D3M: Data-Driven Design of Materials

Welcome to the website of the “Data-Driven Materials Design” research group at the Chair of Experimental Methods in Materials Science, Saarland University!

In short: what we are doing

• Machine learning for predicting micromechanical and macroscopic material properties based on microstructural data
• Inverse materials design to identify the microstructure required for improved properties
• Machine learning powered microscopy for microstructure quantification
• Processing, linking and annotation of multimodal, heterogeneous data sources (images, tabular data, simulation, literature) using large language models

Our Research

Microstructure quantification using machine learning

The first pillar of our work is microstructure quantification based on machine learning (ML): we establish ML-based microstructure analysis as a tool for large-scale data collection, with a particular focus on scenarios with limited data. Our methodological contributions include:

• the generation of synthetic data,
• the use of self-supervised learning and generative AI,
• and the automation of image annotations based on multimodal, correlative microscopy.

Microstructure-property modelling

The microstructure quantification work feeds directly into the second pillar: microstructure-property modelling in the context of processing-structure-property links (PSP). Focusing on micro-scale and microstructure-dominated properties:

• we generate our own experimental datasets, from nano- and micro-scale to macroscopic mechanical testing
• we use and compare explainable microstructure descriptors and deep, latent features for our ML modeling,
• and explore cross-property transfer learning for small experimental datasets.

CircularSaar consortium

The third pillar is our role in the CircularSaar consortium: there, we serve as a central point of contact for materials data science, for collaborations both within the university and with external industry partners. Our tasks include:

• exploratory data analysis,
• development of strategies for data collection and annotation,
• and training of machine learning models and the interpretation of the results in the context of process-structure-property correlations.

We envision the development of a structured, quality-assured material database as the foundation for data-driven research across the entire consortium, promote a common language and better understanding between materials science and computer science, and, most importantly, want to establish data-driven materials design as a methodological enabler that can be specifically applied to implement the principles of the circular economy and make sustainable materials decisions.