arXiv subjects

Cashen Diniz

Publications and source records attributed to Cashen Diniz.

5 recordsLinked to original sources

arXiv2026

Transferable Low-Dimensional Representations of Aircraft Surface Fields

Analyzing flow fields on aerodynamic surfaces is critical to designing next-generation aircraft. Raw computational representations are high-dimensional and costly for complex three-dimensional bodies, while conventional low-dimensional representations apply only to the systems from which they derive. We introduce an approach that learns fluid flow representations from simple two-dimensional geometries and transfers them to complex real-world three-dimensional aircraft. Comparing the proposed Least Volume autoencoder (LVAE) to proper orthogonal decomposition (POD), we show both yield compact 28-41 dimensional representations of two-dimensional flow fields, with the nonlinear LVAE producing more accurate in-domain reconstructions at matched dimensionality. These models transfer zero-shot to three-dimensional systems, and counterintuitively the linear POD transfers more accurately than LVAE. Applied to extruded wings, POD-reconstructed lift and drag remain within 0.1% and 0.7% of true values, while transfer to complex blended wing body aircraft preserves lift and drag to within 0.3% and 0.7% without any three-dimensional training data. We further show that three-dimensional flow patterns missed by the pretrained model can be isolated to train a disentangled 3D latent space via a dual encoder-decoder transfer learning strategy, performing accurately in previously inaccessible data-scarce regimes. Using only five blended wing training cases, mean squared reconstruction error drops by 5x relative to the 2D pretrained models and 70x relative to models trained from scratch. These results establish that transfer learning from readily available two-dimensional datasets unlocks compact, interpretable representations of complex three-dimensional aircraft surface flows, enabling design insights in data-starved early phases where conventional approaches require hundreds to thousands of simulations.

physics.flu-dyn↗

arXiv2025

OptiWing3D: A Diverse Dataset of Optimized Wing Designs

OptiWing3D is the first publicly available dataset of high-fidelity shape optimized 3D wing geometries. Existing aerodynamics datasets are either limited to 2D simulations, lack optimization, or derive diversity solely from perturbations to a single baseline design, constraining their application as benchmarks to inverse design approaches and in the study of design diversity. The OptiWing3D dataset addresses these gaps, consisting of 1552 simulations resulting in 776 wing designs initialized from distinct extruded airfoil cross-sections. Additionally, a majority of the optimized wings in the dataset are paired to 2D counterparts optimized under identical conditions, creating the first multi-fidelity aerodynamic shape optimization dataset. Moreover, this structure allows for a direct comparison between 2D and 3D aerodynamic simulations. It is observed that 3D optimized designs diverge most prominently from the 2D-optimized designs near the wingtip, where three-dimensional effects are strongest, a finding made possible by the paired nature of the dataset. Finally, we demonstrate a constraint-aware conditional latent diffusion model capable of generating optimized wings from flow conditions, establishing a baseline for future inverse design approaches. The dataset, containing wing geometries and surface pressure distributions is publicly released to advance research in data-driven aerodynamic design.

cs.CE↗

arXiv2025

Least Volume Analysis

This paper introduces Least Volume (LV)--a simple yet effective regularization method inspired by geometric intuition--that reduces the number of latent dimensions required by an autoencoder without prior knowledge of the dataset's intrinsic dimensionality. We show that its effectiveness depends on the Lipschitz continuity of the decoder, prove that Principal Component Analysis (PCA) is a linear special case, and demonstrate that LV induces a PCA-like importance ordering in nonlinear models. We extend LV to non-Euclidean settings as Generalized Least Volume (GLV), enabling the integration of label information into the latent representation. To support implementation, we also develop an accompanying Dynamic Pruning algorithm. We evaluate LV on several benchmark problems, demonstrating its effectiveness in dimension reduction. Leveraging this, we reveal the role of low-dimensional latent spaces in data sampling and disentangled representation, and use them to probe the varying topological complexity of various datasets. GLV is further applied to labeled datasets, where it induces a contrastive learning effect in representations of discrete labels. On a continuous-label airfoil dataset, it produces representations that lead to smooth changes in aerodynamic performance, thereby stabilizing downstream optimization.

cs.LG↗

arXiv2025

EngiBench: A Framework for Data-Driven Engineering Design Research

Engineering design optimization seeks to automatically determine the shapes, topologies, or parameters of components that maximize performance under given conditions. This process often depends on physics-based simulations, which are difficult to install, computationally expensive, and require domain-specific expertise. To mitigate these challenges, we introduce EngiBench, the first open-source library and datasets spanning diverse domains for data-driven engineering design. EngiBench provides a unified API and a curated set of benchmarks -- covering aeronautics, heat conduction, photonics, and more -- that enable fair, reproducible comparisons of optimization and machine learning algorithms, such as generative or surrogate models. We also release EngiOpt, a companion library offering a collection of such algorithms compatible with the EngiBench interface. Both libraries are modular, letting users plug in novel algorithms or problems, automate end-to-end experiment workflows, and leverage built-in utilities for visualization, dataset generation, feasibility checks, and performance analysis. We demonstrate their versatility through experiments comparing state-of-the-art techniques across multiple engineering design problems, an undertaking that was previously prohibitively time-consuming to perform. Finally, we show that these problems pose significant challenges for standard machine learning methods due to highly sensitive and constrained design manifolds.

cs.CE↗

arXiv2025

GrainPaint: A multi-scale diffusion-based generative model for microstructure reconstruction of large-scale objects

Simulation-based approaches to microstructure generation can suffer from a variety of limitations, such as high memory usage, long computational times, and difficulties in generating complex geometries. Generative machine learning models present a way around these issues, but they have previously been limited by the fixed size of their generation area. We present a new microstructure generation methodology leveraging advances in inpainting using denoising diffusion models to overcome this generation area limitation. We show that microstructures generated with the presented methodology are statistically similar to grain structures generated with a kinetic Monte Carlo simulator, SPPARKS.

cs.GR↗