Search

The complexity of the structure and operating process of the large-scale petrochemical equipment makes it difficult to realize effective health management. In this article, a natural gas dehydration process digital twin (DT) is proposed for condition evaluation. The framework and detailed DT models are introduced. The geometric, process, behavioral, and rule models in DT map physical entities geometric information, working process, control process, and hidden condition evolution process, respectively. A triethylene glycol dehydration DT application system is developed and the overall reality-virtual mapping process of DT for equipment condition evaluation is demonstrated. The ability of the DT to represent the petrochemical processes is verified with real monitoring data. The condition evaluation ability of the DT application system is validated by failure event and typical process parameters. The methodology proposed in this article can be applied to petrochemical fields to improve the safety of the operational process.

Trajectory data mining can help us mine valuable information such as human activity patterns and movement characteristics, and provide data support for transportation planning, urban design, crime prevention, and behavior analysis. Graph data have been gradually applied in the field of trajectory data processing due to their advantages in easy construction and ability to solve complex relationship problems quickly. However, existing algorithms that combine graph data for trajectory data processing only consider movement patterns and traffic from a spatial perspective, with less attention paid to human behavioral characteristics. The algorithm for processing trajectory data is complex and lacks interpretability when faced with results. Aiming at the above problems, this article proposes a new graph construction method based on trajectory clustering and label classification and performs multiscale analysis based on graph wavelet transform. Experimental results show that this method can effectively extract the behavioral characteristics from trajectory data, achieving higher accuracy in abnormal behavior detection and group partitioning than existing methods.

Prediction of the case depth is an important indicator to evaluate and control the surface quality of the mechanical parts. However, a large transition zone with a smooth hardness gradient is the typical feature of deep induction-hardened steel, limiting the application of nondestructive methods. In this article, a novel approach combining the linear regression method and pulse-echo transverse-to-transverse (PE T-T) backscatter technique is proposed. 12 cylindrical 40Cr steel specimens were prepared using various conditions of induction hardening treatment and classified into three groups with different case depths ranging from 2.7 to 4.6 mm. The hardening of the surface region forms a stratified structure with layers of different microstructural characteristics caused by changes in grain size. We employed a transverse-to-transverse scattering model to extract the gradient feature of the grain size changes in the depth direction. The extracted gradient profiles of different case depths exhibit an S-shape behavior due to the extension of the transition zone. As the case depth increases, the positions of the midpoint increase and rising gradients become steeper. Based on these observations, a four-parameter sigmoid function was used to parameterize the gradient profile curve. The relationships among the obtained gradient feature parameters and case depth measured from microhardness testing have been established. The predicted results show that the combination of the linear regression method and PE T-T technique can be a potential tool to investigate differences in gradient microstructure and to determine the case depth even for the smooth hardness gradient.

Due to the complex transmission structure, fault diagnosis of helicopter transmission systems poses a significant challenge. Data privacy has aroused widespread attention for the interest conflict, especially for some special occasions such as helicopters. Furthermore, the existing graph construction methods overlook the chain structure of helicopter transmission systems, leading to limited model generalization ability, and unknown weight distribution between the source and target models during federated generalization. To address these issues, we propose multisensor graph adaptive federated generalization (MGAFG) for helicopter transmission system fault diagnosis. First, the graph is constructed by multisensor data. Next, the domain shift of different clients is measured by auxiliary domain-based adversarial learning. Specifically, self-supervised learning is introduced to learn the latent distribution features of target client data. Finally, the different fault types are classified by the generalized model. Two different datasets are used to verify the effectiveness of MGAFG. The experiment results show that MGAFG has the highest diagnosis accuracy than other methods.