Generative design is a design method in which software generates many part shapes that meet goals and constraints you set, such as loads, mounting points, material, weight target and manufacturing process, and then ranks them for an engineer to choose from and validate. It is not the same as a chatbot that draws pictures, and it does not remove the need for analysis and testing.
How does generative design work?
You describe the problem instead of drawing the part. The usual workflow has six steps:
- Define the design space. Specify the volume the part may occupy, plus regions that must stay solid, such as bolt holes, bearing seats and sealing faces, and regions that must stay clear.
- Set loads and constraints. Apply forces, pressures, temperatures and supports from the real duty cycle, including worst cases.
- Choose objectives. For example, minimise mass, maximise stiffness, or limit cost, subject to a safety factor.
- Choose manufacturing methods. Additive manufacturing, 3-axis or 5-axis machining, casting or sheet metal each allow different shapes. A design for 3D printing may be impossible to machine.
- Generate and compare. The software produces a set of candidates, usually with simulation results for each.
- Select, refine and validate. An engineer picks a candidate, rebuilds it as clean CAD, and verifies it with finite element analysis (FEA) and physical testing.
The value is less in a single clever shape than in exploring options quickly and finding designs a person would not have drawn.

What is the difference between generative design, topology optimisation and generative AI?
The terms are often mixed up.
- Topology optimisation takes a design space, loads and a target, and removes material where it is not needed to produce one optimal layout. It is a mathematical method with a long history in structural engineering.
- Generative design usually wraps optimisation in a workflow that produces multiple alternatives under different constraints and manufacturing methods, and compares them on performance and cost.
- Generative AI (image generators, large language models) produces content from learned patterns. Text-to-3D tools can create shapes quickly, but without physics and manufacturing rules they produce concepts, not parts you can safely make. Treat them as sketching aids.
When a vendor says "AI design", ask what physics solver sits underneath and what manufacturing constraints it respects.
What is a real example of generative design in production?
In 2016 Airbus and Autodesk unveiled the bionic partition, a divider between the passenger cabin and galley in the A320, designed with generative design software and made by 3D printing in Scalmalloy, an aluminium-magnesium-scandium alloy. Airbus reported that it was about 45 percent, or 30 kg, lighter than the current design. The design work was done by Autodesk's studio The Living, and the printing by APWorks, an Airbus subsidiary. Airbus estimated that applying the approach across the A320 fleet backlog could save up to 465,000 tonnes of CO2 a year.
Two cautions: the part was a demonstrator showing what was possible, and the CO2 figure is an estimate of a scenario, not a measured result. Certification of a flight part also requires extensive testing, which is a reminder that generative design shortens the concept stage and not the validation stage.
Which tools and skills do you need?
Mainstream CAD packages now include generative or topology tools, for example Autodesk Fusion, Creo and SOLIDWORKS, alongside specialist software for lattices and additive manufacturing. The software matters less than the engineering skills around it:
- Loads and boundary conditions. Garbage in, garbage out. A beautiful part optimised for the wrong load case is useless.
- FEA literacy. You must judge whether the simulation can be trusted: mesh, contacts, fatigue, buckling.
- Design for additive manufacturing (DfAM). Overhangs, support structures, orientation, powder removal, heat treatment and surface finish all shape the result.
- Design for machining or casting if you are not printing: tool access, draft angles, minimum wall thickness.
- Material and inspection knowledge. Printed metals have anisotropic properties, and internal features can be hard to inspect.
- Cost thinking. A lightweight part that costs five times more to make is not a win unless weight is worth that much.
What is the UK context?
The UK has a research and industrial base in advanced manufacturing. The High Value Manufacturing Catapult describes a national network of centres including the AMRC, the Manufacturing Technology Centre, WMG and the National Composites Centre, and says it supports more than 3,000 SMEs a year. These centres work with companies on additive manufacturing, digital design and process development, so they are worth knowing for any firm exploring generative design with production-grade validation. Made Smarter Adoption also offers manufacturing SMEs advice and grants for digital tools in their region. Check current terms and eligibility with each body.
What does a sensible first project look like?
Pick a bracket, mount or fixture that you already make, with a known load, a known weight and a simple function. A robot end-of-arm tool, a motor bracket or a jig for a machining cell are all good candidates because lower weight has a clear benefit (faster robot cycle, easier handling) and failure is not safety-critical. Record the current part's mass, cost and test results as a baseline. Run the generative study with the real load cases and your machine shop's or printer's constraints, then rebuild the best candidate as clean CAD and check it with FEA. Make two or three prototypes, test them to destruction or to a fatigue target, and compare cost and performance with the original. If the new part wins on the numbers that matter to your business, you have a case for the next one; if it does not, you have learned cheaply.
What are the common mistakes?
- Treating the first generated shape as final instead of validating it.
- Ignoring how the part will be made, so the "optimal" shape cannot be produced or inspected.
- Using generic load cases instead of the real duty cycle and failure modes.
- Forgetting assembly, maintenance access and tolerance stack-up around the part.
- Expecting AI to replace design judgement; the engineer remains responsible for safety.
How can you learn generative design for free?
- Learn solid modelling with the free Fusion 360 3D Design course and SOLIDWORKS 3D Modelling.
- Browse all free CAD courses and the free generative AI courses to understand the language-model side.
- Add Python and optimisation thinking with Python for AI and Engineering Data and Machine Learning with Python.
- Practise model evaluation in the Machine Learning Engineer job simulation, which is practice, not employment.
- See what employers want on the mechanical design and CAD jobs page and the AI and machine learning jobs page.
- For the wider context, read about digital twins in manufacturing.
Start free
Start with the free Fusion 360 3D Design course, then try the first five free days of the Machine Learning Engineer job simulation to learn how models are evaluated, and look at the mechanical design jobs page. If you want a structured self-paced route into industrial AI with graded projects, see the EDWartens AI Module for its current price.




