Case study · With Merck KGaA
Self-driving perfusion development with an autonomous digital twin
A study performed with Merck KGaA connected a hybrid-model digital twin directly to a 24-reactor ambr®250 perfusion system to create an autonomous, self-driving system.
-
<0.5
Prediction error (rRMSE) of the 3-day forecast of perfusion dynamics, for all nine bioprocess variables
-
+50%
Higher VCV with the digital twin vs. standard operation
-
The autonomous digital twin generally outperformed standard operation
The challenge
Automated hardware, manual decisions
Perfusion development on high-throughput systems like the ambr®250 is highly automated at the hardware level, yet scientists are still asked to make every decision. Can a digital twin take over those decisions?
The approach
Autonomous operation vs. standard operation
Twenty-four ambr®250 perfusion runs were designed and executed to train a hybrid model on four bioprocess parameters: perfusion rate, temperature shift, agitation and pyruvate feed. In a second campaign, the digital twin was connected directly to the bioreactors and operated six of them autonomously, each with its own objective, on both the training clone and a second clone the model had never seen. Each day it read the latest data, retrained, optimized setpoints and applied them with no human in the loop, while reactors under standard operation ran alongside as a benchmark.
- 24 ambr®250 perfusion runs
- 4 bioprocess parameters
- 6 reactors run autonomously
Get the full results
Download the full case study, with project outline, analysis, and case results.