Our solution improved defect detection by 320%, reaching 96% precision.
Meet our client
Client:
Industry:
Market:
Technology:
Client’s Challenge
The client required a real-time system to detect defects in carbon fiber sheets during production, such as gaps, fuzzballs, and general anomalies, to enhance their existing quality control process.
Our Solution
We developed a custom pipeline using detection and segmentation neural networks, designed to identify defects. Our team labeled and curated a dataset of over 2.5k defect instances, ensuring the model had high-quality training data. The system supported both supervised and unsupervised models, achieving pixel-level accuracy and analyzing a 3.4 x 3.8 m sheet in under 10 seconds.
Client’s Benefits
Our solution improved defect detection by 320%, reaching 96% precision. It automated the identification of specific defect classes and general anomalies, significantly elevating the client’s manufacturing quality control process.
