Covestro
AI-run autonomous production line
Covestro has let an AI model run a whole production line in Dormagen without operator input since June 2024
Problem
In chemical plants, operators set process parameters by hand, and the best balance of reaction time, energy use and product specification depends on their experience.
Approach
Covestro trained a small open-source algorithm on data from one line at its Dormagen site in Germany, without a pretrained model. Initially staff applied the AI's suggested parameters; once these proved consistently right, operators proposed letting it run the line. Since June 2024 it has run from production planning to finished product without operator input, while people on site can intervene. The company says it has extended AI to production sites in Germany, Spain, China and the United States.
Results
No figures published
Takeaway
Control was handed over in stages, after a period in which the AI only advised and operators checked its suggestions, and the push for autonomy came from the operators. Covestro has published no figures on energy, quality or throughput for the line, only qualitative claims.
Categories
- Size
- Large enterprise
- Industry
- Chemicals and materials
- Business function
- Production and quality
- Innovation type
- Process innovation
- AI technology
- Predictive analytics and MLOptimization
- Adoption
- Built in-house
- Evidence
- Company's own claim
Sources
Source summaries
An interview with Covestro's CIO about a line in Dormagen that AI has run for several hundred days. He explains that the model is a small open-source algorithm trained on that plant, that it first made suggestions staff implemented, and that operators proposed letting it run alone. He cites optimal reaction times, low energy use and on-spec products but gives no figures.
Covestro's 2025 annual report states that an entire Dormagen production line has worked fully autonomously since June 2024, from planning to finished product, and that the AI rollout has reached sites in Germany, Spain, China and the US. It also mentions AI-optimised liquidity management that cut interest costs by several million euros and an AI academy with about 100 participants. No production figures are given.
Full details are in the original sources.