Expert System for Instrumentation Fault Diagnosis in Ethyl Chloride Separation Process

Abstract

The ethyl chloride separation process in the petrochemical industry relies heavily on the reliability of instrumentation equipment, as failures can lead to significant operational and economic losses. Conventional fault diagnosis remains subjective and time-consuming due to its dependence on expert availability. This study aims to develop a prototype expert system with a knowledge base encompassing 9 process units, 130 symptoms, and 128 possible instrumentation faults. The system employs a forward chaining inference method combined with a hybrid approach of Certainty Factor (CF) and Bayesian Inference to manage data uncertainty. The system was implemented in Python and validated through internal logic consistency analysis and sensitivity analysis. Logic consistency testing demonstrated that the system accurately validated individual rules with an actual CF score of 0.9000. In the ambiguity handling test case, the system correctly prioritized diagnoses based on the highest rule weight (CF = 0.9000) and highest Bayesian probability (0.088785). Sensitivity analysis confirmed that variations in user belief weight () proportionally shifted the primary diagnosis, for instance, the symptom 'decreasing heating ΔT' yielded 'Fouling' as the primary diagnosis with a posterior probability of 0.3027 and CF of 0.8900, while increased weight on temperature instability symptoms shifted the diagnosis to 'unstable reboiler steam supply'. The hybrid CF-Bayesian approach proved effective in supporting knowledge-based decision-making, even when operating with limited historical data.