Work

Constraint Based Diagnosis and Recovery Assessment for Interconnected Chemical Processes

Year: 2026
Type: article
Abstract: Diagnosis and recovery in an interconnected process require compatible representations of fault evidence and feasible interventions. We specify a constraint-based architecture, Constraint-based Diagnosis and Recovery (CDDR), that uses a signed process model to generate and rank diagnostic candidates and a steady-state model to assess recovery options. Inclusion-minimal diagnoses are distinguished from minimum-cardinality candidates, and recovery feasibility is treated as an operational assessment rather than a test of whether a diagnosis is true. A reported synthetic campaign comprises an eight-unit plant, 18 measured variables, 16 constraints and 1,000 episodes across five seeds. Reported top-1 unit-plus-fault-type accuracy is 0.573 for CDDR, 0.475 for univariate threshold monitoring, 0.246 for causal reachability and 0.149 for Principal Component Analysis (PCA) contribution analysis. CDDR has a reported false-alarm fraction of 0.015. Recovery validity is reported as 0.5556 for CDDR, 0.3788 for an alarm-driven comparator and 0.8430 for an oracle, with different scoring denominators. These are descriptive results pending verification of episode records, scoring rules and analysis code. CDDR performs poorly on fouling drift (top-1 0.011) and has the lowest accuracy at doubled observation noise (0.033). The campaign does not validate optimal intervention selection, transient safety, performance on the Tennessee Eastman (TE) process or industrial deployment.
Language: English
Open Access status: hybrid
OpenAlex ID: W7225888377
FWCI:
Cites:
Cited by:
Related to: 10