SUMMARY

The design of strongly coupled multidisciplinary engineering systems is challenging since it is characterized by the complex interaction of different disciplines. Such complexity cannot be easily captured by explicit analytical solutions. In this situation, the deficiency motivates the development of surrogate modeling that enables the prediction of the systems’ behavior without analytical solutions. Between exiting surrogate modeling techniques, machine learning has gained significant interests because of the flexibility of non-linear formulation and applicability to data-driven analysis. Notably, a deep surrogate model augments the precision of prediction and estimation of system behavior once image-based inputs that represent physical experiment and simulation are employed. Nevertheless, the feasibility of the deep surrogate model is often flawed due to massive training costs and complicated network structures.

To address those issues, therefore, in the proposed research, physics-informed artificial images (PiAI) will be formed and utilized to train deep neural networks in strongly coupled multidisciplinary domains. To further improve the credibility of deep surrogate models, an optimized stochastic calibration process is proposed to match data distributions of training and test sets. The calibration process can resolve the covariate shift that is occurred when the data distributions are heterogeneous. Lastly, multi-fidelity deep structures will be introduced to enhance computational productivity. Furthermore, the proposed examples will demonstrate the efficacy and applicability of the proposed framework in the practical engineering design application.