KNODI

About the project

Manufacturers collect huge amounts of data through Industrial Internet of Things (IIoT) technologies, but turning it into value is still hard. Production involves many interrelated steps, with variation in products, processes and context. Data from production planning, sensors and operator records has to be combined. Today, answering a single business question can keep a team of experienced data scientists and domain experts busy for 3 to 6 months. An example of such a question is “What factors determine the energy consumption for this specific product?"

KNODI studies how state-of-the-art language-processing AI, combined with structured knowledge representations, can extract, store and use company knowledge. The goal is to support complex production data analyses in two areas: energy consumption and root-cause analysis of product quality. In-house data users can then explore a rich data context and answer complex questions much more efficiently. Answering such questions with less effort is key to reducing energy use, improving efficiency and eliminating scrap.

Key facts
Industrial partners	Bekaert, Materialise, Timeseer
Research partners	Flanders Make, KU Leuven

My contributions

• Document analytics that turn company documents into curated, analysis-ready knowledge.
• Time-series analytics for anomaly detection and root-cause analysis in production data.
• Integration of both into a single analytics pipeline.

Technologies: large language models, knowledge representation, time-series anomaly detection, root-cause analysis.