# AI Vision for Foam Mold Clip Detection Achieve 100% accuracy detecting clips in automotive & aerospace molds despite resin, vibration, and variation. ## TL;DR (Quick Answer) A Tier-1 supplier's legacy vision system failed to detect black clips in foam molds due to resin buildup and vibration. Overview.ai’s [OV20i Vision System](/content/products/ov20i/index.html), using a classifier recipe, achieved 100% accuracy with just a dozen training images, providing a stable solution without needing recalibration. ## Situation Detecting black clips embedded in foam molds may sound like a simple presence/absence task. But in production environments, variation breaks systems long before accuracy does. This is a common challenge in both automotive and aerospace assembly where component verification is critical. ## The Problem A Tier-1 automotive and aerospace supplier faced this issue with eight different mold variations. Their $100K legacy system constantly flagged false rejects and couldn’t be trusted for unattended operation due to several factors: - **Resin buildup** altered the mold’s appearance over time. - **Reflections and lighting** changed with every cycle. - **Vibration from nearby presses** created instability and inconsistent imaging. - **Mold-to-mold variation** made rule-based systems brittle. ## The Overview.ai Solution Using the [OV20i industrial vision system](/content/products/ov20i/index.html), engineers trained a simple classifier recipe. Within hours, the model reached 100% accuracy on presence/absence checks and learned to classify *mis-seated clips* separately. ### Fast Training with Classifier Recipes With just a dozen labeled images, the on-camera AI learned to distinguish between "clip present," "clip absent," and "clip mis-seated," demonstrating the power of small, high-quality datasets. ### Robust Optics & In-Camera Processing Key to success was robust optics and a segmentation-grade sensor that normalized for lighting and surface texture changes. The system stayed stable through resin buildup and vibration—no recalibration required. ### Real-time Feedback with Node-RED The inspection results were integrated with the line controller using Node-RED logic, providing immediate pass/fail feedback and closing the loop for operators. ## Key Engineering Takeaways - Small, high-quality datasets outperform large, noisy ones when lighting is consistent. - Stability against vibration and surface contamination is as critical as accuracy. - Real-time feedback through Node-RED logic closes the automation loop for operators. ## FAQ ### How many samples are typically needed to train for new molds? 10–15 representative samples are usually sufficient if lighting and optical parameters remain stable. ### Can this handle resin discoloration or partial fill? Yes — segmentation features in the OV vision systems can adapt to grayscale shifts and texture changes with minimal retraining. ### Ready to Automate Your Clip & Component Detection? Get started with the OV20i AI Vision System for reliable, high-speed assembly verification. [Learn About OV20i](/content/products/ov20i/index.html) [Get Technical Consultation](/content/contact/index.html)