AI Manufacturing Agent Development

Description

Two production lines making automotive components. Our quality inspection is visual, manual, and at the end of the line — which means when something drifts we find out after we have made four hours of it. The ask is an agent that watches the process data from the line PLCs, spots the drift before the parts are out of tolerance, and tells the line supervisor what changed. We are not asking for automatic intervention. Our operators will not accept a system that adjusts the line, and frankly neither will our safety case. You will need to work with our controls engineer to get data off the PLCs. It is Siemens S7, and the tag naming is inconsistent between the two lines.

• Establish data collection from the S7 PLCs on both lines and reconcile the inconsistent tag naming
• Model normal process behaviour per product variant, not per line
• Detect drift ahead of tolerance breach and alert the line supervisor with the changed parameters named
• Stay strictly advisory — no writes back to the PLCs, ever
• Validate against historical scrap events to show the alerts would have fired in time

• PLC data collection across both lines with a reconciled tag map
• Drift detection per product variant
• Supervisor alerting on the line HMI and mobile
• Historical validation report against known scrap events