Renewable Fuels / Europe / 2022
A digital twin for thermochemical conversion research
An industrial R&D facility was constrained by the cost and duration of physical experiments. We coupled a kinetics-driven simulation engine to live sensor data, giving the team a validated model to explore reaction space before committing to a run.
Results
- 99.8%
- Model accuracy against measured reaction rates
- 75%
- Reduction in experimental cost
Disciplines
- Simulation
- Machine learning
- Instrumentation
Context
Each experimental run on the facility’s thermochemical conversion process consumed material, plant time and analyst attention. The research programme was therefore rate-limited by physical throughput, and the parameter space it could realistically explore was narrow — a constraint that shaped the science rather than the other way around.
Approach
We built a digital twin around the existing kinetics understanding rather than in place of it. A simulation engine driven by reaction kinetics provided the structure; real-time sensor data provided continuous validation, so that model drift was visible rather than assumed away. Reaction pathways and heat profiles could be tracked dynamically as conditions changed.
Where the kinetics were incompletely characterised, a learned component filled the gap, held to the same validation discipline as the rest of the model. We were explicit with the facility about which parts of the twin were mechanistic and which were fitted, because that distinction determines how far the model can be trusted outside the conditions it was trained on.
Outcome
The model reached 99.8% accuracy in predicting reaction rates and heat profiles within three months, and reduced experimental cost by over 75%. Physical runs were reserved for confirming candidates the twin had already identified, rather than for surveying the space.