Multi-Fidelity Bayesian Neural Network for Uncertainty Quantification in Transonic Aerodynamic Loads
Andrea Vaiuso, Gabriele Immordino, Marcello Righi, Andrea Da Ronch
TL;DR
Addressing uncertainty in transonic aerodynamic loads with multi-fidelity data, the paper develops MF-BayNet, a transfer-learning-enabled Bayesian neural network framework. It pre-trains on low-fidelity data, then sequentially fine-tunes on mid- and high-fidelity data, using Monte Carlo sampling to obtain predictive means and uncertainties. The authors show MF-BayNet outperforms Co-Kriging on $C_L$ and $C_M$ with about half the error and tighter uncertainty bounds, validating the approach for aerospace applications with limited high-fidelity data. The work contributes a transparent, open-source framework for uncertainty-aware multi-fidelity learning in transonic aerodynamics.
Abstract
Multi-fidelity models are becoming more prevalent in engineering, particularly in aerospace, as they combine both the computational efficiency of low-fidelity models with the high accuracy of higher-fidelity simulations. Various state-of-the-art techniques exist for fusing data from different fidelity sources, including Co-Kriging and transfer learning in neural networks. This paper aims to implement a multi-fidelity Bayesian neural network model that applies transfer learning to fuse data generated by models at different fidelities. Bayesian neural networks use probability distributions over network weights, enabling them to provide predictions along with estimates of their confidence. This approach harnesses the predictive and data fusion capabilities of neural networks while also quantifying uncertainty. The results demonstrate that the multi-fidelity Bayesian model outperforms the state-of-the-art Co-Kriging in terms of overall accuracy and robustness on unseen data.