Operation Condition Monitoring And Fault Diagnosis Of Air Cushion Furnace Based On Improved SFA

With the development of scientific technology,modern industrial systems are more and more complex.The air cushion furnace plays an important role in the hot rolling process,which can affect product quality directly.So,it is important to do research on the monitoring method of the air cushion furnace.In recent years,data-driven fault detection and diagnosis has attracted widespread attention and has become an important direction in the field of process monitoring.Process monitoring method based on multivariate statistical is a very important branch in the research of process monitoring and fault detection field.Online monitoring and identification of the abnormal situation can be carried out through the analysis and interpretation of the data sets of the process variables.Slow Feature Analysis(SFA),a new feature extraction algorithm,is able to extract the slowest varying components from time series observation data,which can be understood as slow features.Slow features can characterize the inherent properties of the process,so slow feature analysis has great potential for application.In this thesis,the shortcomings of the slow feature analysis algorithm are improved and applied to monitor the process of air cushion furnace.The main contents of this thesis are as follows:(1)SFA method rarely mentions how to select important features to reduce dimensionality,and the number of slow features generally acquired are the same as the original variables.In actual industrial process monitoring,the number of variables monitored is magnanimous,which brings inconvenience to the calculation and affects the real-time monitoring.In order to reduce dimensionality,this thesis proposes ReliefF-SFA(RSFA)method to select out the features which have ability of fault recognition.(2)Moreover,the industrial process is dynamic,and the variables are usually autocorrelated.This thesis introduces canonical variate analysis method(CVA)and combines it with ReliefF-SFA(RSFA)to solve the problem of auto-correlation.The improved CV-RSFA method is applied to monitor the operation process of the air cushion furnace,which proves that CVRSFA operates better than RSFA.(3)A model diagnosis method based on CV-RSFA combined with random forest is proposed.This method effectively combines the ability of CV-RSFA to capture the inherent properties of the system and the excellent classification ability of random forests.The simulation results based on the air cushion furnace show that the model combined with CVRSFA and random forest has good fault diagnosis ability.