Stopping Corrosion Before It Starts: Real-Time Multisensor AI Forecasting for Hydroprocessing Units

Abstract

Reactor-effluent piping in hydroprocessing units represents one of the most integrity-critical environments in modern refining. In Diesel Hydrotreating Units (DHTs), stabilized austenitic stainless steel SS321 is exposed to high-temperature hydrogen, steep thermal gradients, rapid depressurization, and mixed-phase flow transitions immediately downstream of the reactor. These coupled conditions promote hydrogen-assisted degradation and episodic thinning that evolve dynamically and often escape detection between periodic inspections. This paper presents a refinery-validated, multisensor artificial intelligence (AI) framework that forecasts metal-loss excursions 45–60 days in advance in the hot high-pressure loop of a 1.2 MMTPA DHT. High-frequency process data are fused with process-intelligence diagnostics and historical Ultrasonic Thickness (UT) measurements to generate forward-looking integrity predictions. Long Short-Term Memory (LSTM) neural networks are employed to capture temporal degradation behaviour, while XGBoost models quantify nonlinear interactions and rank process drivers. A novel Integrity Forecasting Index (IFI) is introduced to translate AI outputs into actionable integrity risk aligned with refinery Risk-Based Inspection (RBI) workflows. Forward validation against UT measurements from a future inspection cycle demonstrates strong agreement between predicted and measured thinning, achieving 93% predictive accuracy with a Mean Absolute Percentage Error (MAPE) of 6.8% and a 35% reduction in Root Mean Square Error (RMSE) relative to conventional rate-based methods. The framework identified multiple short-duration thinning accelerations that were not apparent from inspection intervals alone, enabling proactive mitigation and measurable operational benefit. The study demonstrates a practical pathway for transitioning refinery integrity management from retrospective inspection to predictive foresight using existing plant data infrastructure, aligning directly with digital transformation and sustainability objectives.