Novel multivariate q-sigma rule focusing on process variation for incipient fault detection in dynamic processes

Abstract

When incipient faults occur in chemical processes, some variables will slightly deviate from original trajectories, and process residuals will gradually be continuously biased toward one side of their mean values, i.e., process variation will occur. Traditional indices are inadequately sensitive to this situation or achieve it at the cost of a high false alarm rate (FAR). To address this situation and explore methods with low FAR in dynamic processes, canonical variate residuals (CVRs) are generated. Then, a novel multivariate q-sigma (Mq-sigma) rule is proposed to monitor CVRs. It considers the process variation mentioned above in a window and sets the control limit for each variable. When tested on a simulated process, the Mq-sigma is highly sensitive to process variations and can detect incipient faults earlier than other methods, i.e., it has the lowest detection delay and FAR.