Error-aware digital twin of a thermal fin in heat sink systems

For effective thermal management of electronic devices, we introduce a digital twin of a heat sink system benchmarked on a thermal fin. The digital twin updates its thermal state through inverse identification, based on sensor measurements of a fin. Because inverse identification requires repeated runs of simulations, traditional Finite Element (FE) model becomes computationally intractable for this task. Instead, we use a Reduced Basis (RB) model which achieves higher computational efficiency once trained. For accurate identification, we propose an error-aware digital twin that accounts for RB model errors during the identification process, where the errors are estimated from Kriging model trained on discrepancy data between FE and RB models. Numerical experiments demonstrate that the error-aware RB-based digital twin achieves higher computational efficiency than the FE-based digital twin, while providing higher estimation accuracy than the RB-based digital twin. After the state estimation, digital twin updates its thermal field based on the identified state, which is monitored using FE simulation. Based on the maximum temperature of the monitored field, digital twin applies control actions when necessary. Thanks to the error-aware RB-based digital twin, appropriate real-time control action can be applied based on the reliable state estimation.