A hybrid AI approach for fault detection and diagnosis in a continuous distillation process

This study investigates the application of artificial intelligence (AI) for fault detection and diagnosis (FDD) in a continuous distillation process used to separate methylcyclohexane from a binary mixture with toluene. A hybrid approach was developed, combining Gradient Boosting (GB) for fault detection and Decision Trees (DT) for fault diagnosis. Process variables such as heating power, reflux ratio, preheating temperature, pressure drop, and boiler temperature were used as input parameters. The GB model achieved excellent fault detection performance, with an accuracy of 0.9881 and macro-averaged F1-score of 0.9980. While fault diagnosis via DT model, obtained an accuracy of 0.9964 and macro-averaged F1-score of 0.9979, demonstrating its ability to reliably classify different fault types. To enhance interpretability, SHAP (SHapley Additive exPlanations) analysis was applied to the DT model, revealing that Feed Flow Rate and Reflux Ratio are the most influential variables governing fault discrimination, consistent with the physical behavior of the distillation process. The results confirm the effectiveness of the proposed hybrid strategy, which combines high predictive accuracy with interpretability. This approach shows strong potential for real-time deployment in industrial environments, enabling proactive monitoring and supporting operator decision-making in automated distillation processes.