[Paper Review] Two-Scale Topology Optimization with Microstructures

Bo Zhu, Mélina Skouras+2 more|arXiv (Cornell University)|Jun 10, 2017

Visual Attention and Saliency Detection15 citations

TL;DR

This paper presents a two-scale topology optimization framework that jointly designs macro-structures and micro-structured materials to achieve high-performance, printable objects. By precomputing a continuous material property gamut using level set fields and optimizing spatially-varying material properties within this gamut, the method enables efficient, large-scale designs—up to trillions of voxels—demonstrated on complex functional objects and structures.

ABSTRACT

In this paper we present a novel two-scale framework to optimize the structure and the material distribution of an object given its functional specifications. Our approach utilizes multi-material microstructures as low-level building blocks of the object. We start by precomputing the material property gamut -- the set of bulk material properties that can be achieved with all material microstructures of a given size. We represent the boundary of this material property gamut using a level set field. Next, we propose an efficient and general topology optimization algorithm that simultaneously computes an optimal object topology and spatially-varying material properties constrained by the precomputed gamut. Finally, we map the optimal spatially-varying material properties onto the microstructures with the corresponding properties in order to generate a high-resolution printable structure. We demonstrate the efficacy of our framework by designing, optimizing, and fabricating objects in different material property spaces on the level of a trillion voxels, i.e several orders of magnitude higher than what can be achieved with current systems.

Motivation & Objective

• Address the scalability limitations of traditional topology optimization in high-resolution, multi-material 3D printing by decoupling macro-structure and micro-structure design.
• Overcome the computational infeasibility of optimizing at the voxel level for objects with billions or trillions of voxels.
• Enable the design of functional objects with spatially-varying, multi-dimensional material properties (e.g., isotropic, orthotropic, cubic) beyond simple density or stiffness.
• Develop a generalizable framework that supports diverse functional objectives such as minimal compliance, targeted strain distribution, and mechanical behavior.
• Bridge the gap between optimized continuous material properties and printable microstructures through a reverse mapping process.

Proposed method

• Precompute the material property gamut of all printable microstructures using alternating stochastic sampling and continuous optimization to cover the full range of achievable bulk properties.
• Represent the boundary of the material property gamut as a level set field to enable continuous, differentiable constraints in the optimization process.
• Reformulate the topology optimization problem in the continuous space of material properties, treating each macro-element as a point in the gamut rather than individual voxels.
• Use a constrained optimization scheme to simultaneously compute optimal object topology and spatially-varying material properties, constrained within the precomputed gamut.
• Map the optimized continuous material properties back to discrete, printable microstructures from a precomputed database to generate high-resolution, fabricable designs.
• Scale the optimization to trillions of voxels by using a coarse macro-lattice where each cell corresponds to a $64^3$ microstructure, enabling high-resolution fabrication.

Experimental results

Research questions

• RQ1 Can a two-scale topology optimization framework efficiently handle high-resolution designs (up to trillions of voxels) with spatially-varying multi-material properties?
• RQ2 How can the full range of bulk mechanical properties achievable by printable microstructures be systematically precomputed and represented as a continuous gamut?
• RQ3 Can a continuous, differentiable representation of the material property gamut enable efficient and general topology optimization across diverse functional objectives?
• RQ4 To what extent can the framework outperform traditional binary or single-property topology optimization in terms of functional performance and design complexity?
• RQ5 How can optimized continuous material properties be reliably mapped to printable microstructures while preserving mechanical performance?

Key findings

• The framework successfully optimized a Stanford bunny with over 100 million voxels under two distinct loading cases, achieving target deformation behaviors.
• A bridge design was optimized at 1 trillion voxels using a 4 million-cell macro-lattice, each cell mapping to a $64^3$ microstructure, demonstrating scalability to extreme resolutions.
• The method enabled the design of functional grippers that achieve grasping via either in-plane or out-of-plane deformation by varying the soft-to-rigid material ratio.
• The precomputed material property gamut achieved stable and comprehensive coverage of isotropic, cubic, and orthotropic material behaviors across the microstructure space.
• The optimization process scaled efficiently from low-resolution initializations (1.4 million elements) to high-resolution results (1 trillion voxels), maintaining convergence and performance.
• 3D-printed prototypes of optimized designs, including the gripper and bridge, validated the practicality and functionality of the framework in real-world applications.