測定固體和液體材料的晶體結構對于理解它們的機械、電磁和熱力學性質是十分重要的。粉末X射線衍射（XRD）是材料表征的重要手段，它編碼了關于晶體對稱性、晶格參數(shù)、類型以及納米級上原子的填充信息。然而，現(xiàn)在的分類方法需要大量的人為干預，根據總體信息綜合評判來完成分類。有許多變量會影響XRD圖案的形狀，如材料的相或晶格，如果沒有已知類似的結構，就很難表征材料。

此外，樣品中存在的一些少量的雜質相可能會導致分類更加困難、耗時和不準確。超快同步X射線衍射和光譜學測量的最新進展在于從數(shù)百萬次測量中產生了極大的數(shù)據集，遠遠超過了人類可以手動分析的數(shù)據量。

Fig. 2 Diffraction pattern comparison.

因此，對XRD數(shù)據進行自適應和自動分析目前存在迫切的需求。目前已開發(fā)的深度學習模型在不同數(shù)據集下的表現(xiàn)差異極大，表現(xiàn)為魯棒性不足的特點。因此，我們需要一個更具有魯棒性的模型，可以對不同材料的動態(tài)和/或看不見的真實XRD數(shù)據進行分類。

來自羅切斯特大學機械工程學院的Niaz Abdolrahim教授小組，開發(fā)了一種用于晶體系統(tǒng)和空間群分類的廣義深度學習模型。由于XRD數(shù)據中的相對峰強度、距離和順序表征了對稱性，研究人員研究了其中是否存在排列不變性和平移不變性，并據此提出了無池化的卷積神經網絡（NPCNN），基于索引峰之間的相對和局部推理來表征材料，以此來完成分類工作。

為了實現(xiàn)廣泛的分類功能，作者還開發(fā)了一個數(shù)據生成流水線來建立高質量的數(shù)據集，該流水線結合了對衍射模式的實驗效應，并且具有模擬經過合金化和/或動態(tài)實驗的材料的能力。最后，研究人員成功使深度學習模型發(fā)揮出了最先進的性能。

該研究也為開發(fā)其他光譜表征技術模型提供了有效的研究思路。相關論文近期發(fā)布于npj Computational Materials?v9:?214?(2023)。

Editorial Summary

Determining the crystal structure of solid and liquid materials is important for understanding their mechanical, electromagnetic and thermodynamic properties. Powder X-ray diffraction (XRD) is an important means of material characterization, encoding information about crystal symmetry, lattice parameters, type, and filling of atoms on nanoscale domains.?

Fig. 7 Lattice augmentation performance.?

However, the current classification method requires a lot of human intervention to complete the classification based on comprehensive evaluation of the overall information. There are many variables that affect the shape of an XRD pattern, such as the phase or crystal lattice of the material. Without a known similar structure, it is difficult to characterize the material. In addition, the presence of some small amounts of impurity phases in the sample may make classification more difficult and time-consuming. and inaccuracies.?

Recent advances in ultrafast synchronized XRD and spectroscopy measurements have generated extremely large data sets from millions of measurements, far exceeding what humans can manually analyze. Therefore, there is an urgent need for adaptive and automatic analysis of XRD data. The performance of currently developed deep learning models on different data sets varies greatly, showing insufficient robustness. A more robust model is needed that can classify dynamic and/or unseen real XRD data obtained from different materials.?

Fig. 9 Scatterplot on MP performance.

A?group led by Prof. Niaz Abdolrahim from the School of Mechanical Engineering, University of Rochester, developed a generalized deep learning model for crystal system and space group classification. Considering that the relative peak intensity, distance and order in XRD data indicate symmetry, the researchers investigated whether there is alignment invariance and translation invariance, and based on this, they proposed a no-pool convolutional neural network (NPCNN). Classification was accomplished by characterizing materials based on relative and local inferences between indexed peaks. To enable extensive classification capabilities, the authors also developed a data generation pipeline to build high-quality data sets that incorporates experimental effects on diffraction patterns. The pipeline also has the capability of simulating materials that undergo alloying and/or dynamic experimentation. The researchers succeeded in making the deep learning model achieve state-of-the-art performance. This study provides a valuable platform for developing models of other spectral characterization techniques.?This article was recently published in npj Computational Materials?v9:?214?(2023).

原文Abstract及其翻譯

Automated classification of big X-ray diffraction data using deep learning models (使用深度學習模型對大X射線衍射數(shù)據進行自動分類)

Jerardo E. Salgado, Samuel Lerman, Zhaotong Du, Chenliang Xu & Niaz Abdolrahim

Abstract In current in situ X-ray diffraction (XRD) techniques, data generation surpasses human analytical capabilities, potentially leading to the loss of insights. Automated techniques require human intervention, and lack the performance and adaptability required for material exploration. Given the critical need for high-throughput automated XRD pattern analysis, we present a generalized deep learning model to classify a diverse set of materials’ crystal systems and space groups. In our approach, we generate training data with a holistic representation of patterns that emerge from varying experimental conditions and crystal properties. We also employ an expedited learning technique to refine our model’s expertise to experimental conditions. In addition, we optimize model architecture to elicit classification based on Bragg’s Law and use evaluation data to interpret our model’s decision-making. We evaluate our models using experimental data, materials unseen in training, and altered cubic crystals, where we observe state-of-the-art performance and even greater advances in space group classification.

摘要 在當前的原位X射線衍射（XRD）技術中，數(shù)據的生成能力超過了人類的分析能力，有可能導致洞察力的損失。自動化技術需要人工干預，并且缺乏材料研究所需的性能和適應性。鑒于高通量自動XRD模式分析的迫切需求，我們提出了一個廣義的深度學習模型來分類不同材料的晶體系統(tǒng)和空間群。在我們的方法中，我們利用來自不同實驗條件和晶體性質模式的整體表示來生成訓練數(shù)據。我們還采用了一種快速學習技術來改進我們模型在實驗條件下的專長。此外，我們優(yōu)化了模型架構，以引出基于布拉格定律的分類，并使用評估數(shù)據來解釋我們模型的決策。我們使用實驗數(shù)據、訓練中未見的材料以及改變了的立方晶體，來評估模型，我們觀察到最先進的性能和空間群分類的更大進步。