Specialty centers · Model-in-the-Loop Engineering
Free engineering briefs where a model is a component with tolerances and failure modes — not a magic accuracy score. Students write the safety case for handing one actuator decision to a model, including when a human must stay in the loop. Every brief has its own PDF, and all twelve are in the bank PDF. No account. No student data leaves your school.
Shop rule
Teacher-approved plant only — one bounded actuator, hardware interlock or operator confirm on anything that can injure someone. Students document safety cases and refusal logs; they do not wire unsupervised model output to production equipment.
How to run one
Worked sample · printable PDF
Your academy lab has a single actuator decision a model might trigger — open a damper, divert a part, enable a motor stage, release a brake, or raise a warn-only alarm. Can you write the safety case for letting the model request that action — including the exact conditions under which a human must remain in the loop, the acceptance tests that must pass first, and what you refused to automate?
Featured · printable PDF
Given a teacher-approved physical or simulated plant and one logged sensor, can you (1) separate physics from filter/model invention, (2) report uncertainty in engineering units, and (3) name a real fault a maintainer would miss if they trusted the processed stream — without claiming the plant is production-ready?
On a public equipment-degradation dataset the teacher approves, can you beat a dumb time-based maintenance rule — and show whether the model’s false alarms cost more than the failures it prevents?
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?
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?
When you inject named sensor faults into an approved loop, does the controller fail gracefully (detected, safe state) or silently (wrong output, no alarm)?
For the same model task on edge vs. cloud paths the teacher approves, what is end-to-end latency — and does it fit inside the control-loop deadline for your named decision?
After a teacher-defined distribution shift (seasonal temperature, wear, or load change), can you detect that the model’s errors drift — and say whether recalibration or human takeover is the right response?
When you design the operator interface for a model-assisted decision, do operators over-trust the score — and can you measure that with a short, ethical user test?
Can you map your design review to one safety or quality standard clause your mentor recognizes — and state what is not covered by your project?
Can you scope one brief from this bank so it hits your academy’s existing capstone checkpoint calendar — without inventing a new course or skipping a required gate?
Pick a use case where the correct engineering answer is not to deploy the model — and defend that refusal with evidence, not vibes.
Six-dimension rubric
Scope unclear? Safety language off? Tell us here — you do not have to leave an email.