1) can classify crystal systems (seven classes) and space groups (230 classes) with high accuracy based on powder XRD patterns and that data-driven quantification of empirical expert knowledge is possible using an interpretable ML model. In this paper, we show that a simple and fast ML technique (Fig. By analysing the model, we could quantitatively specify the rules of thumb that experienced researchers have. If an ML model can classify the crystal class and the space group based on a diffraction pattern, it must have classification rules. While most materials informatics (MI) studies aim high classification or prediction accuracy, we believe that maintaining the possibility of data-driven knowledge discovery by using a human-interpretable model is important as well. However, the complexity of CNN, or deep neural network, makes it difficult to interpret its internal processes to extract meaningful insights. They achieved high classification performance despite data deterioration due to Poisson noise and instrumental resolution. classified crystal systems and space groups by applying a convolutional neural network (CNN) to simulated powder XRD patterns. Among various subtopics such as pattern decomposition and phase identification 15, 16, 17, 18 cluster analysis and phase mapping 19, 20, 21, 22, 23, similarity metrics for comparison of diffraction data 24, 25, 26 classification of a crystal symmetry 27, 28, 29, 30, 31, 32, 33, a paper by Park et al. Therefore, we focus on the classification of crystal systems and space groups using machine learning (ML) approaches, inspired by the fact that experienced researchers can guess the crystal system from a given diffraction pattern.Īpplication of ML and related techniques for diffraction data analysis is a hot research topic in recent times 13, 14. Excluding human involvement in these processes as far as possible improves the situation and helps realise high-throughput (HiTp) experiments. Given that a large number of powder XRD patterns are generated daily at synchrotron facilities around the world, these time-consuming processes performed manually by human experts are obvious bottlenecks in materials research 11, 12, 13. While the most arduous step is structure refinement using the Rietveld method 10, which typically requires manual optimisation of tens of parameters, space group determination at the initial stage of structure analysis also needs manual trial-and-error operations frequently. Decoding powder diffraction patterns to crystal structure information involves several steps, such as peak indexing, space group determination, initial parameter estimation for the crystal structure, and structure refinement 4, 5, 6, 7, 8, 9. Powder X-ray diffraction (XRD) and powder neutron diffraction are principal experimental techniques to elucidate crystal structures data obtained using these techniques are stored in various databases for specific classes of materials, for instance, inorganic materials and proteins 3. A crystal structure is defined in terms of lattice symmetry, lattice parameters, the types and positions of atoms, and site occupancy. Single molecular layer MoS 2 and WS 2 suspensions, prepared by exfoliation, provide excellent randomly oriented two‐dimensional systems for demonstrating the unique features of powder x‐ray diffraction patterns of two‐dimensional materials and for structure identification using Bragg peak profiles.Crystal structure characterisation is one of the most important tasks in materials development because crystal structure determines material properties 1, 2. It is demonstrated that because of structure factor modulation the Warren expression which relates the width of Bragg peaks to layer size cannot be used for a two‐dimensional sheet with more than one layer of atoms, and it is proposed in such cases that measuring the low‐angle side width of half‐maximum intensity can be used for determination of the layer size. This structure factor modulation provides a continuous plot of the structure factor over the range of the diffraction tail and thus provides valuable information about the structure of the layer. For a two‐dimensional structure consisting of more than one monolayer of atoms, the shape of the Bragg peaks is modulated by the structure factor. The analytic solution, where the Bragg peaks are strongly asymmetric, is compared to computer simulations using the Debye formula, and is shown to be in closer agreement than earlier numerical solutions by Warren and others. An analytic solution for the normalized intensity for powder x‐ray diffraction has been obtained for a simple two‐dimensional lattice using a linear approximation for the interference function.
0 Comments
Leave a Reply. |