Switched-fidelity modeling and optimization for multi-physics dynamical systems

With the advancement of computational power and modeling techniques, automotive and aerospace companies are beginning to utilize highly detailed models throughout the phases of system design and development. Often these systems consist of highly coupled subsystems that span mechanical, electrical, thermal, hydraulic, and pneumatic energy domains. Highly accurate models are typically developed for each individual subsystem, but are operated in isolation, thus ignoring the coupling between subsystems. This can prevent optimal operation at the system level. For large-scale systems, utilizing high-fidelity subsystem models for entire system simulations can be computationally expensive. As a result, lower fidelity models often replace the high-fidelity models at the expense of simulation accuracy. This paper presents a methodology for dynamically changing the fidelity of component models throughout a simulation to find an optimal balance between simulation speed and accuracy. This strategy is demonstrated for a finite-volume model of a vapor compression system where the model fidelity is based on the number of volumes used for the evaporator. Switched-fidelity modeling is shown to increase simulation speed by 64% from the baseline speed of the high-fidelity model, while reducing accumulated error by 69% for secondary flow exit temperature and 76% for primary flow exit pressure from the baseline of the low-fidelity model.