Causality validation of multilevel flow modelling. (2nd September 2020)

• Record Type: Journal Article
• Authors: Nielsen, Emil Krabbe; Gofuku, Akio; Zhang, Xinxin; Ravn, Ole; Lind, Morten
• Is Part Of: Computers & chemical engineering. Volume 140(2020)
• Publication Date: 2020-09-02
• Subjects: Causality -- Multilevel flow modelling -- Validation -- Causal inference; Chemical engineering -- Data processing -- Periodicals; 660.0285
• DOI: 10.1016/j.compchemeng.2020.106944
• Languages: English
• ISSNs: 0098-1354

Abstract

Highlights: An overview of literature on model validation and evaluation of MFM is given. An approach for validating causality of multiple propagation paths is introduced for MFM. A causal comparison method is described mathematically. Conditional probabilities are used for describing qualitative causality. Two MFM models are improved based on a set of introduced revision rules that follow MFM syntax. Abstract: Multilevel Flow Modeling is a methodology for inferring causes or effects of process system anomalies. A procedure for validating model causality is proposed, as interest has increased from industry in applications to safety-critical systems. A series of controlled experiments are conducted as simulations in K-Spice, a dynamic process simulator, by manipulating actuators to analyse the response of process variables. The system causality is analysed stochastically under a defined range of randomly sampled process conditions. The causal influence of an actuator on a process variable is defined as a probability of a qualitative and discrete causal state. By testing an MFM model, and interpreting the propagation paths produced by MFM, the results from MFM are compared to the stochastic causality analysis to determine the model accuracy. The method has been applied to a produced water treatment system for separation of liquid and gas, to revise the causal relations of the model.