Dana

Dana, a global driveline and sealing products manufacturer, faced persistent quality challenges at one of its high-volume axle assembly plants. Each axle unit passed through more than 20 discrete operations, generating over 200 sensor signals per unit — a data volume that overwhelmed conventional Statistical Process Control (SPC) systems, which could only analyze a fraction of collected data. Backlash and noise, vibration, and harshness (NVH) failures had multiple overlapping root causes, each tied to different signals and operations. Manual root cause analysis took days, and on some assembly lines more than 10% of parts required rework — directly eroding First Time Through (FTT) yield and adding significant scrap and labor costs.

Dana deployed Acerta's LinePulse platform, a predictive ML system purpose-built for high-dimensional manufacturing data, targeting the assembly lines with the lowest FTT rates first. LinePulse applied non-polynomial feature extraction to reduce the 200+ signal space to the subset most predictive of backlash and NVH failures, without requiring pre-labeled failure data. Once critical signals were isolated, Dana configured custom real-time alerting within LinePulse so engineers received notifications when signals trended abnormally — enabling intervention before control limits were breached. Following measurable improvement on the initial lines, Dana expanded the deployment across multiple facilities worldwide, including driveline and e-Propulsion plants, as part of its broader digital manufacturing strategy.

Deploying LinePulse eliminated the primary sources of backlash and NVH failures across targeted axle lines, delivering measurable, sustained improvements:

• 65% reduction in axle failure and rework rates
• $2.5–3M in estimated cost savings from reduced scrap and rework labor
• Rework rate held below 4% consistently, even at plants producing over 1 million parts per year

Beyond the headline numbers, root cause diagnosis time dropped from days to minutes. The success of the initial pilot drove a multi-year global rollout, with Dana's VP of Global Continuous Improvement citing LinePulse as a core component of the company's digital manufacturing strategy.

• Pilot on the worst-performing lines first — it creates the clearest ROI signal and builds internal confidence for a broader rollout.
• High signal volume is not the obstacle; the right ML approach (non-polynomial feature extraction) can isolate the 5–10 signals that actually drive failures.
• Real-time anomaly alerting changes operator behavior: teams shift from reactive rework to proactive process correction.
• Unsupervised ML can deliver quality gains even without labeled historical failure data, which is often unavailable in brownfield plants.
• Scalability depends on platform architecture — verify that the solution can replicate models and alerting configurations across geographically distributed facilities.