Xiaohu Zhao’s research while affiliated with China University of Mining and Technology and other places

Face recognition technology, despite its widespread use in various applications, still faces challenges related to occlusions, pose variations, and expression changes. Three-dimensional face recognition with depth information, particularly using point cloud-based networks, has shown effectiveness in overcoming these challenges. However, due to the limited extent of extensive 3D facial data and the non-rigid nature of facial structures, extracting distinct facial representations directly from point clouds remains challenging. To address this, our research proposes two key approaches. Firstly, we introduce a learning framework guided by a small amount of real face data based on morphable models with Gaussian processes. This system uses a novel method for generating large-scale virtual face scans, addressing the scarcity of 3D data. Secondly, we present a dual-branch network that directly extracts non-rigid facial features from point clouds, using kernel point convolution (KPConv) as its foundation. A local neighborhood adaptive feature learning module is introduced and employs context sampling technology, hierarchically downsampling feature-sensitive points critical for deep transfer and aggregation of discriminative facial features, to enhance the extraction of discriminative facial features. Notably, our training strategy combines large-scale face scanning data with 967 real face data from the FRGC v2.0 subset, demonstrating the effectiveness of guiding with a small amount of real face data. Experiments on the FRGC v2.0 dataset and the Bosphorus dataset demonstrate the effectiveness and potential of our method.

The timeliness of shared data is one of the important factors affecting efficient collaboration among smart factories (SFs). Although the blockchain-based edge computing data sharing scheme has achieved certain results in data sharing efficiency and security, the PBFT consensus algorithm widely used in the industrial Internet of Things (IIoT) still faces the problems of large communication overhead, single selection of master nodes, and lack of processing of malicious nodes. In order to solve the above problems, we propose a secure and efficient intelligent grouping PBFT (IG-PBFT) consensus algorithm for the industrial internet of things. First, we design and implement the consensus group intelligent construction method, divide the simulated SF into different consensus groups, and reduce the communication overhead of the blockchain network through two consensus processes within and among the consensus groups. Then, we propose a group leader smart factory intelligent selection method, which selects a SF for each consensus group based on the trust degree of the SF, and the SF represents the consensus group to participate in the consensus among the consensus groups, which improves the security of the blockchain network. Then, we design and implement a constrained smart factory intelligent labeling method that labels SFs based on the category of SFs in the current consensus cycle to prevent constrained SFs from participating in the consensus process. Finally, we conducted simulation experiments on the IG-PBFT consensus algorithm. Through theoretical and simulation experiments, it is proven that the IG-PBFT consensus algorithm can reduce the communication overhead of the traditional PBFT consensus algorithm, improve the security of the consensus process, and have obvious advantages in the IIoT.

Data sharing provides necessary data support for collaborative production among Smart Factories (SFs) and improves the production efficiency of the manufacturing industry. However, the quality of shared data cannot meet the needs of collaborative production of high-precision manufacturing processes among SFs, which hinders in-depth cooperation among SFs to a certain extent. In order to solve the above problems, we propose a lightweight sensing data integrity detection method for the Industrial Internet of Things that performs integrity detection on the shared sensing data before the SF shares the data. Firstly, we design and implement a sensing data integrity feature extraction method to extract the integrity features of sensing data. Then, we design and implement a lightweight sensing data integrity detection method based on fuzzy support vector machines that detects the integrity of sensing data based on the integrity characteristics of sensing data. Finally, we conduct simulation experiments on the proposed method. Through theoretical and simulation experiments, it has been proven that compared with the traditional method, the accuracy of the proposed method is improved by 11.40%, the False Alarm Rate (FAR) is reduced by 79.49%, the Missing Alarm Rate (MAR) is reduced by 71.80%, the detection time is reduced by 72.40%, and the energy consumption is reduced by 12.04%.

(1) Background: Rolling bearings are important components in mechanical equipment, but they are also components with a high failure rate. Once a malfunction occurs, it will cause mechanical equipment to malfunction and may even affect personnel safety. Therefore, studying the fault diagnosis methods for rolling bearings is of great significance and is also a current research hotspot and frontier. However, the vibration signals of rolling bearings usually exhibit nonlinear and non-stationary characteristics, and are easily affected by industrial environmental noise, making it difficult to accurately diagnose bearing faults. (2) Methods: Therefore, this article proposes a rolling bearing fault diagnosis model based on an improved dung beetle optimizer (DBO) algorithm-optimized variational mode decomposition-convolutional neural network-bidirectional long short-term memory (VMD-CNN-BiLSTM). Firstly, an improved DBO algorithm named CSADBO is proposed by integrating multiple strategies such as chaotic mapping and cooperative search. Secondly, the optimal parameter combination of VMD was adaptively determined through the CSADBO algorithm, and the optimized VMD algorithm was used to perform modal decomposition on the bearing vibration signal. Then, CNN-BiLSTM was used as the model for fault classification, and hyperparameters of the model were optimized using the CSADBO algorithm. (3) Results: Finally, multiple experiments were conducted on the bearing dataset of Case Western Reserve University, and the proposed method achieved an average diagnostic accuracy of 99.6%. (4) Conclusions: Experimental comparisons were made with other models to verify the effectiveness of the proposed model. The experimental results show that the proposed model based on an improved DBO algorithm optimized VMD-CNN-BiLSTM can effectively be used for rolling bearing fault diagnosis, with high diagnostic accuracy, and can provide a theoretical reference for other related fault diagnosis problems.