Tier-1 supplier cuts transmission warranty costs 30% with unsupervised ML on EOL test data

A leading Tier-1 transmission supplier needed to reduce warranty claims by detecting manufacturing defects using end-of-line (EOL) test data. Each EOL test involved over 100 steps, but their existing SPC system analyzed only 10% of collected data. Engineers had to manually inspect thousands of signals to detect subtle issues, and there were only 100 training units with no recorded failures.

Acerta deployed LinePulse's unsupervised learning algorithms to calculate abnormality scores for each transmission. Used non-polynomial feature extraction to reduce dimensionality. The model was trained on 100 units without labeled failures, detecting anomalies based on single-signal and multi-signal behavior patterns, integrated into the real-time EOL process.

Reduced the number of signals requiring root cause analysis by 99.8%, allowing engineers to focus on truly anomalous behavior. Cut warranty claim costs by up to 30%. The system operates in real-time during EOL testing without delaying production.

• Unsupervised anomaly detection can identify defects even with no labeled failure data
• Reducing signal dimensionality by 99.8% transforms root cause analysis from hours to minutes
• ML-augmented EOL testing catches subtle defects that SPC-based systems miss