The digital transformation of manufacturing is generating unprecedented amounts of data from sensors, machines and production systems. However, collecting more data does not automatically mean creating more knowledge: industrial data streams are often massive, heterogeneous and affected by variability, disturbances and hidden sources of uncertainty. We develop new in-situ data mining and AI methods designed to extract meaningful information from complex manufacturing environments.
Our research focuses on computationally efficient algorithms capable of processing high-dimensional and multimodal data sources — including images, videos, signals and machine data — while remaining robust against noise, process variability and changing operating conditions. Our models allow embedding physical process knowledge and identifying hidden relationships between process signatures, defects and final product quality. Our goal is to transform raw manufacturing data into actionable intelligence, supporting the transition from traditional quality inspection towards autonomous and data-driven manufacturing systems.
