Model-in-the-Loop Engineering · Capstone brief
ENG-05 · Digital twin skepticism
For a teacher-provided simple physical system, when does a learned residual beat a physics-first model — and when is the twin “good enough” only for some decisions, not all?
The question
For a teacher-provided simple physical system, when does a learned residual beat a physics-first model — and when is the twin “good enough” only for some decisions, not all?
System / materials
Same approved plant as ENG-04 or a lighter spring-mass / DC motor bench. Physics model may be algebraic; learned piece must show where it helps and where it lies.
Expected failure modes
Twin fits training maneuvers only. Claiming digital twin replaces inspection.
Done looks like
Side-by-side prediction error for physics-only vs. hybrid, with a written list of decisions each is allowed to inform.
Five C's
CT: fit-for-purpose, not fit-for-slide-deck. CR: choosing what the physics model skips. CO: peer re-runs one maneuver. CM: decision table for a mentor. CZ: who trusts the twin wrongly.
Mentor role
Engineer reviews modeling split at Checkpoint 1 and decision table at Checkpoint 2. School-supervised.
Rubric calibration
R1: one plant. R2: logged maneuvers. R3: physics-only baseline. R4: out-of-maneuver failure shown. R5: decision table clear. R6: refuses full autonomy.
Two ways this goes wrong
(a) Hybrid model hides a physics sign error. (b) No physical comparison — simulation only without teacher approval.
Credit lane fit
Lane A immediately. No verified credit claim.