Honeywell’s predictive AI proves itself at TotalEnergies

The delayed coker unit (DCU) at TotalEnergies’ Port Arthur refinery's is a complex system that thermally cracks heavy residual oils into lighter, more valuable oil products and petroleum coke.

As Frederic Robert, TotalEnergies’ senior engineer responsible for instrumentation, analyzers, integrated control and safety systems, explained during his presentation at the 2025 Honeywell Users Group EMEA event, the DCU faced persistent operational headaches. Frequent power fluctuations disrupted steam generation systems, while pressure disturbances triggered compressor trips that forced plant shutdowns.

These incidents resulted in excessive flaring, regulatory fines, increased operational costs and compromised safety. Compounding these issues was the fact that operators lacked sufficient time to perform strategic load shedding of non-critical equipment when such problems arose.

Getting help from AI

A pilot project between TotalEnergies and Honeywell sought to bring Honeywell’s machine learning-powered HALO (Highly Augmented Lookahead Operations) Operator Advisor analytics into the control room at the Port Arthur Refinery. By gathering data from the refinery’s distributed control systems and presenting the resulting insights through dashboards, the technology gave operations managers and supervisors new levels of visibility into performance patterns and improvement opportunities.

The project focused on three critical areas to the DCU’s operations:

Steam Disturbance Prediction. Using Honeywell's Experion Operations Assistant (a feature of the HALO system), the system successfully provided notice of five pressure dip events with an average 12-minute advance warning, which was well within the desired target range of 10-15 minutes. This advance notice enabled operators to proactively shed non-critical processes, preventing shutdowns of the steam turbine-driven compressor and ensuring continuous DCU operations.

Drum Cycling Prediction. This addressed management of coke drum processing as they cycle between being filled with heavy oil and being cleared of residual coke before being filled again. Since coking rates vary with pressure, temperature, composition and feed rate, operators needed reliable alerts about drum readiness. HALO's hourly modeling provided cycle status updates across the operation’s 24-hour drum cycles, delivering warnings when drums fell out of sync or decoking cycles experienced delays.

Spalling Prediction. Here, the project looked to tackle the manual intervention required when coke buildup is steam-blasted from feed coils. Though the HALO technology was tested on only one spalling event at TotalEnergies’ DCU, the prediction proved accurate, providing early alerts that allowed supervisors to coordinate operator availability and resources without impacting production.

In addition to the HALO predictions testing, the project implemented Honeywell’s Enhanced Alarm Decision Support (EADS) to address information overload in the control room. New operators often struggled to identify evolving situations they'd never encountered. EADS uses historical data, alarm logs and process diagrams to provide guidance on previous alarm occurrences and proven response actions, essentially capturing institutional knowledge in an accessible format.

Lessons learned and next steps

Robert noted that key to this project’s success was the combination of TotalEnergies' process expertise with Honeywell's technological capabilities. Early involvement of operators and supervisors proved essential, Robert said, as their frontline experience shaped practical, usable insights that drove feedback throughout the project.

The team identified clear pathways for improvement at the refinery based on the project’s results. For example, steam loss predictions could benefit from extended 15- to 25-minute warning windows and predictions could be improved by incorporating external factors, such as weather patterns, into the AI analysis. Robert added that this also held true for drum cycling predictions, which could be improved by adding external variables into the analysis, such as severe weather, equipment failures in connected units and ongoing maintenance activities. For the spalling process, TotalEnergies is looking to expand the model used here to analyze process efficiencies for additional operational value.

TotalEnergies plans to complete pilot evaluations for all three prediction applications this year, while further implementing and assessing the alarm guidance use case. Robert pointed out that the project’s success at the DCU has generated interest in extending these Honeywell technologies to other Port Arthur Refinery units and potentially other TotalEnergies’ facilities globally.

All in all, Robert said this project proved to TotalEnergies how Honeywell’s industrial AI technologies move beyond theoretical benefits to deliver measurable operational improvements, such as preventing shutdowns, reducing emissions and empowering operators with actionable intelligence.