引导式测验 · 中等 等级

制造业中的人工智能测验

测试生产限制、代表性数据、误报、安全联锁、漂移、操作员审查、恢复和性能测量。

相关引导路径人工智能制造
问题 1 的 8

一家工厂添加了一个相机模型来标记可能的表面缺陷。模型的角色应该定义为什么？

预览每个问题8 问题

当您参加测验时，就会出现答案和解释。

  1. 一家工厂添加了一个相机模型来标记可能的表面缺陷。模型的角色应该定义为什么？
    • 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. 为什么制造测试数据应涵盖班次、产品和环境条件？
    • 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. 缺陷模型可以捕获更多缺陷，但会创建团队无法处理的检查队列。这说明什么？
    • 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. 为什么传感器读数的随机分割会夸大新生产线的性能？
    • 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. 哪些保护措施属于可能改变机器设定点的人工智能建议？
    • 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