Simulation study demonstrates automated emergency classification and resolution synthesis in industrial process models, highlighting scalable decision support.
Key Points
The study aims to develop and validate a hybrid intelligent decision support method that identifies unprecedented emergency conditions and synthesizes resolution strategies to reduce cognitive overload for human decision makers.
Formalized industrial process states using normalized parameter vectors and dependency trees separating independent and dependent variables.
Employed a neural network classifier to forecast normal, abnormal, and emergency conditions across a lead interval τ.
Evaluated the method using a computational experiment on the Tennessee Eastman Process simulation model across 28 failure modes and 200 repeated simulations, synthesizing solutions via proximity graphs and evolutionary optimization.
The neural network classifier achieved an accuracy of 0.88 and a macro-averaged F1-score of 0.87 across a test set of 200 situations.
Topological analysis of precedent graphs across 50 synthetic emergency situations demonstrated stable metrics, with a mean vertex degree of 5.62 ± 1.18 and a closeness centrality of 0.43 ± 0.09.