Faulty variable selection based fault reconstruction for industrial processes

Fault reconstruction has been widely applied for fault diagnosis recently, which, however, treats the fault data as a single subject without analyzing the specific effects of different variables. To address this problem, a faulty variable selection method is proposed which can extract more meaningful discriminant directions and probe into the specific faulty variables. By pairwise performing nested-loop Fisher discriminant analysis (NeLFDA) algorithm on normal and fault process data, some projection directions that are useful for classification between normal and fault cases are extracted. Then along these directions, an iterative faulty variable selection procedure is designed by evaluating ratio of variable contribution to the fault variations so that process variables that are significantly fault-relevant are identified and distinguished from those general variables. Based on variable selection results, fault reconstruction model is developed for faulty variables, which is used for fault diagnosis. Online fault diagnosis is then performed by dually checking the characteristics of fault samples for faulty variables and general variables. Its performance is illustrated with the industrial process data from the cut-made process of cigarette.