A data science project focused on the interpretation of X-ray images of crystalline materials. The goal is to leverage data science techniques to efficiently analyse large sets of raw X-ray images and thus extract valuable insights into molecular structure and reasons for subsequent physical properties. By developing a method based on convolutional neural networks (CNNs), rapid interpretation of X-ray images can be achieved.
The project targets researchers, scientists, engineers, manufacturers, and quality control personnel involved in the development and production of high-tech materials and devices. It aims to enhance the design space for profitable materials by enabling high-throughput analysis and non-experts to perform high-quality interpretations.
One of the main challenges is the shortage of labelled X-ray image data available for training. The project explores machine-learning engineering considerations, including the potential use of artificial general intelligence systems for interpreting X-ray images.
EDA is conducted on experimental and simulated images to ensure data quality. Statistical errors associated with simulations are also evaluated using quantitative metrics.
Two different CNN architectures are investigated, with the smaller model architecture deemed more suitable for deployment due to its better performance in interpreting real X-ray data.
The project's success in interpreting real observed data and addressing the lack of labelled data enables the implementation of an automated X-ray interpretation CNN-based device. This pipeline will provide detailed knowledge of the structure and details of desirable materials, leading to increased demand and net profits for related products.
Overall, the project combines data science, machine learning, and X-ray imaging to unlock the potential of atomic or molecular-scale structure analysis in crystalline materials, benefiting various industries and driving innovation in materials development and quality control.