Materials Data Science and Informatics (IAS-9)

In situ TEM has long been the leading method for imaging dislocations in crystalline materials—but until now, the lack of quantitative image analysis has limited what researchers could learn. Our new solution changes the landscape.

Leveraging state-of-the-art deep learning and AI-driven data analysis, we’ve built a powerful digital twin of an in situ TEM straining experiment. This technology not only automates precise image interpretation but also enables fully matched simulations for unprecedented analytical depth. With it, we extract high-resolution spatio-temporal data on dislocation motion in a Cantor high-entropy alloy and uncover the universal behavior of plastic strain avalanches.

Our approach allows clear observation of single-dislocation “stick–slip” events and delivers robust, scale-free avalanche statistics—independent of driving stress—through advanced data analytics.

This machine-learning–powered framework is fully generalizable, setting a new standard for quantitative, reproducible, and data-rich TEM analysis and accelerating material discovery like never before.

Related Publications:

• H. Song, B. D. Nguyen, K. Govind, D. Berta, P. D. Ispánovity, M. Legros, S. Sandfeld, Enabling quantitative analysis of in situ TEM experiments: A high-throughput, deep learning-based approach tailored to the dynamics of dislocations, Acta materialia 282, 120455 - (2025) [10.1016/j.actamat.2024.120455]