Nonlinear Fault Diagnosis Method Based On High-Order Statistics

Higher-order statistical method is a method widely used. Higher-order statistics (HOS) combining interpretative structural modeling (ISM) is used to diagnosis nonlinear factors which lead to control loop fault in chemical process. Since the occurrence of fault is often accompanied by changes in non-linear, non-linear characteristics of the diagnostic loop to determine the incidence of fault, and the application of the model is to explain the structure of complex control systems and hierarchical decomposition, thus simplifying and guiding fault detection and diagnosis.The contents of this study made as follows:①Researched traditional HOS methods, and proposed a dynamic HOS methods. Derivation algorithm of bispectrum and bicoherence is analyzed. According bicoherence, led to three indicators:non-Gaussianity index (NGI), non-linearity index (NLI) and Total non-linearity index (TNLI), which are used to quantify the degree of intensity of non-linear characteristics of control loops. Using a moving window in the traditional HOS method which is called dynamic HOS method, so that it can detect real-time dynamic nonlinear characteristics;②Research based on partial correlation coefficients and priori knowledge approach to construct signal directed graph, which are introduced in the structural model (ISM), improved ideas of traditional ISM construction to be suitable for specific industrial processes, and describe process variables causal factors;③ISM-HOS diagnostic method is proposed, which is used to nonlinear diagnosis, and determine the location and time of fault occurrence process. The proposed HOS-ISM fault diagnosis framework is verified by Tennessee Eastman process and presents improvement for highly non-linear characteristic for selected fault cases.