Quality-related Intermittent Fault Detection Methods Based On Multivariate Statistical Analysis

In recent years,the quality of final products has been a key indicator of industrial processes,and the quality-related monitoring problem has become a hot research direction.Multivariate statistical analysis is a significant data-driven process monitoring method.It has been widely used in the fault detection of industrial processes and has shown excellent detection characteristics.Intermittent fault is a type of non-permanent fault,which has the characteristics of random,intermittent,repetitive,and time cumulative effect.Therefore,it is difficult to solve the problem of intermittent fault detection by using traditional multivariate statistical analysis methods directly.An intermittent fault detection method is investigated based on quality-related multivariate statistical analysis in this thesis to detect the occurrence and disappearance of intermittent faults quickly and accurately.Then an improved method is proposed based on the characteristics of intermittent faults.The main work of this thesis contains the following three parts:1.The traditional quality-related multivariate statistical analysis methods such as partial least squares(PLS),kernel partial least squares(KPLS)and dynamic partial least squares(DPLS)are applied to the research of intermittent fault detection,respectively.For the intermittent faults in Tennessee Eastman(TE)process,the above three methods are simulated respectively,and the simulation results are compared.The difficulties of intermittent fault detection and the limitations of traditional methods are analyzed.2.Considering that the spatial oblique decomposition of the PLS method cannot detect quality-related intermittent faults quickly,an intermittent fault detection method is investigated based on total projection to latent structures(T-PLS).In this method,the two subspaces obtained from PLS are further decomposed into the quality-related part and the qualityindependent part,which makes it more sensitive to quality-related faults.The effectiveness of the method in this chapter is illustrated by using a numerical example and intermittent fault simulation results of TE process.3.The disappearance of intermittent faults brings unsteady processes to the system,then a new method of intermittent fault detection is proposed based on two-step dynamic least squares(TS-DLS).Firstly,the dynamic components and innovation components of the normal process data are estimated,and the constant mean value and variance for standardization are obtained to avoid the influence of approximation errors caused by the time-varying characteristics of the dynamic process.Then the input and output matrices are augmented and the regression coefficient matrix is calculated.The principal component subspace and residual subspace of the input matrix are constructed,and the statistics are calculated by using the principal components of the residual subspace to realize the detection of intermittent fault.The feasibility and superiority of the proposed method are illustrated by using numerical cases and TE industrial process simulations.