AI-Driven Automated Metallography Analysis Platform (AMP): Accelerating Microstructural Characterisation Through Computer Vision and Explainable AI

Microstructural characterisation remains a critical yet time-intensive activity across materials research, quality assurance, and advanced manufacturing. Conventional metallographic workflows often rely on manual image interpretation, repetitive measurements, and subjective assessments, creating significant bottlenecks in laboratory productivity and data reproducibility. As the complexity of advanced alloys continues to increase, there is a growing need for automated and intelligent analysis tools capable of delivering rapid, consistent, and traceable results.

This presentation introduces the AI-Driven Automated Metallography Analysis Platform (AMP), a web-based software platform developed to automate key metallographic image analysis workflows using computer vision, machine learning, and explainable artificial intelligence (AI). The platform integrates automated grain boundary segmentation, ASTM-compliant grain size estimation, phase quantification, crack and porosity detection, inclusion analysis, and automated report generation within a unified user interface.

A key feature of the platform is the incorporation of deep learning models for microstructural phase classification and defect identification. Unlike traditional threshold-based approaches, the system is designed to support complex multi-phase microstructures and varying imaging conditions while reducing the need for extensive manual intervention. To improve transparency and user confidence, explainable AI techniques based on Gradient-weighted Class Activation Mapping (Grad-CAM) are integrated, enabling users to visualise the image regions that influence AI-driven predictions.

The presentation will demonstrate how automated image-processing workflows can significantly reduce analysis time while improving consistency and repeatability across laboratories. Example case studies will illustrate automated grain sizing, phase fraction estimation, defect detection, and mechanical-property prediction from microstructural descriptors.

A live demonstration of the hosted prototype will be provided using a Bring Your Own Data (BYOD) approach. Attendees will have the opportunity to upload their own metallographic images and observe real-time automated analysis, including segmentation, quantification, and report generation.