Predicting the Strength of Composites with Computer Vision Using Small Experimental Datasets

Abstract

Composite materials offer versatile properties, but predicting their mechanical behavior remains challenging due to complex morphology-performance relationships. We address this challenge using convolutional neural networks (CNNs) to analyze X-ray computed tomography (CT) images of cold-sintered polymer-ceramic composites. Traditional machine learning models with morphological features as inputs yielded limited accuracy, while transfer learning from pretrained CNNs improved predictions. Bayesian hyperparameter optimization and ensemble learning further refined the model, achieving R² values of up to 0.94 on unseen data. Leveraging the z-stack nature of CT imaging, a meta-learning approach enhanced predictions, improving R² to 0.95. This study demonstrates alternative machine learning approaches using small datasets to uncover morphology–structure–property relationships in composites and highlights the potential of computer vision in materials development.