Adverse Condition and Critical Event Prediction in Cranfield Multiphase Flow Facility

Today's complex processes and plants are vulnerable to different faults, misconfiguration, non-holistic and improper control and management which cause abnormal behavior and might eventually result in poor and sub-optimal operation, dissatisfaction, damage to the plant, to personnel and resources, or even to the environment. To cope with these, adverse condition and critical event prediction plays an important role. Adverse Condition and Critical Event Prediction Toolbox (ACCEPT) is a tool which has been recently developed by NASA to allow for a timely prediction of an adverse event, with low false alarm and missed detection rates. While ACCEPT has shown to be an effective tool in some applications, its performance has not yet been evaluated on practical well-known benchmark examples. In this paper, ACCEPT is used for adverse condition and critical event prediction in a multiphase flow facility. Cranfield multiphase flow facility is known to be an interesting benchmark which has been used to evaluate different methods from statistical process monitoring. In order to allow for the data from the flow facility to be used in ACCEPT, methods such as Kernel Density Estimation (KDE), PCA-and CVA-based contribution plots, two-sample Kolmogorov-Smirnov test are used. The ACCEPT results are compared with those obtained from Canonical Variate Analysis (CVA) which has been used for the same test bed. CVA is known to be one of the best multivariate data-driven techniques in particularly under dynamically changing operational conditions. The results are evaluated and discussed.