Gains:
- Ability to position AI correctly in design idea generation, parametric thinking and CAD automation
- Ability to convert free text needs into measurable constraints and criteria with a structured prompt
- Ability to apply the discipline of validating designs and scripts produced by AI with engineering constraints
Mechanical engineering is the art of translating an idea into physical reality: coming up with a safe, manufacturable design that works, given a need, a set of constraints, and the laws of nature. Artificial intelligence (AI for short: software that produces text, code and suggestions like a human by learning from given texts) enters this process as a "design partner". Generates alternative ideas, accelerates parametric thinking (the approach of building the design with changeable variables instead of fixed dimensions), writes CAD macros and drafts documentation. But AI does not confirm a strength calculation, does not guarantee whether a tolerance will hold in assembly, and cannot "send" a design to production. All of these decisions are at the discretion and legal responsibility of the engineer. In this unit, you will learn where to put AI in the design and CAD (computer aided design) workflow and how to validate its output.
Where Does AI Stand in the Design Process?
Let's roughly divide a design process into four phases and clearly position the roles of the AI and the engineer. This distinction forms the basis of the question, “What would I make AI do, and what would I never do?”
Stage
The role of AI (accelerator)
Engineer's role (decision)
Requirement/concept
Alternative concept, similar solution examples, list of questions
Determine constraints, priority and acceptance criteria
preliminary design
Parametric relationship, rough dimensioning sketch
Validate calculation, assumption, and load path
detail design
CAD macro/script, standard part recommendation
Geometry, tolerance, material decision
Verification/approval
Report draft, checklist, summary
Analysis, testing, signature and liability
Critical principle: AI is strong at divergent thinking (generating large numbers of ideas) but weak at convergent decision-making (questioning which of these is right, safe, producible). Convergence; It requires knowledge of physics, engineering intuition and responsibility. This is core engineering work that cannot be automated.
Tip: Use AI as an “idea multiplier,” not an “answer machine.” Saying "suggest 5 different design approaches for this link, write down the pros/cons and weak points of each" is much more valuable than saying "give the best design"; because it leaves the decision and justification to you.
Step by Step: From Need for Free Text to Validated Design
- Make the requirement measurable. "Be light and strong" is not a requirement; "mass < 2 kg, factor of safety at 500 N vertical load ≥ 2, outdoor corrosion" is a requirement.
- Mark the ambiguities. Don't let AI make up values that aren't given; Have it listed as "needs clarification".
- Generate lots of concepts. At least 3 alternatives and each of them has its weak point with its load path logic.
- Create a verification plan. A "which calculation should I do" list for each concept.
- Add CAD automation. Have macros/scripts generated for repetitive modeling.
- Verify and confirm. Independently check measurement, unit, mass and critical calculations; then sign.
Prompt that converts the requirement into constraints and criteria
Role: You are an experienced mechanical design engineer. Task: Translate the following free-text requirement into measurable design requirements (constraints + criteria). Requirement: "An easy-to-assemble carrier bracket for a conveyor that will operate outdoors with a 30 kg load." Output format: - Functional requirements (what to do) - Load/environmental constraints (force, temperature, corrosion, vibration) - Measurable criteria (mass, safety coefficient, cost target)- Points that remain unclear and need to be clarifiedRule: Numerical value DO NOT FIT; Mark omitted values as "must be clarified".
The last rule is important: if the AI spontaneously "assumes" a load value that is not given, your design is on the wrong footing. Marking uncertainties strengthens requirements gathering discipline.
Prompt that produces many concepts
Suggest 3 different concepts for the following element: vertical static load of 30 kg, wall mounting with two M8 bolts, outdoor environment (corrosion), target mass < 1 kg.For each concept:- load path (how the force is transferred to the bolts)- possible material family- estimated weakest point- list of calculations I need to doRule: Number fitting; I will verify the accounts.
Parametric Thinking and CAD Automation
Modern CAD tools (SolidWorks, Fusion 360, CATIA, FreeCAD) work parametrically and script-based. AI is very powerful at macro/script generation for repetitive modeling tasks: generating a bolt pattern, deriving a part family with a table, bulk extracting mass properties. Thus, hours of repetition work is reduced to minutes.
CAD script generating prompt
Write a script for the FreeCAD Python console: create a box profile with parameters length, width, height and corner radius, and print its mass properties (volume, center of gravity). Let the material density be the parameter (kg/m^3). Explain the code with comments.Rule: Specify the unit of measurement (mm) at the beginning; Maintain unit consistency. Rule: Calculate the volume with the analytical formula and print it on the screen so I can verify.
Caution: Even if the AI's CAD script "works" geometrically, it may be wrong from an engineering perspective: wrong unit (mm/m confusion), wrong axis, an intersection that doesn't happen. After running the script, compare the dimensions and mass of the produced model with a known reference.
Design review prompt
Critique the following design decisions through the lens of a design review. Don't disagree with me; List weaknesses, risky assumptions, and missing calculations. Write "why risk" and "how do I verify" for each item. Design: [bracket's material, thickness, bolt, load information]
Weak Prompt / Strong Prompt
Weak prompt:
Design me a bracket.
No loads, materials, mounting interfaces and constraints; AI produces a general, dimensionless and unverifiable explanation. The output is apparently full but the engineering value is low.
Powerful prompt:
Suggest 3 different concepts for the following bracket: 30 kg vertical load, wall mounting with two M8 bolts, outdoor environment (corrosion), target mass < 1 kg. For each concept: load path logic, possible material, estimated weak point and what calculation I should make. Don't make up numbers; I will verify the account.
The second prompt gives the constraints, asks for multiple concepts, and leaves verification to the engineer with the question "which calculation should I do?" This positions AI as a tool that accelerates, not takes over, responsibility.
Three Mini Cases (By Numbers)
Case 1 - Pump leg, resonance trap. Engineer Ayşe is designing a vibration-damping stand for a pump rotating at 1450 rpm (i.e. 24.2 Hz). AI recommends elastomer wedge and gives a hardness of k = 200 N/mm. Ayşe calculates the natural frequency herself for a 40 kg mass: f = (1/2π)·√(k/m) = (1/2π)·√(200000/40) ≈ 11.3 Hz. Isolation works since the excitatory frequency (24.2 Hz) is approximately twice the natural frequency, but Ayşe improves the isolation by going down to f ≈ 7.5 Hz with a softer wedge (k = 90 N/mm). The initial value of AI “worked” but was not optimal; accounted for the decision.
Case 2 - Part family with CAD macro. A company produces flanges of 12 different sizes. The engineer writes a Fusion 360 script to the AI and automatically derives the table and 12 variants; The work that takes 3 days manually is reduced to 2 hours. But in the DN80 flange produced by script, the bolt hole diameter is 18 mm; whereas for catalog M16 the 18 mm hole is correct. The engineer verifies the measurements of a random variant against the standard, then accepts all.
Case 3 - Unit trap. An intern mistakenly left the density as 7.85 g/cm³ instead of 7850 kg/m³ but the volume in mm³ in the script given by the AI. The resulting mass deviates by a factor of 1000: 3200 kg instead of 3.2 kg. The order of magnitude check ("this bracket cannot be 3 tons") catches the error immediately. Lesson: unit consistency is the most frequent and dangerous source of error in AI output.
Common mistakes
- Leaving convergence to AI: Saying "give me the best design" and handing over physics and responsibility.
- Leaving uncertainty to be assumed: Letting the AI make up the load/temperature not given.
- Not verifying the CAD script: Not comparing the size and mass of the produced model with the reference.
- Unit confusion: setting up the model/calculation regardless of mm/m, kg/N or g/kg interference.
- Locking in a single concept: Thinking that the first idea of AI is the only option; Not producing alternatives.
- "It worked = correct" fallacy: Confusing the script's error-free operation with the result being correct.
In summary
- In design, AI is strong in divergent thinking (idea generation) and weak in convergent decision (physics, security); The decision is up to the engineer.
- Good design begins with translating the vague desire into measurable constraints and criteria; AI helps in this translation.
- AI in CAD automation generates powerful script/macro; But be sure to verify the units and dimensions of the model.
- Have the AI generate multiple concepts and ask “which calculation should I do?”; undertake verification.
- AI does not approve a design or send it to production; This responsibility remains with the engineer.
Application task
Write a free text requirement for a simple machine element (bracket, headstock, cover, etc.). Have the AI translate this into measurable requirements (constraint + criteria) and mark any values not given as “must be clarified”. Then ask the AI for at least three concepts and a list of “calculations I need to do” for each. Critical verification: Select a numerical value (size, hardness, material property) suggested by the AI and verify it with a hand calculation or catalog/datasheet; If it doesn't match, write why. Finally, list in one sentence what engineering approvals are required before putting this design into production.
checklist
- [ ] The need for free text has been translated into measurable constraints and criteria.
- [ ] Values not given are marked as "must be clarified"; AI didn't make it up.
- [ ] At least three concepts and verification (calculation) lists for each were produced.
- [ ] The units and dimensions of the CAD script/macro output were compared with the reference.
- [ ] Sanity check has been performed.
- [ ] The final decision and approval was left to the engineer; It has not been delegated to AI.