Faurecia

Faurecia's Yancheng facility — Asia's largest automotive slide production site — faced persistent quality challenges that traditional inspection methods could not resolve. Automotive interior components, particularly seating slide mechanisms, must meet exacting tolerances across both visual and acoustic dimensions: a rattle or abnormal noise in an assembled unit is as disqualifying as a visible surface defect. Conventional vision-based quality systems had no mechanism to catch acoustic failures, leaving noise-related defects to surface only after delivery. The resulting customer complaints, scrap rates, and quality costs were compounding under intense OEM pricing pressure, making defect reduction both a margin and a customer-retention imperative.

Faurecia Yancheng deployed a multimodal AI quality control system spanning more than 40 production use cases, combining machine learning, deep learning, and generative AI to address the full sensory spectrum of defect types encountered in slide assembly. The multimodal architecture was the defining design choice: by fusing acoustic sensor data with visual inputs, deep learning models could classify noise-related defects — rattles, vibrations, and abnormal mechanical sounds — that purely optical inspection systems cannot detect. Rather than a contained pilot, the rollout was systematic across production lines, integrating AI inspection checkpoints into existing manufacturing workflows at scale. Models were trained on multi-sensory defect signatures representative of real production conditions, enabling automated screening at throughput rates that manual auditory testing cannot match.

The system delivered measurable improvement across every targeted dimension. Customer complaints fell by 94%, effectively eliminating the post-delivery quality escapes that had driven warranty costs and OEM relationship risk. Total quality-associated costs declined by 62.5%. Scrap costs dropped by 75.8%, directly improving material yield and strengthening contribution margin. Overall equipment effectiveness rose by 10.2%, reflecting simultaneous gains in quality rate, equipment availability, and throughput performance. The breadth of outcomes — spanning defect prevention, cost reduction, and utilization — resulted in the site's recognition by the World Economic Forum's Global Lighthouse Network as a benchmark for AI-driven manufacturing.

• 94% reduction in customer complaints
• 75.8% reduction in scrap costs
• 62.5% reduction in quality-associated costs
• 10.2% increase in OEE

• Multimodal AI — fusing acoustic, visual, and other sensor streams — is required for complete defect coverage in mechanical assemblies; vision-only systems will miss noise-class defects entirely.
• Systematic rollout across 40+ use cases drove site-wide KPI movement, including OEE; isolated pilots rarely produce outcomes at this scale.
• Scrap cost reduction (75.8%) typically surfaces faster in financial reporting than complaint reduction — use it as the primary ROI anchor when building the internal business case.
• Acoustic defect detection requires dedicated training data infrastructure; budget for sensor installation and labeled data collection before model development begins.
• WEF Lighthouse recognition demonstrates that rigorous documentation of AI outcomes can itself become a competitive signal with demanding OEM customers.