Defining the Engineering Question Before Building the Model
The most important FEA best practice: understand what engineering decision the model must inform before choosing geometry, elements or mesh density.
Principle
The model should be driven by the engineering question, not by the available geometry or software capability. A more detailed model is not automatically a better model.
Why it matters
Finite-element analysis can produce precise-looking results from poor assumptions. If the engineering question is not clearly defined before modelling begins, the analyst risks producing a model that is elaborate but irrelevant — answering a different question from the one the engineering decision actually requires.
Good practice
- Write down the specific engineering question before opening the pre-processor: "What is the maximum stress at the blend radius under ultimate load?" not "Analyse the bracket"
- Identify the decision the result will inform: sizing, qualification, root-cause investigation, optimisation, correlation
- Determine the required output quantity and location before selecting mesh density, element type and model fidelity
- Identify what accuracy is needed — a sizing study may tolerate 10% error; a qualification margin of +0.05 may not
- Consider whether a hand calculation or simpler model would answer the question more transparently
Warning signs
- The analysis plan says "run a stress analysis on the bracket" without specifying the output location or decision
- The model includes every geometric feature when only one region is critical
- The analyst cannot state what failure mode governs the design
- The model fidelity is chosen based on available CAD geometry rather than the engineering question
Verification checks
- Can the engineering question be stated in one sentence with a specific output and location?
- Does the model fidelity match the required accuracy of that specific output?
- Could a simpler model answer the same question with less uncertainty?
- Has the analyst identified the expected result before running the model (an order-of-magnitude estimate)?