Evonik Oxeno deploys AI-based process optimization in live plants
- COMPAS uses more than one million pre-calculated simulations to train gray-box AI models that encode thermodynamics and kinetics.
- Models produce realistic plant-control recommendations within seconds and speed decisions when feedstock streams change.
- Technology is running in live operations at Evonik Oxeno sites in Marl and Antwerp and has reduced energy consumption and CO₂ emissions.
- Methodology published in Chemie Ingenieur Technik and Chemical Engineering & Technology; team is expanding applications and seeking patent protection.
Project and methodology
Evonik Oxeno’s COMPAS project (Control and Optimization using AI Models for Process Advancement and Cost Saving) uses a library of more than one million pre-calculated process simulations to train so-called gray-box models. These models combine programmed thermodynamics and kinetics with AI components to reproduce process behaviour and generate optimisation recommendations.
Deployment and operational impact
The AI models have been integrated into live plant operations in Marl and Antwerp. They deliver realistic control recommendations within seconds, help operators respond faster to changes in feedstock streams, and have reduced the energy required to produce the targeted product, thereby lowering the plants’ CO₂ footprint. Work is under way to automate transfer of AI model results into the plant control system.
Validation, transferability and protection
The COMPAS methodology and findings have been published in the journals Chemie Ingenieur Technik and Chemical Engineering & Technology. The team is extending the application scope across other chemical and petrochemical processes and is seeking patent protection for the technology as it progresses toward a self-optimizing C4 Verbund.
Source: Evonik