Study On Fault Diagnosis Of Rolling Mill Hydraulic System Based On LMD-SVM Application

Hydraulic system is one of the necessary equipment for steel enterprises.With the increasing degree of the enterprise integration,the requirement of the hydraulic system's working condition becomes more strict.Stable hydraulic system can not only provide a reliable working environment,but also effectively guarantee the quality and output of steel products.But the hydraulic system is complex,which is highly coupled with other system.So the abnormal hydraulic system pressure fluctuation is not necessarily caused by hydraulic system failure.If it has been stopped production for troubleshooting blindly,it tends to waste manpower,material resources and reduce production capacity.Therefore identifying abnormal hydraulic system pressure fluctuations and accurately finding abnormal reasons can not only ensure the hydraulic system normal working,but also improve the work efficiency and reduce maintenance costs.The pressure signals of hydraulic system in a hot rolling mill are taken as research project.Based on the original data acquisition system,the IBA-PDA hydraulic data collection and monitoring system has been added.The system can acquire,storage,analysis and process hydraulic head pressure signal to analysis and process the data online.The head pressure signal identification method based on the improved local mean decomposition(LMD)algorithm and multi feature combination is proposed for identifying hydraulic signals and finding the reason of the abnormal signal fluctuation.Firstly,several common hydraulic signal processing methods are discussed.By simulating on experimental signal and actual hydraulic signal,these algorithms' advantages and disadvantages are analyzed.Then the improved LMD algorithm is used to decompose the hydraulic signals.Secondly,energy,margin and other characteristic parameteres are extracted and formed as multi feature vectors which are put into the support vector machine(SVM)for doing learning and identifying.Finally,multivariate generalized correlation analysis is used to find the reason of abnormal signal fluctuation.The result of simulation shows that the method can identify head pressure signals effectively.It has also low false positive rate and good recognition rate in processing similar signals.