Abstract
We propose the use of deep learning techniques, specifically convolutional neural networks (CNNs) for delivering microstructure optimization solutions to a lightweight material, AZ61 magnesium alloy that is critical for use in applications within the automotive and aerospace industries. Microstructural parameters of the material under different conditions (as-cast and heat-treated at 100°C, 200°C, 300°C, and 400°C) are fed to a robust CNN to predict the key mechanical properties. The outcomes achieved 100% correct classification of property predictions, confirmed through a confusion matrix, showcasing the model's prowess in detailing complex microstructure features like grain size, phase compositions, and border characteristics. The novelty include the application of class activation maps and convolutional filters to visualize the contribution of β-Mg17Al12 phase redistribution, grain refinement, and homogenization to mechanical performance enhancement mitigating an interpretability gap in prior reports for tuning over AZ61 alloys.