Summary
Computer simulations can take a long time, so people use neural networks as faster guessers called neural surrogates. But when designing things, it's important to know how uncertain these guesses are, especially when the shapes or conditions change a lot. The authors studied different ways to measure this uncertainty in neural surrogates across several big engineering problems like car aerodynamics and crash safety. They found each method can identify where predictions might be wrong and respond well to new, unfamiliar situations, but the best method depends on the specific task and data. This means engineers should choose their uncertainty measurements based on what decisions they need to make.
What this means in practice
- •For automotive design engineers: Use uncertainty estimates from neural surrogates to safely guide aerodynamic and crash performance predictions during vehicle design.
- •For aerospace simulation teams: Incorporate geometry-conditioned uncertainty measures in surrogate models to improve reliability of external aerodynamics assessments.
Authors
Kaustubh Tangsali, Mohammad Amin Nabian, Kelvin Lee, Carmelo Gonzales, Sanjay Choudhry
Abstract
Neural surrogates can substantially accelerate computer-aided engineering (CAE) workflows, but their use in design requires uncertainty estimates that remain meaningful across varying geometries, spatial prediction fields, and engineering quantities of interest. We investigate how established uncertainty quantification (UQ) approaches behave when adapted to geometry-conditioned neural surrogates. We compare one closed-form and two sampling-based approaches-a Gaussian process (GP)-based method, concrete Monte Carlo (MC) dropout, and deep ensembles-and evaluate them on three large, industry-relevant CAE datasets for external aerodynamics and crash dynamics. We examine whether predicted uncertainties have credible magnitudes, identify locations with larger prediction errors, respond to unfamiliar inputs, and remain informative for derived engineering quantities. On the DrivAerStar dataset, where all three methods are compared, each generally assigns higher uncertainty to locations with larger prediction errors, and validation-based rescaling brings interval coverage close to nominal on a disjoint in-distribution test set. Results on AirFRANS and automotive crash also show useful error ranking and interval estimates, but the relative performance of the methods changes with the dataset and evaluation criterion. UQ methods and evaluation metrics should therefore be selected based on the intended downstream CAE decision.