Back to case studiesCase study · 01 / 03

Defect detection across a 12-line production floor.

39%

Less downtime

96.6%

Defect detection

5 wks

To production

12

Lines covered

The challenge

Where operations were breaking down.

Visual inspection was a manual handoff between shifts, and defects often surfaced downstream, at packing, or worse, at the customer site. The recall risk was rising alongside warranty cost, and quality engineers spent half their week chasing root cause instead of preventing recurrence.

Downstream maintenance compounded the problem. Unplanned line stoppages averaged 3.4 hours per week per line, with no central record of which faults preceded which failures. Operators had instincts, but no shared signal.

Leadership wanted a system that could flag both quality defects and equipment anomalies before they triggered a stoppage, without ripping out the existing MES or PLC tooling.

Our approach

How Aelix Forge was deployed.

01

Computer-vision quality inspection

Camera arrays at four critical inspection points feed a defect-classification model trained on three years of historical reject samples. Confidence scores route ambiguous cases to a human reviewer.

02

Sensor telemetry ingestion

PLC and SCADA streams pipe into the operational data lake. Anomaly-detection models flag deviations in vibration, temperature, and cycle time before they cross failure thresholds.

03

Shift-level dashboards

Floor supervisors get real-time dashboards showing defect rates per line, top recurring root causes, and predictive maintenance windows. Mobile-responsive for walking the floor.

04

MES writeback + audit

Every detection writes back to the existing MES with full audit lineage, so quality and compliance reporting stays intact without parallel systems.

Architecture in production

From source data to operational action.

Stage 01

Vision + telemetry

Cameras and PLC sensors stream raw signals continuously from every line.

Stage 02

Ingest + normalize

Edge gateways unify camera frames, sensor packets, and shift logs into a single feature store.

Stage 03

Defect + anomaly models

Vision and time-series models score each frame and signal window in real time.

Stage 04

Alerts + writebacks

Floor supervisors get pushed alerts; MES records every detection with audit lineage.

Measured outcomes

Operational gains after going live.

0%

Less downtime

Predictive alerts catch sensor anomalies before they cascade into line stoppages.

0.0%

Defect detection rate

Vision models outperform manual inspection on consistency across shifts.

0%

Fewer recalls

Caught upstream during inspection instead of downstream at the customer site.

5 wks

From kickoff to production

Integration with existing MES and PLC infrastructure required no replacements.

3.4 → 0.9

Hours/week of unplanned stoppage

Per line, averaged across the 12 lines covered in the initial rollout.

$1.8M

Annualized savings

Combined recall avoidance, scrap reduction, and uptime improvement in year one.

See what Aelix Forge would do for your operations.

Talk to our team about the bottlenecks slowing you down, and the system we'd deploy to remove them.

See more case studies