How a Tier 1 Auto Parts Supplier Reduced Scrap Rate by 42%

The supplier runs three progressive die stamping lines producing structural brackets for a major OEM. At 3.8% scrap rate on their primary bracket family, they were scrapping roughly 340 parts per 8-hour shift. Material cost on these parts runs $4.10-$6.80 depending on gauge. That was $1,400-$2,300 in scrap per shift before labor and disposal. Across three shifts, five days a week, the annual scrap cost for that bracket family alone was approaching $1.8 million.

Their quality team knew where the defects were coming from. Roughly 60% of scrap was being caught at end-of-line inspection - a station with two inspectors evaluating stamped parts for burrs, edge cracks, surface marks, and dimensional compliance against a tolerance spec. The other 40% was being caught by the customer or discovered at assembly, both significantly more expensive failure modes.

The Inspection Bottleneck

End-of-line inspection was a genuine constraint. The two inspectors were evaluating approximately 1,100 parts per hour at a cycle time that matched line output. Adding more inspectors would have cost them $85,000-$95,000 per year in labor and would not have solved the fundamental detection gap. Their own internal audits showed manual inspection missing 17-22% of surface defects that were within their reject criteria.

The root problem was that defects were also escaping upstream. Without in-process data, the quality team had no visibility into when a stamping defect pattern was developing until it showed up at end-of-line. By then, they had run 200-400 defective parts and accumulated scrap before anyone adjusted the tooling or press parameters.

Deployment Configuration

We installed two inspection stations: one after the progressive die press and one at end-of-line. The in-process station used a four-camera configuration with coaxial LED illumination to catch surface marks and edge burrs immediately after forming. The end-of-line station added two additional cameras for full-perimeter coverage and dimensional verification against nominal dimensions.

Camera hardware: 12MP global shutter area scan sensors at 300mm working distance, giving approximately 18-micron pixel pitch on the 220mm x 180mm bracket face. Frame rate at 15fps with 120-microsecond strobed LED exposure to handle 250mm/second part indexing on the exit conveyor. Detection threshold calibrated to reject parts with surface marks exceeding 0.4mm width or edge burrs above 0.6mm height.

Model training used 4,200 labeled images across six defect classes. The customer's quality team had good historical records of defect morphology, which accelerated training significantly. Initial calibration took four days on-site. Production qualification ran for two weeks with parallel manual inspection to validate false reject rates before removing the manual station.

What the Data Showed

Seven months post-deployment, scrap rate on the primary bracket family had dropped from 3.8% to 2.2%. That is a 42% reduction. More importantly, the composition of the remaining 2.2% scrap changed. End-of-line now catches almost everything - customer escapes dropped from an average of 14 per month to 2. The two escapes in the most recent full quarter were both the same root cause: a new defect type not in the training set that required a model update.

The in-process station delivered something the end-of-line station could not: early warning. During the monitoring period, the system generated 23 tooling alerts when burr frequency on specific die stations crossed a threshold. Average time from alert to press adjustment: 18 minutes. Average number of defective parts produced before the alert: 47. Before deployment, the same tooling degradation would have run undiscovered until end-of-line pulled a cluster of rejects, typically after 200-400 parts.

Financial Outcome

The 1.6 percentage-point scrap reduction translates to approximately 144 fewer scrapped parts per shift. At average material cost of $5.30 per part, that is $763 per shift, $2,289 per day across three shifts. Annually, $594,000 in material cost avoided on this bracket family alone.

The two manual end-of-line inspection positions were not eliminated. One was redeployed to a different line that needed additional coverage. One role transitioned to a vision system operator position responsible for monitoring detection performance, managing model updates, and handling quarantined parts. Net labor change: zero, with significantly better utilization of both positions.

Total deployment cost including hardware, installation, training, and first-year support: $218,000. Payback period on material savings alone: approximately 4.4 months. That calculation does not include warranty claim reduction or the value of catching defects in-process rather than at the customer.

What Did Not Go Smoothly

The first six weeks had a higher false reject rate than targeted. The primary cause was part-to-part variation in surface reflectivity from the stamping lubricant - parts with heavier lubricant application were triggering the surface mark threshold on areas that were clean. Adjusting the coaxial illumination intensity and recalibrating the detection threshold for the lubricant reflectance range resolved it. False reject rate settled at 0.4% by week eight, within acceptable range for their operation.

The in-process station also required a housing modification after week two. Stamping produces significant oil mist, and the original IP54-rated enclosure was accumulating contamination on the lens port within 12 hours. Upgrading to IP67-rated enclosures with positive air purge resolved the contamination issue permanently.