Visualization of small-angle X-ray scattering datasets and processing-structure mapping of isotactic polypropylene films by machine learning

Abstract

With the rapid development of the synchrotron radiation X-ray characterization techniques, the preprocessing of large small-angle X-ray scattering (SAXS) datasets and the data mining become urgent requirements for researchers. In this work, we apply the variational autoencoder (VAE) and the conditional variational autoencoder (cVAE) to visualize a large SAXS dataset of hard-elastic isotactic polypropylene (iPP) films in 2- and 1-dimensional latent spaces. The low-dimensional representations enable us to capture key features of the dataset rapidly, such as the similarity among SAXS patterns and the structural evolution trends. The preprocessing of the dataset points out the further direction of data analysis so that researchers can focus on the most valued regions in the dataset. Then, we develop a hybrid VAE-multilayer perceptron (MLP) neural network to realize the processing-structure mapping of iPP films. The robustness of the hybrid VAE-MLP network is verified. Finally, SAXS patterns in the temperature-strain space are generated, which allows us to explore the processing parameter space not involved by previous experiments. These capabilities indicate that the developed machine-learning methods are valuable artificial intelligence toolset to assist in the preprocessing of large-scale SAXS datasets and the establishment of comprehensive processing-structure relationship of hard-elastic iPP films.