Quantcast


CURATOR
A pinboard by
Siyu Tao

Research Assistant, Northwestern University

PINBOARD SUMMARY

In simulation-based engineering design optimization practices, computer simulation analyses/models (e.g., aerodynamics analysis) are typically run thousands of times for different designs (e.g., shapes of airplane wings) to find the optimal design that achieves the best performance (e.g., lifting performance). Because of the repetitiveness of simulation runs, only cheap simulation models can be used in this process so that the time cost is affordable, but their precision or fidelity is typically low. To enhance the fidelity of such simulation models, they are usually calibrated against simulation data from high-fidelity but expensive simulation models. We are developing a new calibration method that achieves higher fidelity of the calibrated simulation models than traditional methods. Our key idea is to introduce more calibration parameters in the cheap models by allowing a linear transformation of the model inputs. In this way, the cheap models have more flexibility in "learning" the high-fidelity data. Thus, their fidelity can be increased more significantly by our calibration method than by traditional ones. Ultimately, the cheap simulation models calibrated by our method can facilitate design optimization processes finding the optimal design with higher accuracy and efficiency. Since our method is developed within a general model calibration context, it can be applied to and benefit any engineering disciplines that involve model calibration processes.