Efficient faulty variable selection and parsimonious reconstruction modeling for fault diagnosis

In the present work, an efficient faulty variable selection algorithm is proposed by developing iterative selection procedure. The significant faulty variables that cover the most fault effects and thus highly contribute to alarm monitoring statistics are distinguished from those general variables that are deemed to follow normal rules and thus uninformative to reveal fault effects. Then to further reveal the fault characteristics, the selected significant faulty variables are chosen to get parsimonious reconstruction model for fault diagnosis where relative analysis is performed on these selected faulty variables to find the relative changes from normal to fault. The faulty variable selection can not only pick up those responsible variables but also exclude the influences of uninformative variables and thus more effectively explore fault effects. It can also help to find more interesting and reliable model representation, better identify the underlying fault information and get enhanced process understanding. Its feasibility and performance are illustrated with simulated faults using data from the Tennessee Eastman (TE) benchmark process.