A first principles approach to automated troubleshooting of chemical plants

Abstract A prototype first principles-based troubleshooting system capable of analyzing the dynamic behavior of individual process units or small subsystems is presented. Our system differs from previous qualitative model-based systems in two ways. Firstly, it reasons with a first principles understanding of physical and chemical phenomena such as reaction, heat flow and liquid/vapor equilibrium, allowing analysis and explanation at a more fundamental level. Secondly, it can automatically modify unit models, allowing identification of faults which significantly change unit behavior without the need for explicit fault scenarios. The troubleshooter uses the qualitative process theory (QPT) representation of Forbus which provides a framework to qualitatively model and simulate the behavior of dynamic systems. Our diagnostic system exploits QPT's explicit representation of the conditions under which phenomena are believed to hold and the constraints they contribute to the model. The system takes as input the assumed unit model and a set of discrepancies between the observed and expected unit behavior. Diagnosis is based on a hypothesis—test strategy of four steps: identify assumption changes which can account for a single discrepancy, determine the resultant changes in the model, simulate the effect of these changes on the unit behavior and compare with observed behavior.