Shu Zhan

The inverse design of materials represents a pivotal technology in the domains of materials science and information engineering. It facilitates the prediction of material properties and structures from desired functionality, thereby accelerating the development of new materials. Nevertheless, conventional material design methodologies frequently necessitate the utilization of extensive computational simulations and experiments, which are inherently time‐consuming and resource‐intensive. A novel approach to materials inverse design is put forth, namely a rotationally invariant crystallographic representation (RICR) coupled with a variational autoencoder (VAE) generative model. The RICR effectively encapsulates the salient features of the crystal structure while preserving rotational invariance through the incorporation of rotational coordinate features. By conducting inverse design experiments on data from the Materials Project database, new crystal materials with user‐specified band gaps and formation energies are successfully design and validated. The results demonstrate that RICR achieves improvements in reconstruction accuracy, performance in mapping regression branch attributes, and the success rate of generating target crystal designs. These findings indicate the effectiveness of RICR in inverse materials design and suggest that this method can provide robust theoretical support for inverse design across diverse materials, thereby advancing the development and innovation of materials science.

With the development of science and technology and the improvement of hardware computing power, the application of large models in the field of artificial intelligence (AI) has become a current research hotspot Among the focal points in the field of deep learning, AI for science is one of the highlighted areas, utilizing deep learning methods for pattern recognition, anomaly detection, predictive analysis, and more on a large scale of scientific data. In the realm of materials science, the structure of crystals is composed of edges and nodes, making it readily representable as a graph. In previous research, some typical models, such as the MEGNet model, utilized their graph neural network features to fit computational results based on density functional theory for predicting various material properties. Building on this concept, the authors propose a novel graph neural network (GNN) model, optimized with a Multi-Head Self-Attention (MHSA) mechanism, for predicting materials data with crystal structures. This model is named self-attention enhanced graph neural network. The model segments the input data into three parts: edges, nodes, and global features. The graph convolutional layer module is primarily used for aggregating node, edge, and global features, learning node representations, and capturing higher-order neighborhood information through multiple layers of GNN. The MHSA component allows nodes to learn global dependencies, providing different representation subspaces for the nodes. In comparison with other machine learning and deep learning models, the results indicate an improvement in the predictive accuracy of this model. A new graph neural network (GNN) model called Self-Attention Enhanced Graph Neural Network (SA-GNN) is proposed for predicting the properties of materials with crystal structures. This model incorporates multi-head self-attention to allow nodes to learn global dependencies and generate different representational subspaces. Compared to other machine learning and deep learning models, the results show improved predictive accuracy, demonstrating the potential of graph networks combined with self-attention for modeling crystal material data.

In cognitive science, the bottleneck of information processing capabilities promotes humans to selectively focus on part of the visible information while ignoring the rest. This is usually called the attention mechanism. Among different attention mechanisms, spectrum attention focuses on assigning every frequency channel obtained by DCT an appropriate weight to achieve adaptive filtering and sends the reweighting frequency components into a real-valued neural network. However, DCT is a real-valued transformation for signal processing thus phase information tends to be omitted. Here we consider a departure from these conventions and propose a spectrum attention complex neural network assisted by DFT, to make a holistic frequency analysis in a completely complex domain. Our proposed network demonstrates outstanding performance on a one-dimensional hyperspectral dataset and MNIST dataset, compared to both traditional complex neural networks and DCT based frequency attention methods.

Serious leakage will cause environmental pollution and safety problems, so strict leakage measurement is necessary for sealed products before leaving the factory. In this article, a semantic-segmentation-based micro-leakage measurement framework (MLMF) is proposed, and the core modules in MLMF are designed and analyzed. First, the DeepLabv3+ algorithm is improved from three aspects, including the design of the fusion layer, the introduction of the attention mechanism of the segmentation model, and the reconstruction of the loss function, which improve the precision of image segmentation tasks. Then, the volume of the intermediate cavity (ICV), which affects the accuracy of leakage measurement, is studied, and the reasonable selection interval of the ICV has been obtained. Finally, through the experimental analysis of the influencing factors of the bubble characteristics in the two-phase flow, the key parameters affecting the bubble shape are found, and the control of the initial shape, generation period, and oscillation process of the bubble is realized. The experimental analysis is carried out through the established experimental platform to verify the effectiveness of the proposed MLMF. The bubble method is innovatively combined with deep learning, and a series of problems, such as bubble shape control and liquid media selection, are solved. It provides a novel low-cost dry measurement mechanism for the field of micro-leakage measurement of sealed vessels.

In the process of measuring micro-leakage based on the bubble method, the physical parameters of the liquid phase have a great impact on the formation and movement of bubbles, thus affecting the measurement accuracy of the system. Therefore, it is of great significance to study the physical parameters of the liquid phase. At present, it is difficult to independently study the parameters of the liquid phase by experimental methods. In this paper, based on the Cahn-Hilliard phase field model, a gas-liquid two-phase flow model for the expression of leakage is established to independently study the influence of different physical parameters on the formation and movement of bubbles in the measuring system. Firstly, the effects of liquid medium density, dynamic viscosity coefficient and surface tension on the initial shape (aspect ratio), the generation period and the shape oscillation of bubbles after they leave the nozzle are analyzed. Then, an experimental platform for visualization of bubbles in two-phase flow based on machine vision is built, through which the morphological parameters of bubbles in the liquid phase can be obtained in real time. Finally, the results of visualization experiment and numerical simulation are compared. The results show that the initial aspect ratio error of bubbles obtained by visual experiment and numerical simulation is 1.8%, and the average oscillation period error is 12.3%. The results are basically consistent, which proves the rationality and correctness of the mathematical model established. It is found that the liquid with larger dynamic viscosity coefficient μ and density ρ and smaller surface tension σ can be selected as the liquid medium of the measurement system. It is more conducive to the accurate expression of micro-leakage in the measurement system, and the bubble shape can quickly reach a stable state. It provides theoretical support for improving the accuracy and stability of the micro-leakage measurement system.

Single Shot Multibox Detector (SSD) uses multi-scale feature maps to detect and recognize objects, which considers the advantages of both accuracy and speed, but it is still limited to detecting small-sized objects. Many researchers design new detectors to improve the accuracy by changing the structure of the multi-scale feature pyramid which has proved very useful. But most of them only simply merge several feature maps without making full use of the close connection between features with different scales. In contrast, a novel feature fusion module and an effective feature enhancement module is proposed, which can significantly improve the performance of the original SSD. In the feature fusion module, the feature pyramid is produced through iteratively fusing three feature maps with different receptive fields to obtain contextual information. In the feature enhancement module, the features are enhanced along the channel and spatial dimensions at the same time to improve their expression ability. Our network can achieve 82.5% mean Average Precision (mAP) on the VOC 2007 [ ], 81.4% mAP on the VOC 2012 [ ] and 34.8% mAP on COCO [ ]-[ ]2017, respectively, with the input size [ ]. Comparative experiments prove that our method outperforms many state-of-the-art detectors in both aspects of accuracy and speed.