Data-driven Method For Multivariable Fault Diagnosis In Industrial Process

With the continuous expansion of modern industrial process scale and the increasing complexity of control system,rapid and effective monitoring and diagnosis technology is very important for ensuring the safety and reliability of production process,improving product quality and increasing enterprise profits.The rapid development of computer technology and the wide use of sensors not only facilitate the collection and storage of industrial data,but also promote the development of data-driven process monitoring and fault diagnosis technology.However,the existing data-driven methods still have many shortcomings,such as "pollution effect",low diagnostic efficiency,high misdiagnosis rate and too much reliance on historical fault data.Therefore,this thesis conducts research for the problem of multivariable fault diagnosis,and also discusses the problems of micro-fault and diagnostic efficiency.Aiming at the problems of multivariable fault and micro-fault in the complicated industrial process,a fault diagnosis method based on bayesian information criterion(BIC)is proposed.Firstly,by combining the bayesian information criterion,the multivariable fault diagnosis problem is successfully transformed into the mixed integer nonlinear programming problem.Then,in order to reduce the difficulty of calculation,the forward selection strategy is adopted to simplify the original problem into a series of nested mixed integer quadratic programming problems.Finally,the branch and bound algorithm is used to solve these problems,and the optimal solution of fault diagnosis problem is obtained.In view of the phenomenon that the amount of collected industrial data is huge and the key characteristic information is easy to be "submerged",a multivariate fault diagnosis method based on between-class difference analysis and multidimensional reconstruction-based contribution(RBC)is proposed.First,two kind of strategies,i.e.,principle component analysis(PCA)and fisher discriminant analysis(FDA),are adopted to obtain the sensitivity indicators of the variables to the faults respectively,which is taken as the basis for determining the optimal reconstruction direction.Then,multidimensional RBC is used to determine the number of fault variables based on the selected reconstruction direction.Finally,the primary and secondary fault variables are distinguished according to their fault sensitivities,and the accurate fault diagnosis is realized.Aiming at the complicated high-dimensional industrial scenarios that fault variables are difficult to diagnose and the diagnosis efficiency is low,a fault diagnosis method based on signal-noise ratio(SNR)and multidimensional reconstruction-based contribution is proposed.First of all,this method first combines the idea of SNR,and proposes a new variable screening index,namely deviation factor.Then,by calculating and comparing the deviation factors of each variable,the variables with greater changes during the fault can be screened out and used as the basis for finding the optimal reconstruction direction.Finally,on the basis of the optimal reconstruction direction,a multidimensional reconstruction-based contribution method is introduced to measure the contribution of each fault variable,so as to accurately locate fault variables.The methods mentioned above are verified in numerical simulation case,Tennessee Eastman process and continuous stirred tank reactor process.Compared with the existing fault diagnosis methods,the experimental results demonstrate the effectiveness and practicability of these new approaches.