Application Researches On Relevance Vector Machine-Based Fault Detection Methods For Complex Industrial Processes

Safety has always been the primary consideration in the industrial process.The complex relationship between each unit and the node in the industrial process improves the difficulty of ensuring the security of the system.When the industrial equipment fails,if the measures are not taken in time,it may cause great casualties and property loss.The complex industrial process is difficult to establish a precise mathematical model,so some traditional mechanism modeling methods are not suitable.With the development of computer technology and storage technology,a large amount of data in the industrial process can be preserved,which provides the hardware base for the data driven fault detection method.Aiming at complex industrial processes,the following research is carried out by improving the existing fault detection methods:(1)In view of the unreasonable and high demand selection of the control limit in the traditional method,this paper introduces the idea of classifier,avoids the choice of control limit,reduces the number of indexes that need attention in the process of fault detection,and improves the accuracy of fault detection.(2)Because most of the industrial process data is not satisfied with the Gaussian distribution,the ICA-RVM fault detector is proposed in this paper,which avoids the shortcomings of the calculation of the kernel density estimation method in the traditional ICA method,reduces the process monitoring index and improves the fault detection precision.(3)In view of the possible Gaussian information in the industrial process data,this paper proposes a ICA-PCA-RVM fault detector,which is used to extract non-Gaussian information and Gaussian information,and applies the RVM classifier to automatically calculate the control limit,reducing the energy dispersion of the operator and further improving the fault detection precision.(4)This paper proposed RVM classifier to build fault detector.Comparing with other classifiers,such as SVM classifier,the fault detection speed of the fault detector constructed by the RVM classifier is improved by an order of magnitude than the fault detector constructed by the SVM classifier,and the accuracy of the fault detection is also improved.The results of this paper show that the ICA-PCA-RVM fault detector can detect the fault quickly and accurately,can assist the operator to make the correct decision.