The determination of morphological properties in metal foams is a major undertaking often relying upon time-consuming and extensive techniques like computer tomography for high quality data acquisition, even employing machine learning techniques. This amount of commitment is one of the reasons why metal foams are not widely used in industry, since quality control of the batches produced is limited to computer tomography or destructive methods. With recent advances in artificial intelligence and neural networks, approaches can be taken regarding information disentanglement from previously unattainable data.

In this work a new method of analysis of morphological 3D properties based on 2D X-ray radiograms and the employment of a new Convolutional Neural Network (CNN) architecture is proposed. The training of this model is based on a combined approach of simulating simplified foams as pre-training data and the acquisition of real experimental data, extracted from X-ray computer tomograms. The network was trained on simulated Voronoi foams and 41 XCTs of real metallic foams. The foam data was subjected to an augmentation, which resulted in a data set of 5000 simulated and 6600 experimental X-ray radiographic foam images. The training was aimed at the regression of the cell size distribution and the sphericity distribution from these images. In addition, tests were carried out to get an insight into the robustness of the model when confronted with similar data that was not included in the training process. It was found that the effectiveness of the neural network increases with a larger number of cells in the observed volume. The presented method allows for a fast in-line application in quality control for industrial settings without interrupting the production process. Additionally, the application is not necessarily limited to metallic foams, but rather can be adapted to other cellular media as required.