An efficient polynomial chaos expansion strategy for active fault identification of chemical processes

Abstract

This paper is concerned with a highly efficient active fault detection and isolation (FDI) framework. An auxiliary, fault-revealing input is derived by solving an optimization problem. As we implement a model-based approach, the active FDI framework is robustified against model parameter uncertainties, including parameter correlations which are common for experimentally derived parameters. Moreover, critical safety limits are considered, and an optimal process performance is fulfilled in parallel. In this work, which is an extension to our ESCAPE-2017 contribution, a novel highly effective polynomial chaos expansion (PCE) approach is used to address parameter uncertainties and to include process design parameters directly. To reduce the computational load, we combined the PCE with a least angle regression (LAR) strategy. The overall effectiveness of the novel one-shot sparse polynomial chaos expansion (OS2-PCE) concept is demonstrated by analyzing a tubular plug flow reactor illustrating the need for uncertainty and parameter correlation analysis in FDI while ensuring an optimal and safe process operation, respectively.