Skip to content

Design · 25 min read

Engineering Organic Parts: Topology Optimization & Generative Design for 3D Printing

How topology optimization and generative design work, why organic shapes carry load so well, and what it takes to turn an optimized shape into a reliable FDM-printed part.

Portfolio projectOrganic Bracket Design — Conventional vs OrganicSee the project →

Organic parts — the smooth, branching, bone-like brackets and mounts you see more and more often — are not a style. When they are done properly, every curve is there because of a load. This article explains where those shapes come from, why they work, and what has to happen between an optimization result and a printed part that you can bolt into a machine and trust. It is written for engineers and product designers, but we have tried to explain each idea from the ground up.

One note before we start: this is general engineering guidance. We don't quote weight savings or strength gains, because those depend entirely on the part, its loads and how it is made. Anyone who promises a fixed percentage before seeing your loads is guessing.

What "organic" design really means

An organic part is one whose shape follows the way forces travel through it. Material is placed along the paths that loads take between the points where the part is held and the points where it is pushed, pulled or twisted. Everything that does not help carry those loads is left out, and the transitions between members are smooth rather than sharp.

There are three common ways to get there:

  • Topology optimization: a computer method that decides where material should exist inside a defined space so that a goal (usually stiffness) is met with a limited amount of material.
  • Generative design: a workflow, offered by tools such as Autodesk Fusion, that runs many optimization studies with different materials, manufacturing methods and settings, and presents the results for an engineer to compare.
  • Engineer-led organic modelling: an experienced designer shapes the part by hand along known load paths, often guided by simulation, without running a formal optimizer.

All three can produce excellent parts. All three can also produce parts that look impressive and fail early, if the loads are wrong or the result is not adapted to how the part will be made.

Topology optimization, explained

Topology optimization answers a deceptively simple question: given a block of space, the places where a part is held, and the loads it must carry, where should the material go? The field was put on a rigorous footing in the late 1980s (Bendsøe & Kikuchi, 1988) and is covered comprehensively in the standard textbook by Bendsøe and Sigmund.

The ingredients of a problem

  • Design space: the volume the part is allowed to occupy. The optimizer can only remove material from inside it, so a design space that is too small limits the result before anything is solved.
  • Non-design space (keep-in regions): regions that must stay solid no matter what — bolt bosses, bearing seats, mating faces, sealing surfaces.
  • Keep-out regions: volumes that must stay empty, such as clearance for fasteners and tools, the swept path of a moving part, cables, or a hand.
  • Boundary conditions: how the part is held. A bolted face, a pin in a bore and a bonded edge each behave differently, and modelling a support as perfectly rigid when it is not will pull material toward it.
  • Load cases: each set of forces and moments the part must survive — operating loads, but also assembly, transport, a drop, a dropped tool or someone standing on it. A part optimized for one load case can be surprisingly weak in another.
  • Objective: what is being minimized or maximized — most often compliance (the inverse of stiffness), sometimes mass, stress, displacement or a natural frequency.
  • Constraints: the limits the result must respect — usually a volume fraction (the share of the design space that may be filled), plus stress, displacement, symmetry, minimum member size or manufacturing rules.

Density methods and SIMP

The most widely used approach is the density method. The design space is divided into small finite elements, and each element gets a density between nearly zero (void) and one (solid). The optimizer adjusts those densities to improve the objective.

Left alone, the optimizer would happily produce a fog of half-dense material, which cannot be made. SIMP (Solid Isotropic Material with Penalization) solves this by making intermediate densities a bad deal: an element's stiffness is scaled by its density raised to a penalty power, commonly 3, so half-dense material gives much less than half the stiffness for half the material (Bendsøe, 1989). The optimizer is pushed toward clear solid-or-void decisions.

The classic problem is minimum compliance with a volume constraint: make the stiffest possible structure using, say, a set share of the design space. The optimizer repeatedly runs a finite-element analysis, calculates how sensitive the compliance is to each element's density, and updates the densities — using an optimality-criteria update or a general method such as MMA. Sigmund's well-known "99 line" code, and its faster 88-line successor, show the whole loop in a few dozen lines of MATLAB and remain the best way to understand it (Sigmund, 2001; Andreassen et al., 2011).

Filters, length scale and checkerboards

Raw density optimization has well-documented numerical problems: checkerboard patterns of alternating solid and void elements, results that change as the mesh is refined, and convergence to poor local optima (Sigmund & Petersson, 1998). The standard cure is filtering — averaging densities or sensitivities over a small radius (Bourdin, 2001) — often followed by a projection step that sharpens the result toward pure solid and void.

The filter radius matters more for 3D printing than it first appears: it sets a minimum length scale, a lower limit on how thin members can become. For an FDM part, that radius should be tied to the thinnest member the printer can produce reliably, which we come back to below.

Level-set and evolutionary methods

Density methods are not the only option. Level-set methods describe the part's boundary as the zero contour of a smooth function and move that boundary according to shape sensitivities, which gives crisp, smooth edges throughout the optimization (Wang, Wang & Guo, 2003; Allaire, Jouve & Toader, 2004). Evolutionary methods such as ESO remove lightly stressed material step by step (Xie & Steven, 1993). Sigmund and Maute's comparative review is the best overview of the strengths and weaknesses of each approach.

What compliance minimization does — and doesn't — give you

Minimum compliance means maximum stiffness for the material allowed. It does not directly mean maximum strength. A stiff structure usually has low stresses, but compliance-based results can still contain local stress peaks, slender members that buckle, or joints that are poor in fatigue. Stress-constrained, buckling-aware and fatigue-aware formulations exist, but they are harder to solve, so most commercial workflows still start from stiffness and then check stress afterwards. That check is not optional.

Generative design: many answers instead of one

Generative design wraps topology optimization (and related solvers) in a workflow that explores many combinations — materials, manufacturing methods, safety targets — and returns a set of outcomes to compare. Autodesk's documentation describes setting up preserve and obstacle geometry, loads, constraints and manufacturing methods, then reviewing the results side by side. Altair OptiStruct (now part of Siemens as Simcenter OptiStruct) and nTop offer topology optimization inside their own toolchains; nTop pairs it with implicit modelling, which handles complex smooth geometry especially well.

NASA's Goddard Space Flight Center has described using AI-driven generative design to produce what it calls "evolved structures" for mission hardware — an example of the approach being used where mass and reliability both matter. The lesson from that work, and from every serious user of these tools, is the same: the software proposes shapes, and engineers still define the problem, judge the results and validate the final part.

The engineering workflow, step by step

An optimized part is only as good as the problem it was solved for. This is the workflow we follow, and the one we would recommend to anyone.

1. Requirements

Start with what the part must do, not what it should look like:

  • The interfaces it must connect to, with their dimensions and tolerances
  • Every load it will see, how often, and for how long — sustained, cyclic or impact
  • The environment: temperature range, sunlight, moisture, chemicals
  • The space available, and what must stay clear
  • Stiffness limits (how far it may deflect) as well as strength
  • Quantity, budget, and whether it must ever be made another way

2. Load cases and safety factors

Turn the requirements into load cases with real magnitudes and directions. Where loads are uncertain, bracket them: an expected case and a credible worst case. Choose a safety factor that reflects how well the loads are known, how variable the material and process are, and what happens if the part fails. For printed polymer parts, process variation and direction-dependent strength argue for a more conservative factor than you would use for a machined metal part with certified material data.

3. Design space

Model the largest volume the part may occupy, then cut in the keep-out regions and define the keep-in regions around every interface. Give the optimizer room: a cramped design space is the most common reason an "optimized" part ends up looking like the original.

4. Solve

Set the objective and constraints, include all load cases (as a weighted combination or a worst-case formulation), choose a filter radius matched to the minimum printable member, and add manufacturing constraints where the software supports them — symmetry, overhang limits or minimum member size. Run a few volume fractions rather than one; the best part is often not at the lowest mass the solver can reach.

5. Interpret, smooth and reconstruct

The raw result is a field of element densities. To get geometry, you take an iso-surface at a chosen density threshold, and the threshold itself changes how much material survives. That surface is faceted and noisy, so it has to be rebuilt as clean, manufacturable geometry. The main options are:

  • Subdivision and T-spline modelling: a smooth cage is fitted over the result and refined by hand. This gives very clean, editable freeform surfaces and full control over every transition.
  • Implicit modelling: geometry is defined by mathematical fields rather than surfaces, which makes smooth blends, variable thickness and lattices robust even on very complex shapes.
  • Parametric rebuild: the load paths are interpreted as conventional CAD features — sweeps, lofts and fillets — which is slower for complex shapes but keeps the part easy to edit and dimension.

Whatever the method, the interfaces are restored as exact, dimensioned features: bores, flat mating faces and bolt holes are not left as smoothed approximations.

6. Validate with FEA

Reconstruction changes the part, so the final geometry is analysed again, under every load case. Check peak stresses against allowables that include the safety factor, check deflection against the stiffness requirement, and check slender members for buckling. Where loads are cyclic or sustained, consider fatigue and creep as well.

For FDM parts, be honest about the material model. Most simulations assume an isotropic material, the same in every direction. A printed part is not, so either use conservative allowables based on the weakest (across-layer) direction or use a direction-dependent material model.

7. Design for the print

Orientation, supports, minimum members, holes and inserts are decided now, before the first print, and can feed back into the design. The next section covers this in detail.

8. Test

Print the part in its intended material and orientation and test it in conditions as close to real use as possible: a proof load, a fit check, a cycle test or a temperature soak. Analysis narrows the risk; testing closes it. If you need standardized test data — tensile properties to a published test method, for example — that is work for an accredited test laboratory, and it should be planned with specimens printed the same way as the part.

Why organic shapes carry load so well

Load paths

Forces travel through a part along paths, much as water finds its way downhill. In a conventional bracket — a flat plate, a web and a boss — the load is forced through whatever material happens to be there, and much of the volume sits in places where the stress is low. An organic shape puts material along the paths and removes it elsewhere. Members carrying mainly tension or compression can be slender; members in bending get depth in the direction they bend.

Smooth transitions and stress concentrations

Sharp internal corners concentrate stress: the smaller the fillet radius, the higher the local stress for the same load. The standard reference on how strongly geometry concentrates stress is Peterson's Stress Concentration Factors (Pilkey & Pilkey). Organic designs replace sharp corners with large, smoothly varying transitions, which lowers those peaks — important for any part, and especially for printed polymers, where a crack that starts at a notch can run along a layer line.

Trees and bones

Nature has been solving this problem for a long time. Claus Mattheck's work on trees showed that they grow in a way that tends to even out stress over their surface, adding material at the base of branches and at wound edges where stress would otherwise peak; his shape-optimization methods for engineering notches are modelled on this (Mattheck, 1998). Bone adapts in a similar way, remodelling over time in response to the loads it carries — the idea usually summarised as Wolff's law. The branching, tapering forms of a well-optimized bracket look natural for exactly this reason.

Engineering organic parts for FDM printing

Topology optimization is often presented as a perfect match for 3D printing because printing can make almost any shape. That is only half true. FDM has its own rules, and an organic part that ignores them can be weaker than the plain bracket it replaced.

Anisotropy: the across-layer weakness

An FDM part is built from extruded lines stacked in layers. Its properties depend on direction: parts loaded so that layers are pulled apart — across the layers, along the build (Z) axis — are generally weaker than parts loaded along the deposited lines. Es-Said et al. (2000) showed how strongly layer orientation affects strength, Ahn et al. (2002) showed that even within a layer the direction of the deposited lines matters, and the effect is discussed at length in reviews of FDM mechanical behaviour (Popescu et al., 2018; Dizon et al., 2018; Gordelier et al., 2019).

The cause is usually attributed to the interfaces between layers. Bond quality depends on how hot the material is when the next layer arrives and how long the interface stays hot enough to fuse (Sun et al., 2008). More recent work adds an important nuance: the bond itself can approach the strength of the bulk material, and much of the weakness comes from the shape of the interface — the grooves between rounded extrusions concentrate stress (Allum et al., 2020). Either way, the practical rule is the same: avoid pulling layers apart in the most highly stressed regions.

Layer interfaces are far more sensitive to tension and peel than to compression. A bolt clamping down across layers is generally fine; a bending load that opens the layers on the tension side of a member is where trouble starts.

Orientation strategy

For an organic part, orientation is chosen with the load paths in front of you:

  • Find the members that carry the highest tension or bending stress, and orient the part so those members lie along the build plate wherever possible, with their deposited lines running along their length.
  • Accept compression across layers before tension across layers.
  • Look at bolt and pin holes: a hole whose axis is vertical in the build prints round and accurate; a horizontal hole prints with a slightly flattened or drooping top.
  • Balance strength against overhangs, support contact and surface finish. Sometimes the best answer is to split the part, or to redesign a member so that it can be printed in the stronger orientation.

When the loads point in several directions and no single orientation suits them all, that is useful information to take back into the design — it is far cheaper to fix on screen than after a failed test.

Overhangs and self-supporting angles

FDM builds each layer on the one below. Surfaces that lean out too far from vertical need support, and around 45° from vertical is the common rule of thumb for a self-supporting overhang — though the real limit depends on material, cooling, layer height and speed. Organic parts are full of overhangs, because their members curve in every direction.

There are two ways to handle this. The first is to design for it: several researchers have added overhang constraints directly into topology optimization so that results are self-supporting (Langelaar, 2016; Gaynor & Guest, 2016), and the design-for-AM literature treats overhangs as a core constraint (Thompson et al., 2016; Liu et al., 2018). The second is to accept supports where they do least harm — on non-critical faces, away from bores and mating surfaces — and to plan how they will be removed.

Supports and surface quality on curved surfaces

Supports leave witness marks, and on a curved surface they leave them across a changing angle, which makes cleanup harder than on a flat face. Shallow, gently sloping surfaces also show stair-stepping, where each layer edge is visible. Practical measures:

  • Orient so that the most visible or functional curved surfaces face up or sideways rather than down onto supports
  • Use tree-style supports that touch the part at small points, which suits organic shapes
  • Reduce layer height where shallow curves must look smooth, accepting a longer print time
  • Plan support removal access: an enclosed organic loop can trap support material

Perimeters, infill and thin members

In bending and torsion, stress is highest at the outer surface of a member. In an FDM part, that surface is made of perimeters (walls), so for slender organic members the perimeters carry most of the load. The practical consequence is simple: thin members should print solid, as perimeters all the way through, rather than relying on infill. Infill is useful in thick sections, but the strength of a well-optimized organic part lives mostly in its walls.

The seam — the point where each perimeter loop starts and ends — is a small discontinuity. On a highly stressed member, it is worth placing the seam on the least stressed side.

Minimum member thickness

A printer lays down lines of a certain width, roughly tied to its nozzle. A member that is only one or two line widths across prints as a couple of strands with a weak interface between them, or with gaps. Organic members should be sized to print as several solid perimeters across their thinnest point. That minimum should be set before optimization, through the filter radius or a minimum-member-size constraint, so the optimizer never proposes something that cannot be printed well. Fine tips and thin edges on smoothed geometry deserve the same check after reconstruction.

Holes, eyelets, bushings and inserts

Mounting eyelets are where organic parts meet the rest of the world, and they need particular care:

  • Size and shrinkage: printed holes usually come out slightly undersize, because the material shrinks as it cools and the slicer approximates curves with straight segments. Critical holes are either compensated in the model after a test print or printed slightly undersize and then drilled or reamed.
  • Wall around the hole: an eyelet needs enough material around the hole to carry bearing and bolt loads, and the boss should blend smoothly into its arm.
  • Heat-set inserts: for threads that will be assembled more than once, brass heat-set or threaded inserts are far more durable than threads in plastic. Their pull-out behaviour depends on the insert, the boss geometry and the print settings (Kastner et al., 2023), so the boss is designed around the insert manufacturer's recommendations.
  • Bushings and sleeves: bores that take a rotating pin, a shaft or high bearing loads are best served by a metal or bearing-grade bushing pressed or bonded into the part.

Bolt preload and creep

Polymers creep: under a sustained load they slowly continue to deform, and under a fixed deformation the stress slowly relaxes. For a bolted printed part, that means the clamping force from a tightened bolt falls over time as the plastic under the bolt head relaxes. Creep in FDM parts has been studied directly, and it depends on the material and on how the part was printed (Salazar-Martín et al., 2018; Mohamed et al., 2017).

Good practice for bolted joints:

  • Put a metal compression limiter (a sleeve) through the eyelet so the bolt clamps metal-to-metal and the plastic is not carrying the preload
  • Use washers to spread the load under heads and nuts
  • Avoid clamping across thin or highly curved features
  • For sustained structural loads, choose a material with better creep resistance and check stresses with long-term behaviour in mind, not just short-term strength

Choosing a material

The material decision comes from the loads, the temperature and the environment. In broad terms:

  • PLA: stiff and easy to print, excellent for prototypes and fit checks, but it softens at fairly low temperatures and creeps under sustained load — usually not the choice for a structural bracket in service.
  • PETG: tough and forgiving, with good layer bonding and moderate heat resistance. A sound general-purpose choice for indoor functional parts.
  • ASA: holds up to sunlight and outdoor weather and handles more heat than PETG. Our usual starting point for outdoor brackets.
  • Polycarbonate (PC): high toughness and heat resistance; more demanding to print without warping.
  • Carbon-fibre nylon (PA-CF): very stiff for its weight with good heat resistance, which suits stiffness-driven organic parts. Nylon absorbs moisture, which changes its properties, and short fibres tend to increase the difference between in-layer and across-layer behaviour (Brenken et al., 2018), so orientation matters even more.

Our Filament Buyer's Guide compares the materials we print in more detail.

Post-processing

  • Support removal and cleanup: careful trimming, then sanding blended into the curvature so a flat spot is not left on a curved surface
  • Hole finishing: drilling or reaming critical bores to size, then fitting inserts, bushings or sleeves
  • Surface finishing: filler primer and paint where appearance matters; some materials can be vapour-smoothed, which seals the surface but can soften fine detail and slightly change dimensions
  • Annealing: some materials gain heat resistance when annealed, but parts can shrink or warp in the process, so it is planned and tested rather than applied by default

When not to use organic design

Organic design is a tool, not a default. A conventional part is often the better choice:

  • Simple loads: a plate in pure tension, a spacer or a shim gains little from optimization.
  • Cost of engineering: setting up load cases, solving, reconstructing and validating takes time. For a one-off part with a generous safety margin, a well-designed conventional part may be cheaper overall.
  • Editability: organic geometry is harder to modify, dimension and revise than a feature-based CAD model. If the design will keep changing, that matters.
  • Other manufacturing processes: if the part may later be machined, moulded or cast, organic shapes can be hard or impossible to make that way without redesign — they ignore tool access, draft angles and parting lines unless those are built into the optimization.
  • Inspection: freeform surfaces are harder to measure and check than flat faces and simple features.
  • Unknown loads: if the loads are guesses, an optimizer will confidently optimize for the wrong thing. A robust conventional part may be safer until the loads are understood.
  • FDM economics: sparse infill already removes a lot of material from a conventional printed part, so the benefit of an organic design is often stiffness and fewer stress concentrations rather than dramatic savings in print time or filament.

How FORMYQ approaches organic parts

When a customer brings us a part that might benefit from an organic design, this is how we work:

  1. Review first: we look at what the part does, its interfaces, its loads and its environment, and tell you honestly whether an organic design is worth it or whether a simpler part will do.
  2. Define the problem with you: load cases, safety targets, keep-in and keep-out regions and stiffness limits are agreed and written down before any optimization starts.
  3. Optimize and reconstruct: we develop the shape and rebuild it as clean CAD with exact interfaces, sized for the printer and material from the start.
  4. Check the design: the final geometry is analysed under every load case, with conservative assumptions for direction-dependent strength.
  5. Plan the print: orientation, supports, perimeters, inserts and bushings are specified, not left to chance.
  6. Prototype and test: we print in the intended material and orientation for fit and function testing, and recommend an accredited laboratory when certified or standardized data is needed.
  7. Hand over: you receive versioned files, and your IP stays yours. Our design service covers how projects are scoped and priced, and nothing is shown in our portfolio without your written permission.

We don't promise a percentage weight saving or strength increase before we have seen your part, because any honest number depends on the loads, the material and the way the part is made.

Frequently asked questions

Is topology optimization the same as generative design?

Not quite. Topology optimization is the mathematical method that decides where material goes for one problem. Generative design is a broader workflow that runs many such studies — across materials, manufacturing methods and settings — and lets an engineer compare the outcomes.

Will an organic part always be lighter and stronger?

No. An optimized part is stiffer for its mass under the loads it was optimized for. Under loads that were left out, or if it is printed in a poor orientation, it can be weaker than a conventional part. The result is only as good as the load cases and the manufacturing plan.

Can you use the optimizer's output directly?

Rarely. Raw optimization output is faceted and noisy, and its interfaces are approximate. It is almost always reconstructed as smooth geometry with exact bores and mating faces, then analysed again.

Which way should an organic part be printed?

So that the most highly stressed members lie along the layers rather than across them, while keeping overhangs and support contact under control. It is a balance decided for each part, often with small design changes to make a good orientation possible.

Is FDM strong enough for structural brackets?

For many brackets, yes, with the right material, orientation and joint design, and with an honest margin for direction-dependent strength and creep. For safety-critical parts, testing in real conditions is essential, and some applications need another process or material entirely.

Do you provide certified strength testing?

No. We can test prototypes for fit and function, and we recommend an accredited laboratory when you need certified results or data to a published test standard.

What do you need from me to start?

The part's interfaces and dimensions, what it connects to, the loads and how they act, the environment, the space available, and the quantity. Photos, sketches and an existing CAD model all help. You can send it all through a design brief.

Sources and further reading

General guidance for FDM printing. Results depend on your geometry, material and use — ask us about your specific part.

More guides

Put it into practice.

Start a Design BriefDiscuss Your Project