Research Projects

Quantitative microstructural analysis, a fundamental component of materials research, is often aided by tools such as ImageJ. These tools, though widely adopted, rely heavily on human effort to produce accurate outcomes. For example, estimating the average size or size distribution of precipitates in a two-phase microstructure demands first isolating and identifying every individual precipitate across the image. Despite being arduous, such human-dependent workflows introduce limitations in scalability, reproducibility, and objectivity. The advancement of computer vision and deep learning presents a compelling alternative. When appropriately designed and rigorously trained, object detection algorithms enable automated identification and measurement of microstructural features. The proposed project, MicroSdetect, aims to harness this potential by developing an open-source AI-powered online platform for automated detection and quantitative analysis of microstructural features. The platform leverages object detection algorithms where bounding boxes not only localize features but also encode dimensional information. This enables the tool to perform both detection and precise measurement that directly support the establishment of process-structure-property relationships and inform data-driven process control. Rationale MicroSdetect aims to offer an alternate approach to microstructure analysis by providing accurate, scalable, and quantitative assessment of critical and diverse microstructural features like precipitate, grains, pores and junctions. The tool is designed to enable robust process-structure-property correlations and to facilitate data-driven process control in materials research, manufacturing, and failure analysis. Scientific Objectives – Develop an AI-based tool for quantitative assessment, including size and size distribution, of critical microstructural features like Precipitates, Grains, Porosity, Junctions and orientation in fibrous microstructures, Secondary phases, and Phase spacing – Deploy this package called MicroSdetect as an open-source AI-powered online tool with an intuitive graphical interface, enabling users to perform automated microstructural analysis without coding or artificial intelligence expertise. Hypothesis Critical microstructural features, without heavy computational demands, can be accessed by extending suitable object detection algorithms. When rigorously trained, these models not only detect features such as grains, precipitates, and pores but also enable quantitative estimation of their size, distribution, spacing and orientation. This forms a reliable framework for automated quantitative analysis across material systems, supporting robust process-structure-property correlations and data-driven process optimization through computationally efficient online deployment. Main Experiments – Build a diverse microstructure databases . – Evaluate recent object detection algorithms with a focus on balancing accuracy and computational efficiency. – Rigorously training of the selected algorithms for accurate detection. – Extend the detection outputs by extrapolating bounding box dimensions for quantitative estimation. – Deploy the pakackage as MicroSdetect, an efficient open-source online tool with an intuitive graphical interface for widespread, user-friendly access. Significance This proposal builds MicroSdetect, an open-access online platform that employs recent advancements in deep learning for quantitative assessment of critical microstructural features. Its efficient computational design, coupled with online deployment, ensures wide accessibility without requiring high-end resources. By enabling robust process-structure-property correlations across diverse material systems, it offers a transformative tool for materials research, manufacturing, and process optimization.