Topology optimization and lattice structures
📍 Where we are: Part III · Design for AM — Chapter 10 of 15. The last chapter set rules for making a design printable. Here we meet the tools that let software propose shapes no person would sketch: topology optimization and lattices.
Picture a block of soap and a rope pulling on it. You carve away every bit of soap that is not helping to hold the rope. What is left looks like bone or a tree branch: thick where the force flows, absent where nothing happens. Topology optimization does this carving with mathematics. Lattices are the next trick: instead of a solid, you fill the inside with a repeating, open pattern, like a honeycomb, which weighs little and can be tuned.
What this chapter covers
We explain topology optimization (TO) in plain terms, with its best-known formulation, SIMP; say how generative design relates to it; introduce lattice structures built from repeating unit cells (strut-based and TPMS); and discuss how to make the results printable and why they must still be validated. Our running example, the lightweight bracket, is the demonstration piece.
Topology optimization in plain terms
A topology optimization problem has four ingredients:
- a design space: the block of material you are allowed to use;
- loads and supports: where the part is held and where forces act;
- an objective: for example, make the part as stiff as possible;
- a constraint: for example, use no more than 30% of the block's volume (an illustrative figure).
Software divides the design space into many small elements and decides how much material each should hold. In the SIMP method (Solid Isotropic Material with Penalization), each element has a density between 0 (empty) and 1 (solid). Intermediate densities are made unattractive by a penalty (a power law), so the result drifts toward clear solid and void; the power-law idea traces to Bendsøe's material-distribution formulation [1]. Sigmund's well-known 99-line MATLAB program implements exactly this: it uses element densities as design variables, a filter that avoids mesh-dependent results, and an optimality-criteria update, and it was written for teaching [2]. A 169-line three-dimensional counterpart is also published for newcomers [3]. The approach is one of several: level-set, phase-field and other methods are compared in a standard review [4].
The raw result looks organic. It is a map of a good layout, not yet a finished part.
Topology optimization turns a design space plus loads into a layout; lattices and smoothing then make it printable, and strut and TPMS unit cells differ in how they print.
Original diagram by the authors, created with AI assistance.
Generative design
Generative design is the commercial name for tools that run topology optimization (and related methods) under several settings, then show you a set of candidate designs that meet your loads, materials and manufacturing choices. It is a workflow, not a different physics: you still define loads, supports and constraints. The designer's job shifts from drawing to specifying the problem well and choosing among results.
Making it printable
Pure TO does not know about the printer. If you ignore the process, the optimizer will happily propose downward-facing surfaces that need supports. Research in this area, summarised in a review of TO for AM [5] and a wider review of design and structural optimisation for AM [6], adds AM constraints to the optimization. One example: a filter that simulates printing layer by layer so that overhangs stay within a limit angle (45° in the paper's formulation), yielding self-supporting designs [7]. Other constraints control minimum feature size and support or lattice infill. Common practice also includes smoothing the result, removing thin slivers, and checking the final shape against the design rules in design for additive manufacturing.
Lattice structures
A lattice fills a volume with a repeating pattern of unit cells, giving a stiff, light, porous structure. Two families matter:
- Strut lattices are made of beams that meet at nodes (like a jungle-gym frame). They are easy to describe and tune. Their mechanical response has been compiled across many unit-cell topologies, and relative density correlates with properties roughly as the Gibson-Ashby model predicts [8]. Sharp junctions can concentrate stress, and horizontal struts may need support in powder bed metal (a general overhang rule, see the previous chapter).
- TPMS lattices (triply periodic minimal surfaces, such as gyroid and diamond) are built from smooth, continuous surfaces that repeat in three directions. Reviews describe them as having smooth, highly interconnected, mathematically controllable geometry [9][10]. Whether a given cell prints without support depends on cell size, wall angle and process, so check it on your machine.
Lattices are discussed for lightweighting, energy absorption and bone-ingrowth implants [8][9]. Grading the cell size or thickness across a part lets you tune stiffness in different regions.
Their special print issues are powder removal (open cells are better than closed ones, but small cells can trap powder), minimum strut size, and surface roughness that matters at small thickness. See the minimum-feature discussion in design for additive manufacturing.
Validation caveats
An optimized, lattice-filled design is only as good as its model.
- The optimization uses a simplified load case; real use has more loads, fatigue and impact.
- The as-built part differs from the CAD: rough surfaces, residual stress and process-dependent material properties are usually not in the model.
- Reviews of lattices point to open challenges such as computational cost for multi-scale design and the link between process, structure and properties [10]. Reviews of TO for AM likewise list uncertainty and post-treatment as open topics [5].
So test prototypes and use inspection (see quality, inspection and NDT) before trusting a computed benefit.
The bracket test
For our bracket: define the bolt-hole supports and the load, let TO find the load paths, smooth them, add fillets, and check overhangs against the planned orientation. Consider a TPMS or strut infill only where a solid section would be thick. How much weight is saved depends entirely on the load cases, so we make no numerical claim here.
Why it matters
These tools are the clearest example of AM's "complexity is free" promise, and they connect design to the physics of the part. Used without constraints or validation, they produce shapes that cannot be printed or that fail in service. Used well, they give lighter, stiffer parts.
In the real world
TO is a mature research field with a long history [4][1], commonly taught with short educational codes [2][3]. Research on AM-specific constraints, supports and lattice infill continues [5], and the DfAM literature frames this kind of design freedom as an opportunity, alongside constraints [11].
Key terms
- Topology optimization (TO) — software that decides where material should be inside a design space to meet an objective under constraints.
- SIMP — a density-based TO method that penalises intermediate densities so elements become solid or void.
- Generative design — a workflow that runs TO-like methods to propose several candidate designs.
- Design space — the volume in which the optimizer may place material.
- Lattice — a repeating, open structure of unit cells.
- Unit cell — the small repeating building block of a lattice.
- Strut lattice — a lattice made of beams that meet at nodes.
- TPMS — triply periodic minimal surface: a smooth surface that repeats in three directions, such as a gyroid.
- Self-supporting — printable without supports because overhangs stay within the limit angle.
Where this leads
Even a well-designed part needs a build plan: orientation, supports, nesting and distortion control. That is orientation, supports and build preparation.