AI is transforming the manufacturing ecosystem, but its full potential is often limited by one fundamental challenge: data availability. In advanced production processes, collecting large datasets representative of all possible defects, failures and process conditions can be expensive, time-consuming or even impossible. We develop new generative AI approaches to overcome these barriers by creating realistic synthetic data that complements real process information. By combining experimental observations with AI-generated data, we improve the ability to understand complex process signatures, identify rare defects and design corrective strategies before failures occur.
Beyond data generation, we explore the next frontier of multimodal generative AI, where text, signals, images and/or videos are integrated into unified models capable of learning from different sources of manufacturing knowledge. These approaches enable zero-shot and few-shot learning strategies, allowing AI systems to adapt to new materials, processes and defects with minimal additional data. We also investigate agentic AI frameworks where intelligent agents support engineers in strategic decision-making, process optimization and the transition towards autonomous, self-improving manufacturing systems.
