Supervised Machine Learning - Based Method of Data Analysis of X-Ray Absorption Fine Structure Spectra
Implementing supervised machine-learning to various materials' spectroscopy data to gain more insight on the structures of those materials Background: A multitude of methods can be used to analyze the structures of various materials and gather information about them. One of those methods includes first reviewing the numerous spectra of the materials and then making conclusions on the material itself. Two spectra that are commonly looked at are X-ray absorption spectra (XAS) and Electron energy loss spectra (EELS). X-ray absorption spectroscopy is used to discover the local atomic and electronic structure of a material. The data in this spectra is obtained by measuring transmission and/or yield of fluorescent X-rays or secondary electrons of an element in a material. This measurement variable is then made as a function of incident x-ray energy over an energy range with a sufficiently narrow energy band. This narrow energy band corresponds to an absorption edge of the respective element at which the incident x-ray photon has sufficient energy to cause a core electron to get excited. There is a need for a more efficient and accurate process to analyze spectrums like these to gather the relevant information needed to make valid conclusions about the materials' structure. Technology Overview: This invention revolves around a supervised machine learning-based spectrum analysis. By using a neural network that is pre-trained with spectrum information, it can identify specific attributes or parameters within the spectra it analyses. This then can be translated to information regarding the actual material itself such as atomic structure, electron structure, and so forth. The sample of material that is being recorded through the spectrums should be comprised of both: nanoparticles and a set of coordination numbers of a coordination shell that goes along with those nanoparticles. Then when the spectrum data is processed, the important spectral regions that are relevant in determining the structure of the material are identified. This is done by pre-training the neural network to analyze the spectrum data in the desired way. This training method consists of first inputting a training set of data (which is essentially spectrum data that has already been analyzed) to the neural network. The network itself has various nodes that are arranged in layers where each one of them has an associated value. The training data have the "true" associated values, and they are in relation to specified features of the material. Thus, the neural network processes the training data and outputs a value and then the output value and the "true" value are compared and the nodes are adjusted accordingly to minimize the difference between the two. This process is then used to analyze the actual spectrum data of materials. This is mainly oriented around XAS and EELS data, however, it can be modified for other spectra as well. Advantages: - It allows for a consistent and accurate method to determine the structures of various materials from their spectra. For example, using conventional methods to track the structure of heterogeneous catalysts is a challenge, but this process makes it much more novel and precise. - This process is not very time consuming and allows for immediate X-ray absorption near-edge structure and extended X-ray absorption fine structure analyses. - This method is also a promising approach for high-throughput and time-dependent studies when compared to the current methods used. Applications: - This invention allows for the reconstruction of the average size, shape, and morphology of materials from their respective spectra. - It primarily focuses on analyzing XAS and EELS data and can be effectively used to solve the structures for metal catalysts, however, it can be easily generalized to other materials and spectra by modifying the training data and nodes in the neural network. Intellectual Property Summary: Patent application submitted,Provisional patent Stage of Development: US-2020-0003682-A1 Licensing Potential: Development partner,Licensing,Commercial partner Licensing Status: Available for licensing. R# 9007 Additional Information: machine learning,spectroscopy,energy loss,electronic structure,transmission,x-ray,x-ray photon,photon,spectrum,neural,neural network,spectral,heterogeneous catalyst,reconstruction,morphology,metal catalyst,neuromorphic machine learning,spectroscopy analysis,x-ray spectroscopy,heterogeneous catalysis,absorption spectrometry https://stonybrook.technologypublisher.com/files/sites/phjcl5cr22ivjb9l6x8h_paola-galimberti-xexbi9uxasc-unsplash.jpg Source: Paola Galimberti, https://unsplash.com/photos/XeXBi9uxasc, Unsplash License.