Modeling And Fault Detection For Industrial Processes Based On Subspace Identification Method

With the continuous development of the industrial process,the scale of production continues to expand,effective process modeling and monitoring methods are critical to ensure the safe operation of the industrial process,to ensure product quality and to improve the economic benefit.Multivariate data sampled from the complex industrial process often has nonlinear,dynamic and time-varying characteristics.The traditional canonical variate analysis algorithm cannot effectively deal with the data characteristics.In this thesis,aim at the data characteristics of industrial processes,one of subspace identification algorithms canonical variate analysis was adopted,proposed a series of recursive modeling and online detection algorithm for industrial process.Main work and contributions are the following several aspects:(1)The traditional closed-loop subspace identification method produces estimates of the deviation problem in large colored noise.The proposed recursive subspace identification method based on orthogonal decomposition.According to the state space model of closed loop system and the projection relation of data,the deterministic stochastic model is constructed,and the recursive QR decomposition of projection vector is realized by using GIVENS transform;Then,an identification algorithm with forgetting factor is introduced to construct the recursive update form of the generalized observability matrix;Pushed by the least squares method to estimate the parameter matrix recursive;Finally,the numerical example of open loop three order state space model to verify the consistent estimation algorithm can obtain the system matrix eigenvalue,when there is colored noise in the systems,the unbiased estimation of the closed-loop system identification model.(2)According to the traditional subspace identification method are used in batch form,is not conducive to the system in industrial process model online identification and recursive estimation,so we put forward the canonical variable analysis of recursive subspace identification algorithm based on closed loop.By using the sliding window method based on QR decomposition,will join the data and eliminate the effective combination of data manipulation,To achieve a fast sliding window QR decomposition algorithm,reduce the operation steps unnecessary,to further improve the computational efficiency,realize recursive update projection data matrix;Then,based on the Propagator's display signal processing method,the singular value decomposition algorithm is used to update the extended observability matrix,In order to achieve standardized variable analysis method can obtain the system matrix eigenvalue recursive subspace identification scheme based on closed-loop consistent estimation;the proposed method is used for simulation of continuous stirred tank reactor.Simulation results show that this method can not only guarantee the efficiency of the identification process,and compared with the traditional modeling method based on CVA high accuracy,in order to realize the online identification of industrial process system provides a guarantee of recursive estimation.(3)According to the characteristics of dynamic and time varying data,the method is applied to fault detection in industrial process,and the method of on-line fault detection and identification based on the canonical variable analysis is studied.By means of the theory of signal subspace tracking,the recursive singular value decomposition is realized,and the computational load of the algorithm is reduced.The fault identification method based on the weighted variable contribution is designed,and the method is applied to the fault detection and identification of the continuous stirred tank reactor.The simulation results show that the process of fault detection method provides an effective online time-varying analysis method for fault detection reduces the false alarm rate based on the traditional standard variable,has a certain theoretical significance and practical value.