitor the running state of indusrial equpments,but some confounding factors in data collection process will affect the predictionaccuracy. This paper proposed a new wavlet named pulse wavlet and applied it in datapreprocessing, after studing much about wavlet theories. And by reading and learninglots of literature and books about the life prediction algorithms, this paper proposesan improved SVM prediction algorithm. It selects the relevant fault features of onlinesignals using correlation analysis, solves the problem, that SVM parameters’ valueshave a great impact on its prediction ability, appling particle swarm optimization anddetermines the performance status of systems with regression models, in order toavoid the traditional prediction algorithms’ poor generalization ability, poor results ofonline applications and other shortcomings. It shows good generalization ability andresults of online applications.Bearing data collected from bearing bench is applied to test the method. The experimental results show that pulse wavlet has a better performance to extract the valid information, and the method could assess the performance of running bearing accurately, and based on the evaluation results it calculates the remaining life in a smaller error. It got a better improvement in prediction accuracy and generalization ability, compared with the traditional SVM and BP network.