화학공학소재연구정보센터
Journal of Process Control, Vol.95, 75-85, 2020
Generalized moving variance filters for industrial alarm systems
Accurate and rapid detection of variation changes in process variables is of paramount importance to the safety and proficiency of process industries. Moving variance filters detect such variation changes by tracking the variance of the last N samples of a process variable. However, no explicit formulation is available for performance analysis, and design of these filters. We extend these filters by considering the 'generalized variance', which is the same as the conventional variance, but the filter terms are weighted differently. Next, we propose an analytical framework to analyze the filter performance. We prove that the conventional filter indeed is the optimal configuration if only the detection accuracy is considered. But in the case study, via a counter-example, we show that this statement does not hold if we also take the detection delay index into account. In this case, our result gives a straightforward and intuitive measure to obtain the filter accuracy as a function of filter coefficients. This can also be used as a part of the cost function when designing these filters considering multiple criteria. Through a case study on the Tennessee Eastman process, we show that compared to conventional filters, generalized filters make it possible to detect abnormality faster for the same degree of accuracy. (C) 2020 Elsevier Ltd. All rights reserved.