Fault detection based on polygon area statistics of transformation matrix identified from combined moving window data

摘要

Principal component analysis (PCA) has been widely used in monitoring industrial processes, but it is still necessary to make improvements in having a timely and effective access to variation information. It is known that the transformation matrix generated from real-time PCA model indicates inner relations between original variables and new produced components, so this matrix would be different when modeling data deviate due to the change of the operating condition. Based on this theory, this paper proposes a novel real-time monitoring approach which utilizes polygon area method to measure the variation degree of the transformation matrices and then constructs a statistic for monitoring purpose. The on-line data are collected through a combined moving window (CMW), containing both normal and monitored data. To evaluate the performance of the proposed method, a simple numerical simulation, the CSTR process and the classic Tennessee Eastman process are employed for illustration, with some PCA-based methods used for comparison.