Process Monitoring And Fault Classification Base On Local Data Projection Transformation Analysis

With the rapid development of sensor technology and computer technology,industrial production process systems continue to develop in the direction of large-scale and complicated.When a fault occurs during production,it is difficult to judge by the traditional single-variable fault monitoring method.If we can’t find the fault in time and judge the fault type,it will further deepen the fault,which may cause economic loss or even equipment damage and casualties.Therefore,it is important to study industrial process monitoring and fault identification.In order to improve the separability of process data,this thesis conducts research on fault monitoring and fault classification based on data-driven method.The main research contents of the thesis include the following aspects:(1)Aiming at improving the separability between normal data and fault data,we proposed a new data-driven process monitoring method called Locial Data Projection Transformation Analysis(LDPTA).This method obtains the sequence information of the local data by using the sliding window projection transformation form without separating the new base vectors.This method improves the separability of the data and also effectively improves the fault detection rate.(2)Since the proposed LDPTA method can effectively improve the separability between normal data and fault data,we combine it with the random forest method to classify multiple faults.The newly proposed LDPT-RF algorithm combines the advantages of the two algorithms and can effectively classify the fault data.The fault classification ability of the method is proved by comparison with other methods.(3)We apply the proposed method to the monitoring of the operating state of the air cushion furnace during the processing of the copper strip.Through simulation experiments,we can conclude that the proposed method can effectively monitor the operating state of the air cushion furnace,and can quickly classify faults when faults are found.In order to verify the effectiveness of the proposed method,the potential of the window-based LDPTA method in monitoring continuous processes is explored using two case studies(a numerical example and the challenging Tennessee Eastman process).The performance of the proposed method is compared with the existing MSPM methods,such as PCA,DPCA and SPA.The monitoring results clearly demonstrate the superiority of our method.Combining the LDPTA method with the random forest method to achieve identification of multiple fault types,and has a good performance on the TE benchmark dataset.Finally,the method proposed in this thesis is applied to monitoring the operating state of air cushion furnace.Compared with the same type of methods,it is proved that the proposed method is more effective.