BladeSDF : Unconditional and Conditional Generative Modeling of Representative Blade Geometries Using Signed Distance Functions

TL;DR

The paper presents a domain-specific, DeepSDF-based framework for turbine blade modeling that uses a truncated SDF representation to reconstruct smooth, watertight blade geometries with quantified accuracy. It establishes a near-Gaussian latent manifold aligned with blade-relevant parameters and introduces a conditional pathway that maps engineering descriptors, such as maximum directional strains, to latent codes for performance-informed synthesis. Empirically, the approach achieves approximately 1% surface deviation relative to the blade extent on training data, demonstrates robust generalization to unseen designs, and supports unconditional generation via latent interpolation and Gaussian sampling as well as conditional generation through a strain-to-latent map. The resulting pipeline enables manufacturable, interpretable concept generation and demonstrates potential for design-in-the-loop workflows in turbine blade development.

Abstract

Generative AI has emerged as a transformative paradigm in engineering design, enabling automated synthesis and reconstruction of complex 3D geometries while preserving feasibility and performance relevance. This paper introduces a domain-specific implicit generative framework for turbine blade geometry using DeepSDF, addressing critical gaps in performance-aware modeling and manufacturable design generation. The proposed method leverages a continuous signed distance function (SDF) representation to reconstruct and generate smooth, watertight geometries with quantified accuracy. It establishes an interpretable, near-Gaussian latent space that aligns with blade-relevant parameters, such as taper and chord ratios, enabling controlled exploration and unconditional synthesis through interpolation and Gaussian sampling. In addition, a compact neural network maps engineering descriptors, such as maximum directional strains, to latent codes, facilitating the generation of performance-informed geometry. The framework achieves high reconstruction fidelity, with surface distance errors concentrated within $1\%$ of the maximum blade dimension, and demonstrates robust generalization to unseen designs. By integrating constraints, objectives, and performance metrics, this approach advances beyond traditional 2D-guided or unconstrained 3D pipelines, offering a practical and interpretable solution for data-driven turbine blade modeling and concept generation.

Figures (19)

• Figure 1: Ground-truth SDF near the surface : Labeled samples for one blade obtained from the point cloud (convex-hull sign + KD-tree magnitude); visualization restricted to $|s|\le 0.1$ ($\delta=0.1$) to highlight the training band.
• Figure 2: Blade parameterization for automated dataset generation
• Figure 3: End-to-end workflow for DeepSDF-based blade modeling. Data generation: point clouds to ground-truth SDF via convex-hull sign and KD-tree distance; normalized to ($[-1,1]^3$). Training (auto-decoder): jointly learn decoder weights and 256-D per-design latents from truncated SDF pairs with an ($L_2$) latent prior. Latent-space analysis: PCA and marginals reveal interpretable axes and an approximately Gaussian, well-centered code distribution. unconditional generation: synthesize shapes by latent interpolation and diagonal-Gaussian sampling. Conditional generation: NN-map ($g_\phi$) from target maximum strains (($\varepsilon_x$,$\varepsilon_y$,$\varepsilon_z$)) to latents for goal-directed designs. Reconstruction and evaluation: extract meshes with marching cubes; assess accuracy via a surface distance metric and NRMSE.
• Figure 4: Training loss convergence for DeepSDF auto-decoder (8×512, $z=256$) : Clamped SDF L1 loss ($\delta=0.1$) drops >90% by 1̃00 epochs, then plateaus.
• Figure 5: Training-set distribution of surface distance (Eq. \ref{['eq:Distance Metric']}) : Histogram over $n_{\text{train}}=222$ reconstructions; mass concentrates near $5.5\times10^{-2}$ ($<1\%$ of $D_{\max}$), with a shallow tail to $8\times10^{-2}$.