(487f) Model-Based Fault Diagnosis for Safety-Critical Chemical Reactors: An Experimental Study

Over the past 20 years, there has been an increased emphasis on ensuring safety in process industries, leading to a growing interest in fault detection and isolation. Fault diagnosis methods are typically classified into two main categories: model-based and statistical/data-driven approaches. In safety-critical chemical reactors, utilizing first-principles models to develop fault detection and isolation algorithms can provide physically meaningful residual signals from material and/or energy balances that do not close properly. This approach offers reliable fault diagnosis and a better understanding of the system's behaviour, making it a more meaningful approach than statistical methods. However, in real-world experiments, random disturbances and noises from measurement render the system not completely deterministic, a probabilistic statistical method will be incorporated into the model. In the present work, we will demonstrate the application of observer-based fault detection and isolation algorithm under sensor noises with an actual CSTR reaction experiment as an example of which partial simulation we presented last year¹.

The alkylpyridine N-oxidation process is an important reaction, this reaction is highly exothermic and can lead to safety concerns due to the possible decomposition of hydrogen peroxide, generating oxygen. Thus any possible fault in the cooling system should be detected readily to ensure a safe operation. The specific reaction has been studied extensively^2,3, and in this work, the kinetics in homogenous aqueous solution has been inspected experimentally, so a quantitative model is assumed.

Extensive research has been conducted on fault detection and isolation in linear systems⁴. More recently, generalizations have been made to a broad range of nonlinear systems⁵, and most recently, for nonlinear systems in the presence of noises⁶. There are numerous simulations and theoretical works, yet no experimental study has been reported on a nonlinear chemical reaction to the best of our knowledge. In an experiment, the noises and/or deviations from sensors and reaction rate fluctuation drive the observations or residuals generated from the observations to be a statistical time series that corresponds to a specific probability distribution.

In this work, we first broach the experimental setup, and the kinetics of the 3-Picoline oxidation process in a homogenous aqueous solution is investigated. In addition, we introduce the discrete model for this process. Residual generators are built for two specific faults: coolant inlet temperature fault and feed concentration fault. Furthermore, Generalized Likelihood Ratio (GLR) will be applied to distinguish between normal and faulty operating conditions based on exceeding a GLR-based threshold that is determined by the residual data. The fault detection and isolation is achieved on-the-fly with the experiment running.

(1) 2022 AIChE Annual Meeting. https://plan.core-apps.com/aiche2022/abstract/905c3d3d-a6fa-4446-805c-5d... (accessed 2023-03-29).

(2) Cui, X.; Mannan, M. S.; Wilhite, B. A. Towards Efficient and Inherently Safer Continuous Reactor Alternatives to Batch-Wise Processing of Fine Chemicals: CSTR Nonlinear Dynamics Analysis of Alkylpyridines N-Oxidation. Chem. Eng. Sci. 2015, 137, 487–503.

(3) Wang, J.; Huang, Y.; Wilhite, B. A.; Papadaki, M.; Mannan, M. S. Toward the Identification of Intensified Reaction Conditions Using Response Surface Methodology: A Case Study on 3-Methylpyridine N-Oxide Synthesis. Ind. Eng. Chem. Res. 2019, 58 (15), 6093–6104. https://doi.org/10.1021/acs.iecr.8b03773.

(4) Ding, S. X. Model-Based Fault Diagnosis Techniques: Design Schemes, Algorithms, and Tools; Springer Science & Business Media, 2008.

(5) Venkateswaran, S.; Liu, Q.; Wilhite, B. A.; Kravaris, C. Design of Linear Residual Generators for Fault Detection and Isolation in Nonlinear Systems. Int. J. Control 2022, 95 (3), 804–820.

(6) Venkateswaran, S.; Sheriff, M. Z.; Wilhite, B.; Kravaris, C. Design of Functional Observers for Fault Detection and Isolation in Nonlinear Systems in the Presence of Noises. J. Process Control 2021, 108, 68–85. https://doi.org/10.1016/j.jprocont.2021.10.001.