Neural operators adapt to maintain accuracy in changing mechanics conditions
Adapting neural operators for mechanics decisions under changing operating conditions
Computational Engineering, Finance, and ScienceMachine Learning
Summary
Predicting how materials deform under different conditions is hard and slow using traditional methods. The authors tested a way to speed up these predictions using neural operators, which are a type of machine learning model. However, these models lose accuracy when conditions change beyond what they were trained on. By updating the neural operators with some new high-quality data collected during use, the authors showed the models could regain accuracy and make better decisions about how to control materials. Still, correctly judging when model predictions can be trusted remains a challenge.
What this means in practice
- •For mechanical engineers: Improve control decisions in systems involving deformable materials under varying conditions by updating predictive models with incoming high-fidelity simulations.
- •For manufacturing process designers: Incorporate adaptive neural models to speed up simulation-guided adjustments during production of magneto-active soft materials with changing load conditions.
Tested on simulated data.
Authors
Prashant K. Jha, Koffi Enakoutsa, Ian Galloway, Henry Anderson
Abstract
Neural operators can accelerate repeated nonlinear mechanics calculations, but their accuracy can deteriorate as operating conditions move beyond the training range. This work studies whether high-fidelity solutions acquired during use can be reused to adapt a neural operator and improve subsequent mechanics-based command selection. Two hard-magnetic soft-material systems are simulated using high-fidelity finite-element (FE) models, providing reference solutions for evaluating surrogate predictions and selected commands. A neural operator predicts deformation from known material, loading, and magnetic-field inputs, while an empirical error estimator determines which predictions may be used for command selection. Selected FE evaluations supplement these predictions, and their complete loading paths are retained for periodic updates of the neural operator and estimator. In both examples, the fixed operator loses substantial accuracy when stiffness and loading move outside the training range. Updates using 16 acquired paths recover much of the lost accuracy while preserving accuracy in the nominal regime. Under the same FE evaluation budget, the updated operators also improve command selection, although the benefit varies with the operating condition. Error estimation is less consistent, with inaccurate predictions sometimes accepted and accurate predictions rejected. These results demonstrate that reusing high-fidelity loading paths can extend the useful operating range of a neural operator. However, improved forward accuracy alone does not guarantee reliable prediction acceptance, highlighting prediction-specific error assessment as a separate requirement for trustworthy decision making.