In material science experiments, large amounts of data are produced, but only partially captured in a central knowledge base. One of the reasons for this is that many of the existing processes are not yet digitized. Additionally, the already captured data and processes are often stored in various formats. This hinders the retrieval and systematic analysis of the data.
The goal of the DiProMag research project is to digitize the process chain from production to characterization to prototypical application of magnetocaloric materials. As part of this project, an ontology is being developed to represent experiments, experiment data, as well as the goals and reasoning behind the scientists' experiments. Using a novel approach to scalable ontology development that we have developed, all process data and intended outcomes are digitally represented. To achieve this, we utilize OTTR (Reasonable Ontology Templates), a language with supporting tools for representing and instantiating RDF graphs and OWL ontology modeling patterns.
Building upon this foundation, we leverage the collected data to develop an AI system that automatically generates new hypotheses from the available data, such as physical relationships. We are developing a methodology that utilizes both structured and unstructured data to train a high-dimensional embedding space, enabling the extraction of digital knowledge about material-physical connections through analogical reasoning.
The long-term goal is to establish a digital foundation for magnetocaloric materials and utilize it for the discovery and development of improved material properties, finding faster, more effective, and cost-efficient industrial solutions in shorter timeframes.