CYRUS-GenAI- Leveraging HPC and Diffusion-Based Generative AI for Scalable Industrial Surface Inspection

Short description of the innovation study

Deploying CYRUS to a new production line currently requires a two-month data collection and annotation phase, while entering new material markets (e.g. copper, steel) can take up to six months. The CYRUS-GenAI project aims to break this bottleneck by leveraging European HPC infrastructure to train a domain-specific Industrial Vision Foundation Model and a Physics-Guided Diffusion Model, built on D-Cube's existing repository of million high-resolution industrial images. This approach enables the synthesis of infinite, physically accurate samples of rare defects, overcoming the limitations of general-purpose AI models, which lack the physical precision required for industrial applications. The expected outcome is a reduction in new-installation deployment time from 2 months to under 5 days, an increase in annual installation capacity, and the ability to rapidly adapt (few-shot learning) to new materials and industrial sectors.

Organisations involved: 

Main SME: D-Cube