Deep Learning-Based classification of aluminium alloy microstructures using synthetic metallographic data

Abstract The present study investigates the integration of synthetic metallographic data into machine learning workflows for the automated analysis of aluminium alloy microstructures in manufacturing contexts. A bespoke algorithm was developed to generate synthetic micrographs that replicate key metallurgical features associated with various manufacturing processes, including sand casting, die casting, forging, extrusion, and different heat treatment conditions. Convolutional Neural Networks (CNNs) trained exclusively on synthetic datasets were evaluated using real microscopy images subjected to conditions simulating real-world acquisition challenges, demonstrating strong generalization performance. The models achieved a classification accuracy of 99.3% across 21 distinct microstructural classes under optimal imaging conditions. The findings indicate that synthetic data can serve as a substitute for real images in training applications, thereby facilitating the development of robust models that are capable of generalising effectively across process-induced microstructural variations. The proposed approach offers a scalable, cost-efficient solution for microstructural classification and enables the integration of data-driven analysis into advanced manufacturing pipelines, which contributes to enhanced process monitoring and quality control.