Global Automotive Parts Manufacturer · Automotive
The manufacturer was sampling sensor data at 10% across their production lines — a cost-control measure implemented when the volume of IoT sensor data made full processing financially impractical. A progressive fatigue pattern in a safety-critical component was developing across a subset of units produced during a specific production window, generating subtle vibration signatures that were present in the sensor data but invisible at 10% sampling.
Full sensor telemetry coverage deployed across all production lines. LLM reasoning applied to identify progressive failure signatures — patterns that develop over time and only become statistically significant at full data volume.
"The signature was there in the data for 18 days before we would have known anything was wrong. At 10% sampling it was noise. At 100% it was a clear, actionable pattern."
VP of Quality Engineering
The failure signature manifested as a 0.3% deviation in resonance frequency during a specific stage of the production process — statistically invisible in sampled data but consistently present across affected units at full coverage. The targeted recall of 47,000 units was completed without a single field incident. The full production run recall would have cost an estimated $23M and significantly damaged the brand relationship with their primary OEM customer.