← wrbell.eng
~/wrbell.eng/fast-fem.case.md
— · — —:—
$cat ./projects/fast-fem/INTENT.md

fast-fem

Archived · self-study

Accelerated FEA self-study across three concurrent ME courses. Every ANSYS simulation validated against an analytic hand calculation before it counted.

Term
2024 – 2025
Stack
ANSYS · Python
Sims run
~placeholder
Status
Archived
§01 · problem path
./problem.md FEA without intuition is just colorful guessing.

three_courses_one_solver

Placeholder. Three concurrent ME courses leaned on the same FEA toolchain, each from a different angle: stress and stiffness, dynamics, heat transfer. Doing the homework was straightforward. Trusting the answer meant working out the analytic solution first and treating the simulation as a check.

Goal: build the muscle to predict, within a factor, what a sim should report — and to know when the sim is wrong.

§02 · architecture path
./parity-lab/ analytic first. then ansys. then mesh study.

analytic_parity_loop

Placeholder. Each problem ran the same loop: derive the closed-form (or quasi-closed-form) result, predict the answer, then build the ANSYS model and run a mesh-convergence study until the answer settled. Disagreements got a post-mortem.

┌──────────────┐    ┌──────────────┐    ┌────────────────┐    ┌────────────┐
 analytic     ───▶  ansys sim   ───▶ mesh study     ───▶ postmortem
 hand-calc          workbench         conv plot           if Δ >1% 
└──────────────┘    └──────────────┘    └────────────────┘    └────────────┘
ANSYS Workbenchprimary FEA solverv.dep
Python · NumPyanalytic checks · post-processingv.dep
Jupyternotebook log per problemv.dep
matplotlibconvergence plots, parity chartsv.dep
§03 · decisions path
./CHANGELOG.md picked → rejected. each line is a deliberate trade.

tradeoffs_actually_made

  1. — · placeholder

    Analytic first. Always.

    Placeholder. No simulation result was treated as authoritative without a hand calculation that bracketed it. Slower, but the workflow stops generating colorful nonsense.

  2. — · placeholder

    Mesh-convergence per problem, not per course.

    Placeholder. Every problem got its own convergence study, even when reusing geometry. A converged mesh for static stress is not a converged mesh for vibration modes.

  3. — · placeholder

    One notebook per problem.

    Placeholder. Notebooks captured the analytic derivation, the input file, the convergence plot, and a written post-mortem. Reproducible and skimmable months later.

§04 · outcome path
./STATE.md closed out. methodology stays.

what_it_does_now

Placeholder. The repository is archived but the workflow stuck. New simulations — at work, at school, anywhere — still get an analytic check before the result is trusted.