| In actual process industry production,ensuring production safety and stable product quality are the core objectives of process monitoring methods.The implementation of " borderline" control such as advanced control makes the product quality distribution in modern process industries have non-Gaussian characteristics such as heavy-tailed distribution.Most of the traditional process monitoring strategies are based on the assumption of Gaussian distribution,and in practice,such monitoring strategies will inevitably lose some important information.To address the above problems,this paper constructs a monitoring strategy based on higher-order statistics,and specifically carries out the following work:1.To address the problem that the traditional partial least squares(PLS)method is difficult to extract non-Gaussian latent structures from data effectively,this paper proposes a higher-order statistics-based projection to latent structure(HPLS)method.The method extracts latent variables by maximizing the entropy and mutual information and determines the number of latent variables based on the entropy-based cross-validation criterion.2.In this paper,an independent signal correction strategy(ISC)is proposed to minimize the mutual information between process variables and quality variables in order to determine the appropriate rejection fraction,since orthogonal signal correction(OSC)cannot reject the fraction of process variables independent of quality variables.And the ISC strategy is combined with the HPLS model to propose an ISC-HPLS based quality-related process monitoring method.3.To address the problem that the traditional statistics are difficult to analyze and deal with non-Gaussian information in the main subspace and the residual subspace in a reasonable way,this paper proposes an entropy-based statistic construction method,which uses entropy instead of the sum of squares for higher-order analysis of non-Gaussian information.Synthetic data simulation and Tennessee Eastman(TE)process simulation verified the feasibility and effectiveness of the proposed method. |
