Exceptions while the work is still on the floor
A manufacturer. Name withheld.
Quality problems showed up after the line had moved. Reviewers worked from spreadsheets, camera notes, and a shift log. There was no one queue that said this batch looks wrong, and here is why. By the time a person saw it, the work was downstream.
What was deployed
Mapped the quality workflow: which signals exist, who reviews an exception, and what “stop” is allowed to mean.
Deployed anomaly detection on that stream, in the shape of an operating system for the plant. Signals in. A queue of exceptions out.
A person accepts or rejects each exception. The model does not scrap a batch by itself.
What changed
of exceptions a person still accepts or rejects
Exceptions surface while the work is still on the floor, not in a weekly report. We do not publish a defect rate, a scrap percent, or a downtime figure for this client. The return is the queue, and a review that happens before the line ships the problem onward.
No. This engagement is quality and anomaly detection: a queue of exceptions on the line. Forecasting is a different workflow.
No. It puts an exception in front of a person. The person accepts or rejects it.
Start with the workflow.
Week 1 is a mapping of one workflow. If you already know the system and want a read on the shape, ask for an architecture review.