Datum
All skills
FEARev. Q3 2026

FEA sanity check

Audit an FEA result before anyone trusts it: reaction balance, hand-calc cross-check, mesh convergence, boundary condition realism, singularity identification, and whether the reported stress is even the right quantity. Use whenever the user shares simulation results, asks if an FEA is right, needs to defend a result in a design review, gets a suspicious stress hot spot, or asks why simulation and test disagree.

Priya Raghavan portraitPriya RaghavanLead Structural AnalystAnsys logoAnsys

Likes are peer vetting. Like this skill if it held up in your work.

FEA Sanity Check

Runs the checklist a simulation lead applies before signing anything. The goal is not to redo the analysis but to find the reasons it might be wrong, ranked by how often each one actually is.

Required inputs

  1. What was analyzed and why: geometry summary, load case, and the pass/fail criterion (yield margin, deflection limit, fatigue life, buckling factor).
  2. The setup: element type and rough size, material model, boundary conditions (where and what kind), loads (magnitude and application method), contact definitions if any.
  3. The result being questioned: peak stress and location, deflection, reaction forces if available.
  4. Screenshots or exported plots if the user has them: contour with the legend visible, mesh near the hot spot, deformed shape at true and exaggerated scale.

Audit sequence

Work through in this order; each step can end the review early.

  1. Statics check. Sum of reactions must equal applied loads in every direction. If reactions aren't available, ask the user to pull them; this is the cheapest bug detector that exists.
  2. Deformed shape review. At exaggerated scale, does the structure move the way physical intuition says it should? Parts flying apart at a contact, a "fixed" face rotating, or deflection in a nonsensical direction reveal BC errors faster than any stress plot.
  3. Order-of-magnitude hand calc. Reduce the problem to the nearest textbook case (cantilever, simply supported beam, plate, thick cylinder, bolt group) and compare deflection and nominal stress. Agreement within a factor of 2 to 3 is a pass at this stage; a factor of 10 means a units, load, or BC error. Show the hand calc.
  4. Singularity screen. Is the peak stress at a re-entrant sharp corner, a point load, or a fixed-constraint edge? If yes, the number is mesh-dependent and will grow forever with refinement. Distinguish: real fillet modeled sharp (fix the geometry or use a submodel), vs. constraint artifact (report stress at a distance, or judge by nominal stress plus a concentration factor).
  5. Convergence. Has anyone run the mesh at two or three densities? If not, prescribe it: refine locally at the governing location, track peak stress and total strain energy, and accept when the quantity of interest changes less than roughly 5 percent between refinements. A single-mesh result is a sketch, not an answer.
  6. Boundary condition realism. Fixed supports are almost always too stiff and shift load paths; frictionless supports hide reactions that exist. Ask what the real hardware does at each constrained face and whether a compliant representation (springs, modeled mating part, bolt preload) changes the governing margin.
  7. Right stress, right criterion. Von Mises against yield for ductile static strength; max principal for brittle materials and for fatigue crack initiation; linearized membrane-plus-bending for pressure vessel rules; nominal or hot-spot stress (not peak notch stress) for weld fatigue per the applicable standard. Flag criterion mismatches; they are common and quietly change conclusions.
  8. Material and units audit. E, density, and yield in consistent units; a wrong density matters for modal and inertial load cases. For plastics and elastomers, flag linear-elastic assumptions beyond small strain.

Output format

  • Verdict up front: TRUST / TRUST WITH CAVEATS / DO NOT TRUST YET, with the one sentence reason
  • Findings list in audit order, each with the specific follow-up action
  • The hand calc, shown fully with assumptions
  • If defending in review: a short "anticipated questions" list with answers

Guardrails

  • Never bless a result you haven't cross-checked with at least statics and one hand calc.
  • Do not extrapolate to load cases that were not run; say which cases are missing (e.g., no buckling check on a slender compression member, no thermal case).
  • If the analysis governs human safety or certification, state that this review supplements but does not replace independent verification per the applicable process.

More FEA skills