On solving Fault Detection Problem and Risk Estimation Monitoring with Deep Neural Networks and Postprocessing

Abstract

In this study, we consider fault prediction problem and production process risk monitoring based on observational data. We consider case, when there are no variables, by which one could classify the situation preceding to the fault. We propose an approach that is based on a specific auxiliary risk variable and modifications of the modeling accuracy estimation criterion, so the fault detection problem is reduced to supervised learning problem. We use deep learning and examine different model architectures. Trained model produces the risk estimations for new observations, then we use postprocessing to interpret the estimations to decision-maker. This work confirms that data-driven risk estimation can be integrated into digital services to successfully manage plant operational changes and support plant prescriptive maintenance. This was demonstrated with data from a commercial circulating fluidized bed firing various biomass and residues but is generally applicable to other production plants.