Skilled inspection shifts from broad sampling to anomaly disposition.
AI value · Inspection coverage
Company-reported
AI analyzes about 1.5 million welds on 300 vehicles each shift
Operator coverage, not defect escape-rate evidence.
Before
Uses ultrasound to inspect a random sample of resistance spot welds. → Reviews sampled findings and directs investigation or rework.
After
Analyzes process data for approximately 1.5 million welds across 300 vehicles each shift. → Focus on flagged anomalies and continue random ultrasound validation.
Human boundary
Audi sets thresholds; humans disposition anomalies.
Why it matters
Quality can move from ultrasound sampling to AI screening of every weld.
This case is company-reported. Use it for the operating-model shift; do not treat the numbers as independently measured.
How the work changed
Before
How the work ran before the change.
Step 1 of 2
Audi quality technician
Uses ultrasound to inspect a random sample of resistance spot welds.
ControlAbout 5,000 weld checks per vehicle reported; sample-based control remains in parallel.
Step 2 of 2
Quality engineer
Reviews sampled findings and directs investigation or rework.
ControlAudi quality procedures and human disposition.
What changed
Quality can move from ultrasound sampling to AI screening of every weld.
Decision rightHuman sets the threshold; AI decides each instance
After
How the same work runs now.
Step 1 of 2
WPS Analytics
Analyzes process data for approximately 1.5 million welds across 300 vehicles each shift.
ControlMachine-learning screening covers the population; traceability designed with quality/Fraunhofer bodies.
Step 2 of 2
Quality technician and engineer
Focus on flagged anomalies and continue random ultrasound validation.
ControlHumans own disposition; full coverage is not proof of zero defects.
Process model built from the published workflow evidence for Audi Neckarsulm. Every step, actor, and control appears in full below.Every step, actor, and control
Exception path
Flagged or uncertain welds receive manual ultrasound review.
Work removed
Routine review of non-anomalous welds
Decision authority
Audi sets thresholds; humans disposition anomalies.
Before
01
Audi quality technician
Uses ultrasound to inspect a random sample of resistance spot welds.
Control: About 5,000 weld checks per vehicle reported; sample-based control remains in parallel.
02
Quality engineer
Reviews sampled findings and directs investigation or rework.
Control: Audi quality procedures and human disposition.
After
01
WPS Analytics
Analyzes process data for approximately 1.5 million welds across 300 vehicles each shift.
Control: Machine-learning screening covers the population; traceability designed with quality/Fraunhofer bodies.
02
Quality technician and engineer
Focus on flagged anomalies and continue random ultrasound validation.
Control: Humans own disposition; full coverage is not proof of zero defects.
Work that left the path
Routine review of non-anomalous welds
Human role before
Quality technicians ultrasound-tested a random sample of spot welds, and quality engineers reviewed findings and routed exceptions to investigation or rework.
Human role after
Quality staff investigate anomalies and continue ultrasound validation.
AI roleAnalyzes resistance-weld process data for every spot and flags anomalies.
Outcomes
Inspection coverage
Company-reported
Random checks around 5,000 welds per vehicle→AI analyzes about 1.5 million welds on 300 vehicles each shift
2023 rollout · 300 vehicles per shift
Operator coverage, not defect escape-rate evidence.
What leaders can reuse
Anti-pattern
Do not infer zero defects from 100% coverage.
Questions
01Where is the operating threshold set and who can override it?
02What measured result would trigger rollback or retraining?
03Which residual decisions must remain human-owned?
Portability conditions
Traceable data
Validation sample
Certified process
Reputation risk
low
Evidence and authority
What the public record supports.
Current · updated
1 independent, 1 primary; publication outcomes are reported.
Bundle 1.0.0 · reviewed 2026-08-23 · stable ID 8ed0ede9e7831b2f
