FIELD NOTE / TIKTOK
The twin passed because the valve never sticks.
The complete written thought and the evidence behind it. The video edition will follow its public release.
The written argument is here.
This approved TikTok edition is on the journal now. Its video player and original platform link will appear after each public release is verified.
The twin passed because the valve never sticks.
Video caption
The twin passed because the valve never sticks. The process may continue after the output changes. Use it to prepare, then observe the installed system. #EricFieldNotes
Full written post / accessibility read
Picture a fictional process valve. The control logic commands close, and the twin always moves it immediately. The test passes. Now introduce a stuck valve or delayed feedback in the offline model. The physical acceptance question suddenly changes.
A program can set the close bit correctly while a valve fails to move. If the oracle checks only the bit, it will stay green. The owner must specify feedback, process-state and time-window evidence for the actual hazard and plant design.
In an offline model, hold the valve open or delay its feedback. Compare expected states written by a qualified owner with the controller response. If a test still says pass, its oracle is too narrow. Record the gap for isolated bench and authorized commissioning plans.
A twin earns trust for each fault class it represents and validates. Let it find failures early, but require physical acceptance under the site's process for the final claim. Do this because a simulated perfect valve is a model assumption, not a plant fact.
#EricFieldNotes
Four-beat scene transcript
1. The twin passed because the valve never sticks.
Picture a fictional process valve. The control logic commands close, and the twin always moves it immediately. The test passes. Now introduce a stuck valve or delayed feedback in the offline model. The physical acceptance question suddenly changes.
Visual: An absent fault cannot fail a simulated test.
2. Commanded and actual state are different.
A program can set the close bit correctly while a valve fails to move. If the oracle checks only the bit, it will stay green. The owner must specify feedback, process-state and time-window evidence for the actual hazard and plant design.
Visual: The process may continue after the output changes.
3. Mutate the twin, not the real plant.
In an offline model, hold the valve open or delay its feedback. Compare expected states written by a qualified owner with the controller response. If a test still says pass, its oracle is too narrow. Record the gap for isolated bench and authorized commissioning plans.
Visual: A fault test should turn the verdict red or narrow it.
4. Keep the twin's claim bounded.
A twin earns trust for each fault class it represents and validates. Let it find failures early, but require physical acceptance under the site's process for the final claim. Do this because a simulated perfect valve is a model assumption, not a plant fact.
Visual: Use it to prepare, then observe the installed system.
Research and claim limits
- Siemens Automation Framework documentation (S140)
- NIST: Credibility Consideration for Digital Twins in Manufacturing (S141)
The examples identified as illustrative or simulated are design probes, not reported incidents. Vendor specifications do not establish workload performance.