Blogging for MRS 2022 - Kathy Liu
What are you excited to see at the 2022 MRS Fall Meeting?

Symposium DS02: Integrating Machine Learning with Simulations for Accelerated Materials Modeling

N M Anoop Krishnan, Indian Institute of Technology Delhi

Cementron: Cement Clinker Microstructure Segmentation Using Machine Learning

Written by Aashutosh Mistry

Did you know cement (a common building material) production is one of the leading causes of global CO2 pollution? Various researchers across the globe have been trying to rethink the cement industry for a greener future. Production of cement clinker is a key step that can be reformulated. Clinkers are formed via the sintering of limestone and clay, and their microstructure is the key aspect defining their usefulness as a good cement. To reformulate this step, researchers want to correlate various processing attributes to changes in the clinker microstructure. While visualizing the clinker microstructure is not a difficult task and typically optical microscopy gives desired resolution, analysis of the microstructure is effort intensive and typically one manually identifies various materials phases in such images based on an empirical understanding of underlying materials phases. N M Anoop Krishnan and his colleagues are using machine learning to automate this step to accelerate the overall research in rethinking cement production. While multiple machine learning algorithms for image analysis have been proposed by the tech companies like Facebook and used by various researchers for their scientific work, a key challenge in this work was the lack of a reliable image dataset of clinker microstructures. Accordingly, the researchers first had to develop a database of hundreds of microstructural images with corresponding phases identified manually. Once such a database was ready, the researchers tweaked and trained Detectron 2—a convolutional neural network-based algorithm proposed by Facebook to detect objects in images—on such an image database. The results show that the various phases in the clinker microstructure can be predicted with reasonable accuracy. While the researchers will continue to improve this algorithm to decrease errors like missed particles and misclassifications, Krishnan and his colleagues are excited to use these interpreted images for further analysis—especially to correlate different cement manufacturing processes to clinker microstructures.

Comments

The comments to this entry are closed.