(567g) Smart Data Analytics for Fault Detection and Its Application to Biopharmaceutical Manufacturing

Maintaining a high quality of the product is a key factor in most manufacturing processes. In order to ensure quality and maintain safe operations, process monitoring schemes are employed. Fault detection, which is the detection of abnormal operating conditions, is the first step in a chain of steps in process monitoring systems that need to be performed [e.g., see Refs. 1 to 4 and citations therein]. Many new data analytic tools have been introduced in recent years for a variety of objectives including fault detection. The increasing multitude of these powerful tools has motivated the development of smart data analytics approaches, i.e., a decision tree that automatically selects the most suitable method based on metadata and a systematic interrogation of the dataset [5].

In contrast to a past journal article [5] that developed a smart process analytics approach for building data-driven models for process optimization and control, this presentation builds such a decision tree for the objective of process monitoring. More specifically, we develop an automatic method selection strategy for the objective of fault detection, to select and apply the most suitable fault detection algorithm for a given problem. This decision tree is constructed based on (1) theoretical analyses of methods for extracting information on the data characteristic and on the mathematical frameworks underlying various data-driven methods, including latent variable methods, nonlinear methods utilizing kernel transformations, and dynamical systems methods, and (2) numerous case studies constructed to span the types of data encountered in chemical and biological process systems. The smart data analytics approach for fault detection is then demonstrated for a biopharmaceutical manufacturing process. As in past work [5], open-source software that implements the approach will be available for download.

[1] Chiang, Leo H., Evan L. Russell, and Richard D. Braatz. Fault Detection and Diagnosis in Industrial Systems. Springer, London, U.K., 2000.

[2] Isermann, Rolf. Fault-diagnosis Systems: An Introduction from Fault Detection to Fault Tolerance. Springer, London, U.K., 2005.

[3] Qin, S. Joe. Survey on data-driven industrial process monitoring and diagnosis. Annual Reviews in Control, 36(2), 220-234, 2012.

[4] Qin, S. Joe, and Leo H. Chiang. Advances and opportunities in machine learning for process data analytics. Computers & Chemical Engineering, 126, 465-473, 2019.

[5] Sun, Weike, and Richard D. Braatz. Smart process analytics for predictive modeling. Computers & Chemical Engineering, 144, 107134, 2021.