Autonomous Image Analysis to Accelerate the Discovery and Integration of Energy Materials

Abstract

Accelerating the development of electrochemical energy devices (e.g., batteries, fuel cells, electrolysis cells) is pivotal for the transition towards a green and sustainable energy economy. A significant proportion of development efforts in this realm is relying on new functional materials as electrocatalytically active media, ionic media, porous transport media, or multi-functional electrodes. Experimental materials research harnesses a broad spectrum of imaging and visualization methods, based on modern spectroscopy, microscopy, and scattering techniques. These techniques are utilized to quantify structural characteristics of materials and to detect and monitor structural changes during fabrication and normal operating cycles of the device. The extreme diversity and the complexity of materials leave the applications of conventional methods for image processing often empirical and indiscriminate. Recently, deep learning (DL) models based on deep convolutional neural networks (ConvNets) have been transforming the field of computer vision by outperforming the classical algorithms in various problems such as image recognition, reconstruction, synthesis, style transfer, and other tasks. To leverage the advancements in DL-based approaches in materials research, ConvNet models are employed for various analysis tasks like segmentation, object detection, denoising, or super-resolution microscopy. Here, we will present DL-based methodical pipelines to automate the particle size distribution analysis of Pt nanoparticles in catalyst layers [1] and for high-throughput screening of catalyst layer inks [2] in polymer electrolyte fuel cells.