The paper presents a method for learning compact, expressive parameterisations for Bayesian shape optimisation by applying principal component analysis to deformations between existing designs. This approach yields a linear, interpretable search space where the number of components balances expressivity and dimensionality. Experiments on aerofoils, wings, and radio‑frequency cavities demonstrate improved sample efficiency and exploration beyond hand‑crafted baselines.
arXiv:2607. 09763v1 Announce Type: cross Abstract: Engineering shape optimization faces challenges in both expert-dependent problem setup and surrogate-model reliability.
arXiv:2609.36806v1 Announce Type: new Abstract: Modern engineering systems, from automobiles to aircraft, are designed by using precise, continuous parametric computer-aided design (CAD) models. Eval...
Generative machine learning provides an end-to-end framework for aerodynamic design optimization by generating diverse aerodynamic configurations conditioned on target performance. However, the genera...
CANTO is a transformer neural operator that directly maps continuous CAD geometry—represented as sequences of NURBS patches—to physical fields without requiring meshing. It tokenizes control points, knot vectors, and weights to predict continuous surface and volume fields at arbitrary query locations. Evaluated on automotive and aircraft aerodynamics benchmarks, CANTO achieves state‑of‑the‑art accuracy, notably a 19.8% reduction in surface‑pressure error on HiLiftAeroML, and enables gradient‑based inverse design, finding lower‑drag designs on AhmedML.
Optimizing 3D shapes within the latent spaces of deep generative models is fundamental to computer assisted engineering, yet remains prone to a critical failure mode we term manifold drift: the tendency of gradient-based optimization to move latent vectors away from the manifold of valid shapes. This problem is exacerbated in state-of-the-art 3D shape generative models that operate in increasingly high-dimensional latent spaces where valid shapes occupy a vanishingly small fraction of the full space.
arXiv:2507. 17786v2 Announce Type: replace Abstract: We introduce a reinforcement learning (RL) based adaptive optimization algorithm for aerodynamic shape optimization focused on dimensionality reduction.
SABLE is a Bayesian optimisation method designed for high‑dimensional latent spaces where the latent dimension far exceeds the evaluation budget. It uses a Bayesian affine surrogate that decomposes each query into an in‑span component optimized by expected improvement and an off‑span component governed by a scalar norm, with a finite‑horizon schedule to avoid under‑exploration. A recompression scheme reduces the number of pseudo‑observations to ≠O(M), keeping refitting costs manageable while achieving high sample efficiency and rapid decision times on synthetic and latent‑space optimisation tasks.
The paper introduces a joint contour location (jCL) method for actively learning input configurations that simultaneously achieve specified responses across multiple computer experiments. By employing two distinct acquisition schemes—one for exploration and one for exploitation—along with a decision rule, the approach balances learning across multiple response surfaces and provides a natural stopping criterion when no solution exists. The method is demonstrated with Gaussian processes, multitask GPs, and deep GPs, outperforming existing single-response contour location strategies and optimization-based alternatives.
arXiv:2607. 29225v1 Announce Type: new Abstract: Bayesian Optimization (BO) is widely adopted for data-efficient optimization in scientific and engineering applications, yet its computational cost is rarely evaluated alongside optimization performance.
arXiv:2603.06760v2 Announce Type: replace-cross Abstract: Control co-design jointly optimizes a robot's physical design and its controller. Existing techniques are limited in the number of model para...
arXiv:2608. 09938v1 Announce Type: cross Abstract: Hyperelastic deformations are highly sensitive to domain geometry and boundary conditions, making generalization across both a critical capability for neural operators applied to these problems.
The paper introduces Structural Affinity, an explicit regularisation technique for cage-based deformation (CBD) that couples an inner control cage to a coarser outer cage via a precomputed affinity matrix derived from mean‑value coordinates. This method directly influences cage‑handle motion, reducing self‑intersections, mesh distortion, and the need for hand‑crafted loss terms. Experiments on robotic morphology co‑design, neural deformation transfer, and aerodynamic wing optimization demonstrate improved surface smoothness, coherent cage distortion, and reduced surface roughness.