Machine learning (ML) algorithms have taken a big leap in the past few years. Their applications are far-reaching, e.g., for autonomous driving, natural language processing, or speech recognition devices. ML-driven approaches have recently also gained high interest in material science. Such approaches are highly interesting to identify the influence of structures or morphologies ranging over different length-scales in relation to the property of interest.

Further, the recent development in deep learning provides exciting possibilities towards synthetic image generation enabling the possibility to predict material properties for unseen conditions. In particular the prediction of material properties from a given microstructure and its reverse engineering displays an essential ingredient for accelerated material design.  

The Materials Center Forschung GmbH, developed (within the COMET framework K2 Center IC-MPPE, P. No. 886385, P. No. P2.22 ECOSolder and FFG-projects: NanoPore, P. No. 883905, ProQualiCu, P. No. 853467 ) a comprehensive methodology suitable to uncover the underlying  processing-structure-property relationship in porous materials. The team utilizes a multi-method machine learning approach incorporating tomographic image data acquisition, segmentation, microstructure feature extraction, feature importance analysis and synthetic microstructure reconstruction. 

The presented findings highlight not only the importance of synthetic image generation and of accurately retrieving a set of microstructural features with statistical confidence for accelerated material design but also scrutinizing the features physical meaning in context to the material property. The presented methodology provides an essential step for the prediction of material properties, of unseen conditions, for porous materials.

A link to the manuscript can be found here: Analyzing microstructure relationships in porous copper using a multi-method machine learning-based approach | Communications Materials (nature.com)