Honeywell previews Experion Cognition release

The landscape of industrial process control is standing on the precipice of a cognitive revolution. For decades, distributed control systems have managed steady-state operations, allowing operators to oversee hundreds of process variables simultaneously. However, when operations deviate from the norm, the burden falls squarely on human shoulders that are increasingly burdened by labor shortages and rising process complexities.

At the 2026 Honeywell Users Group in Phoenix, Colin Hams, offering manager at Honeywell, provided an in-depth preview of Experion Cognition. Slated for release later this year, Experion Cognition represents a new generation of automation that combines analytics, generative human-machine interface (HMI) and small language models (SLMs) in an on-premise infrastructure. The Experion Cognition platform aims to capture human expertise, to digitalize decision-making and to provide operators with a reliable digital companion for guidance.

Operational performance challenges

Hams opened his presentation with an interactive counting exercise designed to demonstrate the psychological phenomena of selective attention. Just as audience members missed an obvious gorilla while focusing on counting people passing a basketball, control room operators face identical cognitive bottlenecks during peak stress events.

"Our human brain is limited in capacity, so we naturally filter out things we perceive as irrelevant, so that we can focus on the task at hand," Hams explained. "While today's control systems are really good at handling steady state, we still rely on human expertise for those conditions that are not steady state."

Plants currently grapple with several interconnected human-factor challenges:

• Knowledge gaps: Approximately 40% of the current industrial workforce is projected to retire within the next decade. As senior personnel depart, they leave a massive knowledge vacuum.
• Shift-to-shift fluctuations in performance: Because of varying experience levels among operators, plants suffer a 10% to 25% variability in performance between shifts.
• Cognitive attention overload: While operators can supervise hundreds of variables during steady-state conditions, human cognitive limitations restrict them to actively monitoring only three to five items during a startup or abnormal event. Manual interventions can spike from a few times an hour to hundreds, leading to mental fatigue and missed cascading process issues.
• The cost of incidents: Human factors contribute to 40% of abnormal situations and unplanned shutdowns. Even when standard operating procedures (SOPs) exist, abnormal situations still result from human execution errors. The financial penalty is steep: a single unplanned industrial outage can cost more than $2 million.

"We all know these challenges aren't new; they've been around for years," Hams noted. "We've effectively achieved the limits of what we can do with previous technology."

Experion Cognition

To break through these cognitive and experiential barriers, Honeywell developed Experion Cognition. Hams described the platform as an evolution of operational autonomous capabilities brought together into a single, unified offering. It is built on four central pillars:

• digitalizing human decision-making,
• delivering a generative HMI for enhanced human-autonomy teaming,
• deploying an agentic AI architecture, and
• providing a streamlined knowledge-capture engine.

The platform targets four primary operational outcomes:

• bridging the retirement knowledge gap with best-practice guidance,
• eliminating shift-to-shift variability,
• reducing unplanned shutdowns via proactive anomaly detection, and
• boosting overall operator efficiency by automating routine tasks.

A key differentiator of Experion Cognition is its modular, containerized foundation known as the App Hive. Recognizing that artificial intelligence is evolving faster than traditional control system lifecycles, Honeywell has isolated the cognitive layer from the core, deterministic automation layer.

"Industrial process control is really in the very early adoption stages of looking at artificial intelligence. We need a platform where we can deploy and use new capabilities as they're released without impacting our underlying base release and without impacting your control system," Hams said. "The App Hive allows you to unlock new capabilities faster by providing a modular, container-based platform."

Seamless Integration with Experion PKS

A primary concern for plant managers adopting AI tools is the complexity of implementation and the risk of destabilizing existing infrastructure. Hams emphasized that Experion Cognition acts as a secure, on-premise companion system that integrates seamlessly into existing Honeywell environments, specifically targeting Experion PKS Release 530 (R530) and Release 600 (R600).

The software requires no new station node types. For R530 systems, a server and station update via Patch 3 enables integration, while R600 systems are native-ready. From an operator's perspective, the unified HMI obeys all existing workspace rules and adheres to strict design principles.

To shield core control systems from heavy computational burdens, the system runs on dedicated Level 3 hardware, although it can run on Level 2. The topology features an Experion Cognition server utilizing distributed server architecture (DSA) to communicate across multiple clusters, the App Hive node, a Knowledge Base node and a physical GPU node to host the SLM.

The system allows closed-loop, autonomous action, such as automatically hot-swapping a drifted sensor value with a calculated fallback, while maintaining every critical safeguard. It operates fully isolated from the internet. "Experion Cognition does not require any cloud connectivity. It's an on-prem solution," Hams emphasized. "It protects your IP, and it maximizes your cybersecurity."

The autonomous journey

At the heart of Experion Cognition’s autonomous journey is situation handling, executed by a suite of intelligent software agents. These include:

• the sensor malfunction agent, built to flag hidden anomalies like frozen or drifting tags that remain within legal limits and evade standard alarms,
• the process anomaly agent, which leverages machine learning to parse historical data for complex upstream/downstream deviations, and
• the process condition agent, which is designed for deterministic, rule-based logic.

To populate these agents, the platform's knowledge-capture tool leverages an SLM to read text-based SOPs and translate them into automated work instructions. If no written procedure exists, engineers can interview the system, and the AI will prompt them to map out the logic.

Honeywell is mitigating the risk of AI erraticism in high-hazard environments. "What's really important here is we're not generating this on the fly at runtime, because we all know AI can be great, but it can be flawed and hallucinating problems," Hams clarified. "Every situation is verified by your process experts before being available to operators."

Though the full platform launches later this year, components of its agentic architecture have already achieved validated success in the field under early iterations and pilot programs, including:

• Chevron, which has deployed alarm guidance and improved response times and seen more accurate and consistent handling,
• TotalEnergies, where predictive alerts are anticipating steam disturbances and avoiding delayed coker trips, and
• ADNOC Gas, where sensor malfunction and process anomaly agents were trialed late last year and demonstrated potential for reducing unplanned downtime.

Experion Cognition establishes the foundation for incremental autonomy. Plants can start by deploying deterministic rule-based configurations and later scale into complex machine-learning (ML) models as historical data populates the App Hive. By validating automated actions through human experts, Honeywell's Experion Cognition platform offers a predictable, safe and highly scalable path toward fully autonomous operations, said Hams.