Model-in-the-Loop Engineering · Capstone brief
ENG-03 · Tolerance stack-up meets ML
When a model’s output drives a mechanical decision, how does its error add to tolerance stack-up — and what is the total error budget in millimeters, degrees, or newtons?
The question
When a model’s output drives a mechanical decision, how does its error add to tolerance stack-up — and what is the total error budget in millimeters, degrees, or newtons?
System / materials
Teacher supplies a one-page stack diagram (3–5 components) from an academy project or a standard textbook problem made physical. Model may be a regression from measured samples or a provided curve fit — no black-box required.
Expected failure modes
Reporting model RMSE without converting to physical effect at the actuator. Ignoring worst-case stack direction.
Done looks like
Stack diagram, per-component tolerances, model error translated to the same units at the decision point, and a pass/fail against a teacher-set budget.
Five C's
CT: worst-case vs. statistical stack. CR: choosing where to measure. CO: peer checks unit conversion. CM: GD&T-style summary a machinist understands. CZ: who is near the moving part.
Mentor role
Mechanical or manufacturing engineer reviews stack at Checkpoint 1 and budget at Checkpoint 2. School-supervised.
Rubric calibration
R1: one decision point named. R2: measurements reproducible. R3: no-model stack shown. R4: worst case addressed. R5: machinist-readable. R6: refuses “close enough” without units.
Two ways this goes wrong
(a) Percent error with no physical translation. (b) Stack-up on paper only — no tie to the lab artifact.
Credit lane fit
Lane A immediately. No verified credit claim.