Research On Fault Diagnosis Method Of Industrial Process Based On Variable Selection

Fault diagnosis technology is able to identify and locate the faults in the industrial process,and improve the safety and reliability of industrial process.With the increasing degree of automation and intellectualization in modern industrial process,a large amount of data that contains useful process information is collected and obtained.Therefore,the data-driven fault diagnosis methods has received extensive attention.The research of this thesis,based on the variable selection technique,aims to solve the problems,which include unknown directions of the fault,the randomness and uncertainty of the data in the actual industrial process.The practical work and innovations are as follows:(1).The problem of fault reconstruction,which is difficult to be solved,is transformed into an easier variable selection problem.Without faulty directions,the size and directions of the fault can be calculated simultaneously.(2).The practical industrial data usually has the characteristic of randomness and uncertainty.In order to improve the robustness of fault diagnosis,the fault diagnosis problem is transformed into parametric regression problem by constructing prediction matrix.Combining with the Bayesian theory and the variable selection algorithm,the faulty probabilities of the variables can be calculated.With the faulty probability,the variables which are most likely to be the faulty variables can be identified.The experimental studies show the effectiveness of the proposedfault diagnosis method.The proposed methods can not only accurately identify the faulty variables,but also give the faulty probability of the variables,providing more useful information for the process recovery.