LG Innotek Achieves 99.99% Defect Detection with Intel-based AI Vision Inspection

Title

LG Innotek Achieves 99.99% Defect Detection with Intel-based AI Vision Inspection

Content

LG Innotek, a Korean producer of precision components for mobile devices, automotive displays, semiconductors, and smart products, deployed an Intel-based AI-powered automated inspection solution that achieves a 99.99% defect detection rate across its precision component manufacturing lines. The system is built on Intel Core Processors (with integrated GPU), Intel Arc discrete GPUs, and Intel Xeon Processors, combined with OpenVINO as the inference engine. Lee Sang-houn, head of Equipment Technologies Division, Production Innovation Center at LG Innotek, stated that traditional rule-based inspection systems "struggled to adapt to new materials and new product specifications and often relied on the subjective judgment of operators, leading to ambiguous standards." Yang Hee-cheol, head of LG Innotek's AI Inspection Technology Team, added that Intel Arc discrete GPUs "improve cost efficiency by 3-4 times compared to equivalent hardware performance from other vendors LG Innotek had used." LG Innotek manages hundreds of precision-component models, each with different inspection items, standards, and specifications. Under the previous rule-based system, defect judgments varied by operator condition and individual perspective, and even small specification changes created quantification challenges. The deployment replaces those subjective, fragmented inspections with a fully automated AI vision stack that runs edge inference on Intel Core-based embedded Vision Inspection PCs at the manufacturing equipment, with periodic retraining of deep learning models on Intel Xeon servers. The deployment uses an EfficientNet-B3 inference model (about 10M parameters, 240×240 RGB input) running at edge, with retraining on ResNet-50 (100-image dataset, batch size 16, bf16). OpenVINO simplifies model conversion — "by adding just a few lines of code, third-party trained models could be used for inference in Intel environments through OpenVINO, without any burden on the existing development environment," Yang explained. Intel's AMX (Advanced Matrix Extension) is leveraged on Xeon for fine-tuning parallel computation. The system achieves 99.99% defect detection rate in production. Operational outcomes include: Intel Arc discrete GPUs delivering 3-4× the cost efficiency of equivalent hardware from other vendors; full automation of previously subjective inspections; and elimination of the need for visual inspection by human workers as a backup to AI. The system is being scaled across more than 100 LG Innotek processes. Lee Sang-houn's forward-looking statement frames the deployment trajectory: "In the future, starting with the production line of the Flip Chip Ball Grid Array (FC-BGA), a high-value-added semiconductor substrate, in Gumi Factory, we plan to deploy Intel AI solutions to various manufacturing facilities to ensure stable mass production." LG Innotek is also expanding AI beyond inspection into maintenance and other areas, with the goal of becoming the global No. 1 advanced materials and components company.