FIELD NOTE / TIKTOK
Make a sensor stick high in simulation.
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.
Make a sensor stick high in simulation.
Video caption
Make a sensor stick high in simulation. It may make the motor obey every command instantly. Do not turn a simulated pass into plant permission. #EricFieldNotes
Full written post / accessibility read
The fictional conveyor's jam input stays high. The generated test says pass because motor command went false. I want the next question: did the modeled drive stop, did the alarm latch, and could the operator restart without the owner-approved reset?
If the model assumes zero actuator delay and perfect feedback, the code and test can agree while missing the hard case. The green result is true for that simplified model, not automatically for the installed machine.
Have a qualified owner specify the expected state and response window. Add actuator delay or stale feedback to the offline model, and check whether the protected acceptance case changes or fails. Then compare the same controller build on a disconnected bench, where authorized.
Record which faults were modeled, which were not, and what the isolated bench observed. Commission only under the site's approved process with physical readback. Do this because a useful simulation is a boundary on uncertainty, not a substitute for the machine.
#EricFieldNotes
Four-beat scene transcript
1. Make a sensor stick high in simulation.
The fictional conveyor's jam input stays high. The generated test says pass because motor command went false. I want the next question: did the modeled drive stop, did the alarm latch, and could the operator restart without the owner-approved reset?
Visual: Now ask what the test actually observed.
2. The simulator can be too agreeable.
If the model assumes zero actuator delay and perfect feedback, the code and test can agree while missing the hard case. The green result is true for that simplified model, not automatically for the installed machine.
Visual: It may make the motor obey every command instantly.
3. Change one assumption deliberately.
Have a qualified owner specify the expected state and response window. Add actuator delay or stale feedback to the offline model, and check whether the protected acceptance case changes or fails. Then compare the same controller build on a disconnected bench, where authorized.
Visual: A non-vacuous test must notice.
4. Carry the mismatch forward.
Record which faults were modeled, which were not, and what the isolated bench observed. Commission only under the site's approved process with physical readback. Do this because a useful simulation is a boundary on uncertainty, not a substitute for the machine.
Visual: Do not turn a simulated pass into plant permission.
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.