Development of a manufacturing-specialized image AI model trained on official crash test data

Elice, Inc. (CEO Jae Won Kim), an AI education solutions company, has developed an AI-based image solution tailored for manufacturing sites in collaboration with Hyundai Motor Group’s Namyang R&D Center. The solution automatically classifies, searches, and manages crash test images. The project, led by Hyundai Motor Group’s Namyang R&D Center, was carried out over approximately eight months, from late March to late November this year.

Hyundai Motor Group’s Namyang R&D Center initiated this solution development project to address the significant time and manpower required for classification and search when test photos are stored randomly, making analysis difficult. Elice developed a deep learning–based image classification and retrieval model, and then integrated multiple models according to their purpose into a single, unified service. In particular, the models were trained in a private cloud environment that applies Elice’s modular security framework, thereby reducing the risk of data leakage and enhancing overall security.

Customized AI that classifies over 60 types of crash test images with more than 98% accuracy

To tackle the challenge of crash test image classification—which was difficult to solve using off‑the‑shelf models alone—Elice applied a customized AI model retrained primarily on crash test data from official institutions. As a result, the system can now classify more than 60 types of test images with over 98% accuracy. In addition, by retraining existing AI models with public datasets, Elice was able to build a model well-suited to the Namyang R&D Center’s specific test environment, even under conditions where proprietary data is limited.

The image search function allows users to select a damaged area or specific component directly on a photo with a mouse, and then displays visually similar images in order of relevance. This enables much more precise retrieval of comparable test cases than traditional text- or image-based search methods.

Elice packaged image indexing, automatic classification, and search capabilities into a single desktop application, so that the entire process—from the moment test photos are stored to their eventual utilization—can be handled automatically in one streamlined workflow.

Jae Won Kim, CEO of Elice, said, “The team at Hyundai Motor Group’s Namyang R&D Center gave feedback that both the image similarity search feature and the program UI exceeded their expectations,” adding, “Building on this collaboration, Elice plans to continuously expand practical, ready-to-deploy AI solutions across a wide range of industrial fields, including manufacturing and mobility.”