Filters reveal emergent structure in computational morphogenesis

TL;DR

This work introduces a nonperturbative, Laplace-transform-based filter (Pareto-Laplace transform) for computational morphogenesis to study emergent structure in topology-optimization problems. By mapping design densities to particle-like degrees of freedom and treating compliance as a potential energy, the authors define a partition-function-like quantity $Z(\beta)$ whose temperature controls the exploration of the design space; this reveals distinct condensation regimes and site-level importance without requiring the final optimal design. The approach highlights how emergent morphology arises from in-play degrees of freedom and provides both local (site-specific) and global (effective dimensionality) insights, validated on 2D and 3D compliance minimization with open-source implementations. The method generalizes to other topology-optimization problems and non-gradient settings, offering a principled, scalable nonperturbative tool that can guide robust design under manufacturing variation and even inform data generation for AI-assisted design. Overall, it offers a rigorous framework to predict critical design elements and enhance design-realization reliability across domains.

Abstract

Revolutionary advances in both manufacturing and computational morphogenesis raise critical questions about design sensitivity. Sensitivity questions are especially critical in contexts, such as topology optimization, that yield structures with emergent morphology. However, analyzing emergent structures via conventional, perturbative techniques can mask larger-scale vulnerabilities that could manifest in essential components. Risks that fail to appear in perturbative sensitivity analyses will only continue to proliferate as topology optimization-driven manufacturing penetrates more deeply into engineering design and consumer products. Here, we introduce Laplace-transform based computational filters that supplement computational morphogenesis with a set of nonperturbative sensitivity analyses. We demonstrate how this approach identifies important elements of a structure even in the absence of knowledge of the ultimate, optimal structure itself. We leverage techniques from molecular dynamics and implement these methods in open-source codes, demonstrating their application to compliance minimization problems in both 2D and 3D. Our implementation extends straightforwardly to topology optimization for other problems and benefits from the strong scaling properties observed in conventional molecular simulation.

Figures (9)

• Figure 1: The Pareto-Laplace transform for compliance minimization: density dynamics and ensemble generation. (a) Compliance minimization is about distributing a limited amount of material within a Design Domain to minimize compliance (deformation) under Applied Forces, considering boundary conditions like Fixed Nodes. The density of each site, $x_e$, is a value between $0$ and $1$, where a higher density reduces compliance, while a lower density reduces total used material (volume). (c) The density of each site is analogized to the position of a particle, where forces from compliance and volume constraints influence its position. (b) Connecting the particle system to a heat bath via a Nosé-Hoover thermostat generates an isothermal ensemble of design solutions. (d) shows the average design solutions for low and high-temperature ensembles in 2D and 3D cantilever beam problems.
• Figure 2: Three Phases of Condensation for a 2D Cantilever Beam Problem.(a–c) mean material distribution at representative temperature regimes: (a)$T \lesssim 3.3$, showing infill condensation within the frame; (b)$3.3 \lesssim T \lesssim 17.2$, illustrating the formation of a connecting frame; (c)$T > 20.7$, highlighting initial condensation near fixed and load-bearing nodes. (d–e) Compliance ratio $\langle C \rangle / C_\text{min}$ and mean pressure $\langle \lambda \rangle$ as functions of temperature, respectively. For $17.2 \lesssim T \lesssim 20.7$, the mean volume pressure increases sharply, while the compliance ratio remains nearly constant, indicating that the reduction in compliance is due to the addition of material rather than condensation. (f–m) Site-specific Mean density $\langle x \rangle$ and entropy density $S/S_\text{max}$ across four representative sites: (f, g) An essential site to compliance minimization (red square in (a–c)), exhibiting high density and low entropy even at high $T$. (h, i) A site with a contribution to volume reduction rather than compliance minimization (gray square in (a–c)), characterized by simultaneous loss of density and entropy at $T \approx 20$. (j, k) A designable site (light blue square in (a–c)) maintains high entropy and $\langle x \rangle$ until $T \approx 3.3$. (l, m) A sensitive site (dark blue square in (a–c)) converges to its final density at higher $T \approx 20$ while preserving entropy down to $T \approx 3.3$.
• Figure 3: Importance Maps for 2D and 3D Cantilever Beam Design Problems.(a) An optimized solution as solved by andreassenEfficientTopologyOptimization2011. (b) The Condensation Temperature Map, showing the normalized condensation temperature across the design domain with the colormap in (c), assuming a filled density ($x_e=1$) for all sites. (c) shows a 2D colormap where the horizontal axis represents site density and the vertical axis shows the normalized condensation temperature. The vertical axis indicates site importance, with higher condensation temperatures marking essential sites and lower temperatures indicating designable ones. High-density sites (opaque) contribute to compliance minimization, while low-density sites (transparent) aid in volume reduction.(d) The Importance Map for the 2D Cantilever Beam, based on the optimized solution in (a) and the Condensation Temperature Map in (b), using the colormap provided in (c). (e) The Importance Map for the 3D Cantilever Beam.
• Figure 4: Linear Response of Average Compliance to Changing Temperature. (a) Different slopes of the linear response indicate distinct regimes of condensation, each with a different number of effective dimensions. (b) At $T=25$, most sites are unconstrained with maximized entropy, and only the few in-play sites contribute to reducing compliance, resulting in the linear behavior is shown by the red line in (a). As temperature decreases to $T=9$, more sites become condensed, leading to more in-play sites and a higher slope of the linear response, as illustrated by (c) and the grey line in (a). (d) shows the design domain at $T=5$, where most sites are in-play, optimizing the infill design to minimize compliance.
• Figure A1: Mean density vs $T$ for all sites for a cantilever beam in 2D. Fig. \ref{['Fig:CondensationPhases']} (main text) panels f,h,j,l showed mean density vs $T$ for selected material sites. (a) plots mean density vs $T$ for all material sites of the cantilever beam design which is shown in (b). Data points in (a) are shaded by the density of the corresponding site in the ultimate design in (b). (c) magnifies the results from a single site to show the density and temperature scales used throughout the individual site plots in (a).