Honeywell AI Pilot Aids Coker Unit Operations at TotalEnergies Refinery

Key Takeaways

• AI predicts steam pressure dips 10‑18 minutes early, 85% accuracy
• Drum cycling model reached 75% accuracy but not retained for planning
• Heater spalling predictions achieve 75% accuracy with two‑hour updates
• Enhanced alarm decision support deployed on 300+ coker tags, aiding novices
• Pilot aims to reduce flaring events and increase coker unit availability

Pulse Analysis

The refining sector has long wrestled with the tension between high‑throughput operations and the need for precise, real‑time control. Traditional process control relies on static setpoints and human intuition, which can lag behind rapid disturbances such as steam pressure drops. By embedding Honeywell’s Experion Operations Assistant into TotalEnergies’ delayed coker unit, the pilot introduces large‑language‑model‑driven analytics that surface predictive insights minutes before a fault manifests, giving operators a tangible head‑start.

Early results underscore the practical upside of AI‑enhanced control. The steam‑disturbance model now forecasts pressure dips up to 18 minutes in advance, boosting prediction accuracy to 85% and enabling proactive valve adjustments that curb flaring. Parallel models for drum cycling and heater spalling achieve roughly 75% accuracy, delivering two‑hourly updates that align with maintenance windows. The Enhanced Alarm Decision Support (EADS) system, applied to over 300 coker tags, translates historical alarm patterns into actionable recommendations, especially for junior operators who lack extensive on‑site experience. Collectively, these tools promise reduced unplanned shutdowns, lower emissions, and higher unit availability.

Beyond the immediate operational gains, the pilot signals a broader shift toward AI‑first strategies in heavy‑industry process plants. Successful integration requires close collaboration between control engineers and data scientists, as well as continuous model refinement to handle the plant’s dynamic conditions. If replicated across other units—such as fluid catalytic cracking or hydrotreating—the technology could redefine reliability standards, drive cost efficiencies, and accelerate the industry’s digital transformation agenda. Stakeholders watching the TotalEnergies‑Honeywell experiment will likely view it as a benchmark for scaling AI from experimental pilots to enterprise‑wide deployments.