Machine learning‐based risk assessment of heavy fuel oil leakage during handling operations: A case study of a thermal power plant

Abstract Heavy Fuel Oil (HFO) handling in power plants poses significant safety and environmental risks, yet conventional risk assessment tools lack real‐time applicability. This study addresses this gap by integrating Hazard and Operability (HAZOP) analysis, computational simulations, and machine learning (ML) to predict HFO leakage consequences at the HUBCO Power Plant, Narowal. A HAZOP study identified 13 critical deviations, with tank leakage as the most severe hazard due to its potential for toxic exposure, environmental contamination, and operational disruptions. Quantitative simulations using the Process Hazard Analysis Software Tool (PHAST) modeled 2757 scenarios, revealing that large leaks (≥500 mm diameter) under elevated temperatures (40–50°C) and moderate wind speeds (4–6 m/s) generated toxic vapor concentrations exceeding 500 ppm at 60 m. To enable real‐time prediction, an Artificial Neural Network (ANN) surrogate model was trained on PHAST data, achieving high accuracy ( R 2 = 0.9667–0.9670) in estimating vapor dispersion at 60 m and 200 m distances. The ANN reduced computational latency from 5 to 15 min (PHAST) to 3–5 ms, indicating scalability for dynamic risk assessment. While slight heteroscedasticity was observed in high‐concentration predictions, the model exhibited robust generalization, particularly at 200 m. This hybrid framework bridges the gap between physical modeling and rapid decision‐making, presenting a proactive strategy for industrial safety. The study underscores the transformative potential of ML in enhancing hazard prediction, operational resilience, and environmental stewardship in HFO‐dependent facilities.