A control plan is a table. It lists what gets checked during production, how, how often, and what happens when something is off.
In the auto industry, AIAG publishes a Control Plan reference manual. It describes three phases: a prototype control plan, a pre-launch control plan and a production control plan. The production plan is a living document, updated as you learn.
What goes in it
The usual columns:
- Characteristic. What you're checking.
- Specification. The target and the limits.
- Method. How it's measured.
- Frequency. How often. Every piece, every hour, one in fifty.
- Reaction plan. What happens when it fails.
AIAG's manual is specific about the reaction plan: contain the suspect product, stop the process from making more, bring the process back under control, and name who owns the reaction.
Special characteristics
Some characteristics matter more: safety, regulations, critical function. These are special characteristics. AIAG's manual says they all go in the control plan, and at a minimum the design failures with severity 9 or 10 are among them.
How a bot uses it
Take a bot that writes a weekly status report. Its control plan might look like this:
| Characteristic | Specification | Method | Frequency | Reaction plan |
|---|---|---|---|---|
| Figures have sources | Every number links to a file | Bot checks each number | Every report | Hold the report, flag the figure, ask |
| Owners named | No action without an owner | Count blank owners | Every report | Fill from the action log or mark "owner needed" |
| Current file version | Uses the latest plan | Compare revision on file | Every report | Stop, reload, redo |
| Delivered for review | Draft, not sent | Check it's a draft | Every report | Never send. Report the miss to the owner |
| Report accuracy (special) | No wrong figures reach the customer | A person spot-checks | Weekly | Pull the report, fix the cause, log it |
Now "be careful" has become five checks with written reactions.
Why it helps
Bots are told to "double-check" all the time. That isn't a control. It doesn't say what to check or what to do if the check fails. A control plan does both. The reaction plan matters most: it tells the bot to stop and hold, not to quietly patch and carry on.
Where bots differ
A factory control plan measures physical parts with calibrated gauges. A bot mostly checks its own work, which is like asking a machine to inspect its own parts. Useful, but not independent. Keep at least one check that a person or a separate tester does. See Your tester bot is your internal auditor.
Back to Engineering tools for bots.