A Continual Validation, Updating, and Decision-Making Framework for Self-Adaptive Digital Twins via Robust Model Predictive Control: A Case Study in Additive Manufacturing
The paper proposes a drift-aware digital twin loop that updates only a small slice of the surrogate model, then tests whether the update actually helped before deployment.
The framework uses Fisher score vectors to detect distribution shifts in the twin’s inputs and confidence. When drift is found, it applies LoRA fine-tuning to fewer than 1% of parameters. A Mann-Whitney U test is used online to validate predictive improvement before the new surrogate is accepted. The authors report short detection delays and restored accuracy and uncertainty estimates in stochastic-system and additive-manufacturing case studies. ArXiv · AI/CL/LG's note
The framework uses Fisher score vectors to detect distribution shifts in the twin’s inputs and confidence. When drift is found, it applies LoRA fine-tuning to fewer than 1% of parameters. A Mann-Whitney U test is used online to validate predictive improvement before the new surrogate is accepted. The authors report short detection delays and restored accuracy and uncertainty estimates in stochastic-system and additive-manufacturing case studies. ArXiv · AI/CL/LG's note
score 4