A research team from The Hong Kong University of Science and Technology (HKUST) has unveiled GrainBot, an innovative AI toolkit designed to automate the quantitative analysis of microstructures in microscopy images. This development addresses the increasing demand for data-driven methodologies in materials science, facilitating a systematic approach to transforming intricate image data into quantifiable figures. As the field of materials science progresses, tools like GrainBot are crucial for enhancing the discovery and refinement of next-generation materials.
Microstructure quantification has historically posed significant challenges across various disciplines within materials science and engineering. Although modern microscopy can produce detailed images, extracting consistent and scalable insights from these visuals has proven difficult. Traditional methods often focus on basic feature identification, providing limited understanding of how diverse microstructural parameters interact. This limitation can obstruct researchers from fully grasping structure-property relationships, thereby slowing the development and optimization of new materials.
To tackle this issue, the team, led by Prof. ZHOU Yuanyuan, Associate Professor in the Department of Chemical and Biological Engineering at HKUST, created GrainBot. This toolkit offers an integrated solution for image segmentation, feature measurement, and correlation analysis. Utilizing a convolutional neural network for accurate grain segmentation, GrainBot incorporates custom algorithms capable of measuring grain surface area, grain-boundary groove geometry, and volumes of surface concavity or convexity.
By converting microscopy images into extensive numerical descriptors, GrainBot empowers researchers to establish large-scale, standardized databases of microstructures, moving beyond reliance on qualitative observations. The team demonstrated GrainBot’s effectiveness by applying it to metal halide perovskite thin films, essential materials for high-efficiency solar cells. The toolkit processed atomic force microscopy (AFM) images of samples with varying bottom surface morphologies, yielding a database of thousands of individual grains annotated with multiple microstructural parameters.
The subsequent statistical analysis uncovered general distribution patterns and intricate relationships among features such as grain size, groove geometry, and surface roughness. Additionally, the study employed interpretable machine-learning models to explore interactions between different microstructural features. By training gradient-boosted decision tree models on selected grain descriptors and utilizing interpretation tools like feature importance profiles and partial dependence plots, the researchers examined how parameters such as grain surface area and grain-boundary groove angle collectively influence characteristics like surface concavity depth and ridge height.
Prof. GUO Yike, Provost and Chair Professor in both the Department of Computer Science and Engineering and the Department of Electronic and Computer Engineering at HKUST, emphasized the broader implications of this research for AI-driven scientific infrastructures. “GrainBot illustrates how AI can transform complex microscopy images into structured, reproducible datasets that can be readily shared, re-analyzed, and integrated into larger research platforms,” he stated. Such advancements are vital as scientific workflows increasingly become automated and data-intensive.
Prof. Zhou added that GrainBot aims to assist researchers needing consistent, quantitative microstructural descriptors. “Our goal is to lower the barrier for integrating microscopy characterization into data-driven studies and autonomous laboratory platforms. By providing a unified framework adaptable to various perovskite compositions and processing conditions, GrainBot enhances accessibility to microstructure quantification, even for those lacking specialized coding or machine-learning skills.” Understanding grain morphology, including grain-boundary grooves and surface structures, is particularly significant for improving the long-term stability of perovskite solar cells.
Beyond applications in perovskites, GrainBot establishes a strategic framework for analyzing microstructures in other polycrystalline thin films. Looking ahead, the research team plans to integrate GrainBot with various characterization techniques and investigate direct correlations between microstructure characteristics and device performance, as well as long-term stability.
This research, titled “GrainBot: Quantifying Multi-Variable Microstructure Disorder in Materials,” has been published in Matter, a leading journal from Cell Press.
