EngineeringNewVerified Real deployment at a named or precisely described company, with a measured result from a primary source.

Generative AI trained on past designs proposes new product designs that meet cost and manufacturability targets (Eaton)

Added 2026-10-16 · Last updated 2026-10-16

Role

Engineering / R&D (product design) · Typical users: design engineer, product development manager, cost engineer, DFM / manufacturing engineer

The problem

Customer-specific products take months to design because each request needs input from several disciplines and many manual iterations before the design meets cost, performance and manufacturing requirements, and customers wait. At Eaton (power management; 92,000 employees), customers regularly ask for custom components such as a unique passenger-car valve stem or a new lighting fixture; a manual design that is cost-effective and meets manufacturing specifications takes months, and a lighting fixture can take up to 16 weeks because it needs thermal, electrical, mechanical, optical and manufacturing engineering.

How AI helps

Eaton trained its generative AI on actual historical product design data and simulation results, including cost and design-for-manufacturing (DFM) analysis from aPriori that runs programmatically in the cloud. The AI runs thousands of design iterations in minutes and proposes the top five designs; those go through a high-fidelity simulation, and Eaton's digital design and engineering team reviews them against the requirements before any design is released.

Step by step

1. Pick one product family with many past custom variants and measure today's design time per request, from request to released design.
2. Collect the past variants as data: design parameters, requirements, test or simulation results, cost and any manufacturing problems found later. Eaton calls this an authoritative source of truth.
3. Make cost and manufacturability checkable by software: a should-cost model, DFM rules from your plant (minimum wall, bend radius, tolerances you can hold) or a cost tool with an API.
4. Parameterize the design in CAD (configurator or script) so variants can be generated automatically, then generate and screen many candidates against the rules and cost.
5. Send only the top few candidates to full simulation, and have the responsible engineers review each one against every discipline's requirements.
6. Release through your normal design review and prototype or test as usual; record design time per request and compare with the baseline.

Tools

• aPriori aP Pro (cost, DFM and sourcing simulation run programmatically): used at Eaton, with Eaton's own generative AI and simulation tools
• Other tools to evaluate: generative design in Autodesk Fusion, PTC Creo or Siemens NX; nTop; Ansys and Simcenter simulation with AI; should-cost tools; CAD configurators for product families
• Any LLM (ChatGPT, Claude, Copilot) with the prompt below, to frame a design-automation pilot from your past variants before buying software

As reported by the source; most figures are vendor or self-reported and not independently verified.

Human review: cautions

• The figures come from the software vendor's case study and a trade article repeating them; Eaton describes them as early results.
• Generative AI only proposes designs: engineers check every candidate in high-fidelity simulation and design review, and physical tests still apply where safety or certification requires them.
• The approach depends on clean historical data; with few or messy past designs, start by building that data set.
• Cost and manufacturability scores are only as good as the cost model and DFM rules behind them; validate them against real quotes and production results.
• Do not paste customer drawings or confidential designs into a public chatbot; use an approved enterprise tool.

You are a senior product development engineer helping a manufacturer set up a design-automation pilot for one product family.

I will paste: (1) a table of our past variants of this product [variant ID, key design parameters, customer requirements, test or simulation results, unit cost, manufacturing issues found]; (2) a new customer request [requirements and constraints]; (3) our manufacturing rules [processes we use, tolerances and features we can and cannot make].

Do this:
1. From the table, list the design parameters that seem to vary with cost and performance, quoting the variants that show it. Do not fit a model or compute correlations; describe what you see and say how confident you are.
2. List the requirements of the new request by discipline (mechanical, thermal, electrical, manufacturing) and the ones that are missing or ambiguous, as questions for the customer.
3. Propose the parameters to vary in a first set of automated variants and a sensible range for each, based only on our past variants, and say which past variant is the closest starting point.
4. List the checks each candidate must pass before simulation (our manufacturing rules), and what must be simulated or tested before release.
5. Draft a one-page pilot plan: product family, baseline design time to measure, data to clean, tools needed, who reviews.

Do not calculate costs, stresses or temperatures yourself; say which tool or test should produce each number.
Use only the data I paste in; if something is missing, write 'not found' or list it as a question instead of guessing. Do not send, order, post or change anything in any system: produce drafts for my review and approval.

Related cases

• AI predicts simulation and test results from your past data, so engineers run fewer prototypes
• Find existing parts before designing or buying a new one (AI shape search)
• AI quoting from drawings and RFQ emails
• AI pulls BOM data from supplier documents and checks it against stock and specs

All 6 Engineering use cases →

Sources

• https://www.apriori.com/resources/case-study/eatons-generative-ai-cuts-product-design-time-by-87/
• https://www.assemblymag.com/articles/99091-generative-ai-slashes-design-time