Monitoring a segmented fluid bed dryer by hybrid data-driven/knowledge-driven modeling

Many data-driven and knowledge-driven methods for process monitoring have been developed in the last decade. In this study we show that the combined use of techniques from both categories can potentially outperform their standalone use. The proposed hybrid approach for fault detection and diagnosis is grounded in conventional multivariate statistical process monitoring. However, the datasets subject to analytics include not only field measurements, but also data obtained from a state estimator based on a mathematical model of the process. We apply the proposed methodology to a pharmaceutical case study, using the mechanistic model of a segmented fluid bed dryer from gPROMS FormulatedProducts. The hybrid framework demonstrates improved fault detection and diagnosis performances, when compared to data-driven monitoring or state estimation taken in isolation.