DIAGNOSIS OF PROCESS DISTURBANCES BY STATISTICAL DISTANCE AND ANGLE MEASURES

Disturbance and fault diagnosis techniques that rely on statistical me thods traditionally utilize distance based discrimination functions. C omplementary information is contained in the angular relations between data clusters representing process operations under various disturban ces. A novel disturbance diagnosis approach is presented based on angl e discriminants. The diagnosis method is successful in cases where dis tance based discrimination is not very accurate. The methodology is il lustrated by diagnosing various disturbances in the Tennessee Eastman process and compared with the diagnosis utilizing distance based algor ithms. (C) 1997 Elsevier Science Ltd.