Accelerating Materials Discovery: Automated Identification of Prospects from X‐Ray Diffraction Data in Fast Screening Experiments

Abstract

New materials are frequently synthesized and optimized with the explicit intention to improve their properties to meet the ever‐increasing societal requirements for high‐performance and energy‐efficient electronics, new battery concepts, better recyclability, and low‐energy manufacturing processes. This often involves exploring vast combinations of stoichiometries and compositions, a process made more efficient by high‐throughput robotic platforms. Nonetheless, subsequent analytical methods are essential to screen the numerous samples and identify promising material candidates. X‐ray diffraction is a commonly used analysis method available in most laboratories which gives insight into the crystalline structure and reveals the presence of phases in a powder sample. Herein, a method for automating the analysis of XRD patterns, which uses a neural network model to classify samples into nondiffracting, single‐phase, and multi‐phase structures, is presented. To train neural networks for identifying materials with compositions not matching known crystallographic structures, a synthetic data generation approach is developed. The application of the neural networks on high‐entropy oxides experimental data is demonstrated, where materials frequently deviate from anticipated structures. Our approach, not limited to these materials, seamlessly integrates into high‐throughput data analysis pipelines, either filtering acquired patterns or serving as a standalone method for automated material exploration workflows.

Keywords

• deep learning
• fast screening
• high throughputs
• X‐ray diffraction