Topology Optimisation for Manufacturable Structures

Topology optimisation is the most powerful tool available for placing material along the load path. It is also one of the most misused. This flagship article traces the full workflow from design space to verified hardware: design domain, non-design regions, loads, constraints, objectives, manufacturing constraints, engineering interpretation, CAD reconstruction, detailed FEA, manufacturing review and test.

Article 04Lightweight Design16 min read

topology optimisationdesign spacemanufacturing constraintsCAD reconstructionload pathfinite element analysisgenerative design

What Topology Optimisation Actually Does

Topology optimisation determines where to place material within a defined design space so that an objective — typically mass or compliance — is minimised subject to a set of constraints. It is a mathematical method that treats the design space as a field of material density and iteratively removes material that contributes least to the objective. The result is a material distribution, not a CAD model. Understanding this distinction is the single most important step in using topology optimisation for real hardware. The raw result contains valuable load-path information, but it is not a component: it is a density field that must be interpreted, constrained, reconstructed as CAD, reanalysed and manufactured before it becomes hardware.

THE RAW TOPOLOGY OPTIMISATION RESULT IS NOT THE FINISHED COMPONENT.

The Inputs That Determine the Result

The topology optimisation result is entirely determined by its inputs. Change the design space, the load cases or the manufacturing constraints, and the result changes. The inputs are therefore not administrative settings; they are engineering decisions that must be made deliberately and documented. A topology optimisation whose inputs are not clearly defined cannot be reproduced, cannot be reviewed and cannot be defended.

Input	Role	Effect on result	Common pitfall
Design domain	The volume within which the optimiser may place material	Defines the outer envelope of possible solutions; too small a domain constrains the result, too large a domain allows impractical geometry	Defining the design space to match the existing component rather than the functional requirement
Non-design regions	Volumes excluded from optimisation (interfaces, bolt holes, mounting pads)	Preserves features that must remain unchanged; their position forces the load path around them	Making too much of the component non-design, leaving the optimiser nothing to work with
Load cases	The set of forces, pressures, temperatures and accelerations the component must withstand	Determines the load paths the optimiser reinforces; missing a load case produces a structure that is optimal for the wrong requirement	Optimising against a single static load case and ignoring fatigue, buckling and thermal loads
Boundary conditions	The constraints and connections that define how loads are reacted	Governs how the load enters and leaves the component; incorrect constraints produce an unrealistic load path	Over-constraining the model to achieve convergence, producing an artificially stiff result
Volume fraction	The target fraction of the design space to be filled with material	Sets how much material the optimiser may use; too low a fraction produces spindly, unmanufacturable geometry	Pushing the volume fraction too low in pursuit of mass reduction, at the cost of manufacturability and stress
Minimum feature size	The minimum dimension the optimiser may produce	Prevents features thinner than the process can build; controls the discreteness of the result	Omitting a minimum feature size, producing struts and webs below the manufacturable limit
Symmetry	A constraint that the result must be symmetric about a plane	Enforces symmetry, halves the solution space and improves manufacturability; can over-constrain an asymmetric load case	Applying symmetry to a genuinely asymmetric load set, producing a sub-optimal result
Manufacturing constraints	Constraints on draw direction, build direction, overhang angle, member size and extrusion direction	Produces a result that respects the chosen manufacturing process; without these, the result is mathematically optimal but practically unbuildable	Applying manufacturing constraints after the optimisation rather than within it

The Optimisation Workflow

The workflow below is the sequence that connects a structural requirement to verified hardware. Each stage has a defined input and output, and each stage introduces engineering judgement that the optimiser cannot provide. Skipping a stage — particularly the engineering interpretation, CAD reconstruction and detailed FEA stages — produces a component that is not defensible. The ordered list is the process; it is not optional.

Design Space — define the volume within which material may be placed, based on the functional envelope and interface requirements.
Loads & Constraints — define all load cases, boundary conditions and design constraints that the component must satisfy.
Topology Optimisation — run the optimiser with the defined objective, volume fraction, minimum feature size and manufacturing constraints.
Raw Material Distribution — extract the raw density field from the optimiser.
Engineering Interpretation — interpret the density field as load-path geometry, smoothing, merging and pruning features that do not carry significant load.
Manufacturing Constraints — apply build orientation, overhang, support and feature-size constraints to the interpreted geometry.
CAD Reconstruction — rebuild the interpreted geometry as a solid CAD model suitable for manufacture and analysis.
Detailed FEA — reanalyse the reconstructed CAD model against all load cases, including fatigue, buckling and thermal cases not used in the optimisation.
Manufacturing Review — review the CAD model with the manufacturing engineer for buildability, support removal, inspection and machining.
Test — substantiate the final design by test, including coupon, element and component-level testing as required by the qualification basis.

The Conceptual Optimisation Formulation

Topology optimisation can be stated conceptually as the minimisation of mass subject to a set of structural and manufacturing constraints. The formulation below is conceptual: a real optimisation problem includes many more constraints and sub-cases, and the precise formulation depends on the solver, the objective and the qualification basis. It is stated here to make clear what the optimiser is and is not doing. The optimiser is minimising mass subject to the constraints it is given. If a constraint is omitted — fatigue, for example — the optimiser does not know it is missing, and the result will not satisfy it.

Conceptual optimisation formulation:

  minimise  mass  m
  subject to:
    σ              ≤  σ_allowable          (stress)
    δ              ≤  δ_limit              (displacement)
    λ_buckling     ≥  RF_required          (buckling factor)
    f_natural       ≥  f_required           (frequency)
    manufacturing constraints satisfied

where:
  m              = mass of the component
  σ              = computed stress under the load case
  σ_allowable    = allowable stress for the relevant failure mode
  δ              = computed displacement
  δ_limit        = allowable displacement
  λ_buckling     = computed buckling load factor
  RF_required    = required reserve factor for buckling
  f_natural      = computed natural frequency
  f_required     = required frequency

This formulation is conceptual. A real problem includes multiple load cases, fatigue constraints, thermal loads, damage tolerance and process-specific manufacturing constraints. The optimiser satisfies only the constraints it is given.

From a Machined Bracket to Verified AM Hardware

The diagram below traces a single component through the full optimisation-to-hardware workflow. The starting point is a traditional machined bracket — a block of material with most of the interior carrying little load. The design space is defined around the functional envelope; the optimiser produces a raw density field that places material only along the dominant load path. The density field is interpreted as load-path geometry, manufacturing constraints are applied, the geometry is reconstructed as smooth engineering CAD, and the component is built additively, machined at its interfaces and structurally verified. The mass reduction at each stage is real, but only the final verified hardware is a defensible outcome.

[DIAGRAM: Optimisation-to-hardware workflow shown as eight stages, left to right. (1) Traditional machined bracket — a solid block with most of the interior under low stress, shown with a load arrow and reaction arrow. (2) Defined design space — the functional envelope shown as a translucent bounding volume, with non-design regions at bolt holes and interfaces. (3) Topology optimisation density result — a jagged, voxelised material distribution showing material concentrated along the dominant load path. (4) Interpreted load-path geometry — the density field smoothed and pruned into connected structural members. (5) Manufacturing-constrained design — overhang and feature-size constraints applied; self-supporting surfaces and fillets added. (6) Smooth engineering CAD — a clean solid model suitable for build and analysis. (7) Additively manufactured component — the as-built part on a build plate with supports. (8) Machined interfaces and structural verification — the final component with smooth bolt-hole flanges, connected to a test rig. Annotations note where engineering judgement is applied and where independent FEA is required.]

The Density Field Is Not a CAD Model

The raw topology optimisation output is a density field — a distribution of material within the design space, typically on a finite element mesh. It is not a CAD model. It contains jagged boundaries, intermediate densities, thin connecting features and artefacts of the mesh. Before it can be manufactured, it must be interpreted by an engineer who identifies the load-carrying members, removes numerical artefacts, smooths the boundaries, adds fillets and transitions, and reconstructs the geometry as a solid CAD model. This interpretation is where the engineer’s understanding of the load path, the manufacturing process and the inspection requirement is applied. The optimiser cannot do this: it has no understanding of the manufacturing process beyond the constraints it was given, and no judgement about which features are essential and which are numerical noise.

The raw density field contains useful load-path information, but it is not a CAD model. The transition from density field to manufacturable geometry involves engineering judgement that the optimiser cannot provide.

Independent Verification Is Not Optional

The reconstructed CAD model is a new geometry. It is not the geometry the optimiser analysed, and it is not the geometry that the optimiser’s convergence history refers to. It must be reanalysed independently against all load cases — including those not used in the optimisation — with a mesh that is appropriate for stress, fatigue and buckling, not for the optimisation. The optimiser’s own result is not a substantiation of the final hardware. Taking the raw topology optimisation result directly into production, without engineering interpretation, CAD reconstruction and independent verification, is not a defensible engineering process. The optimiser is a design tool; the verification is a separate, independent step that must be performed by an engineer who was not responsible for the optimisation.

TAKING THE RAW TOPOLOGY OPTIMISATION RESULT DIRECTLY INTO PRODUCTION WITHOUT ENGINEERING INTERPRETATION, CAD RECONSTRUCTION AND INDEPENDENT VERIFICATION IS NOT A DEFENSIBLE ENGINEERING PROCESS.