IA ci Quiz defar

Sooy def quiz bi, tontu yi ak leeral yi dañuy feeñ.

1. A factory adds a camera model to flag possible surface defects. What should the model’s role be defined as?
   • Detection support, recommendation, or control—because each has different risk
   • A single category called “AI”
   • A replacement for every safety interlock
   • A guarantee that products meet every specification
2. Why should manufacturing test data cover shifts, products and environmental conditions?
   • A large training set guarantees coverage
   • A camera sees the same image under every condition
   • Those conditions can change the inputs and expose performance gaps
   • The model learns calibration without data
3. A defect model catches more flaws but creates an inspection queue no team can handle. What does this show?
   • More alerts always improve quality
   • The model is correctly calibrated
   • Human inspection is no longer needed
   • Detection performance alone may not improve the overall process
4. Why can a random split of sensor readings overstate performance on a new production line?
   • Randomization always prevents leakage
   • Correlated readings from the same machine or period may leak across train and test
   • The model cannot use sensor data
   • New lines have no operating conditions
5. Which safeguard belongs around an AI recommendation that could change a machine setpoint?
   • Deterministic interlocks, authorized procedure, and an operator override
   • A larger dashboard
   • Automatic execution without logging
   • Removing the recovery plan
6. What should a team check when a manufacturing process changes?
   • Only whether the model file size changed
   • Assume previous evaluation still applies
   • Delete all original signals
   • Whether data distributions or defect labels changed and whether the model still performs
7. Which measure helps distinguish a useful detector from one that creates too many interruptions?
   • Only training accuracy
   • Only the number of cameras
   • False alarms, missed defects, downtime, scrap, and worker burden
   • Average model latency alone
8. Why preserve the original sensor signal and model version for an incident review?
   • To make the model more accurate automatically
   • So the team can reconstruct what evidence and model produced the alert
   • To remove the need for operator notes
   • To guarantee no defect occurred