A numerical-structural approach to failure detection and isolation in complex plants

The detection and isolation of plant failures is based on the parallel statistical testing of residuals, generated from an input-output model. The primary model is transformed to yield a great variety of secondary variants. Structural constraints are applied to attain isolability. Measures of sensitivity and robustness are introduced and used to seek the best variants over a finite discrete set of models. Design strategies are outlined that include the on-line selection of models and require heuristic search techniques.