https://www.ece.lsu.edu/mcu/lawss/add_materials/FaultDetectionPart1.pdf

A Review of Process Fault Detection and Diagnosis Part I : Quantitative Model-Based Methods

A review of process fault detection and diagnosis: Part III: Process history based methods

http://web.mit.edu/braatzgroup/35_Fault_diagnosis_in_chemical_processes_using_Fisher_discriminant_analysis_discriminant_partial_least_squares_and_principal_component_analysis.pdf

Fault diagnosis in chemical processes using Fisher discriminant analysis, discriminant partial least squares, and principal component analysis

http://web.mit.edu/braatzgroup/36_Fault_detection_in_industrial_processes_using_canonical_variate_analysis_and_dynamic_principal_component_analysis.pdf

Fault detection in industrial processes using canonical variate analysis and dynamic principal component analysis

Multivariate statistical methods such as principal component analysis (PCA) and partial least square (PLS) have been successfully used in modeling multivariable continuous processes (Kaspar and Ray, 1992; Kourti and MacGregor, 1995; Chen and McAvoy, 1998) Several extensions of the conventional PCA/PLS to batch processes have also been reported, among which multiway PCA (MPCA) model is the most widely used (Wold et al, 1987; Nomikos and MacGregor, 1994, 1995; Wise et al, 1999; Smilde, 2001) The MPCA model is ill-suited for multistage batch processes, as it takes the entire batch data as a single object, and it is difficult to reveal the changes of process correlation from stage to stage Considering that the multiplicity of the operation stage is an inherent nature of many batch processes, each stage has its own underlying characteristics and the process can exhibit significantly different behaviors over different operation stages; it is desirable to develop a stage-based model that can reflect the inherent process stage nature to improve the process understanding and monitoring efficiency Kosanovich et al (1994) and Dong and McAvoy (1995) developed two MPCA/nonlinear MPCA models, utilizing the two-stage nature of a jacketed exothermic batch chemical reactor Their results show that the two-stage models are more powerful than a single model Their stage models, however, inherit the common weakness of the MPCA model that the unavailable future data in an evolving batch should be estimated for on-line monitoring A new stage-based sub-PCA modeling method is proposed in this article for multistage batch processes, based on the recognition of the following: (1) a batch process may be divided into several “operation” stages reflecting its inherent process correlation nature; (2) despite that the process may be time varying, the correlation of its variables will be largely similar within the same “operation” stage Changes in the correlation may be used to indicate changes in the process “operation” stages We have placed a quotation mark around “operation” to indicate that the operation referred to in this article may not, and does not have to, have the exact correspondence to the physical operations of the process Based on the above recognition, a representative model can be built for each stage, using the conventional two-way PCA model This allows two-way PCA to be “directly” applied to a batch process after a proper stage division; a stage division algorithm is also developed in the article A three-tank process, as an experimental verification system, is finally introduced to illustrate the effectiveness of the proposed

Sub‐PCA modeling and on‐line monitoring strategy for batch processes

Dynamic PLS modelling for process control

A huge amount of data is collected by computer monitoring systems in the chemical process industry. Such tools as principal component analysis and partial least squares have been shown to be very effective in compressing this large volume of noisy correlated data into a subspace of much lower dimension than the original data set. Because most of what is eliminated is the collinearity of the original variables and the noise, the bulk of the information contained in the original data set is retained. The resulting low dimensional representation of the data set has been shown to be of great utility for process analysis and monitoring, as well as in selecting variables for control. These types of models can also be used directly in control system design. One way of approaching this is to use the loading matrices as compensators on the plant. Some advantages of using this approach as part of the overall control system design include automatic decoupling and efficient loop pairing, as well as natural handling of nonsquare systems and poorly conditioned systems.

Michael H. Kaspar

Papers

Dynamic PLS modelling for process control

Chemometric methods for process monitoring and high‐performance controller design