GDL (China)

Construction quality, durability and service life of China Railway Track System III (CRTS III) is influenced by the performance of the self-compacting concrete (SCC) used for filling layer. In this paper, the effect of unit water consumption, sand ratio and water reducer components on working performance of SCC filling layer is studied by laboratory and field tests. Appropriate mix proportion parameters are obtained. The results indicate that unit water consumption can significantly improve the in-service performance of SCC. However, it is essential to consider both the dosage of water-reducing agents and the unit water consumption to ensure the stability of the mixture. Sand ratio of SCC is between 49% and 52% used for filling layer, which have excellent flowability and stability. It is recommended that the content of the defoamer should be 3/1 000, and the content of the air-entraining agent should be 5/1000. These research results will play an important role in improving the construction quality of SCC filling layer.

We propose an integrated model predictive control (MPC) scheme for steering-robot path tracking that directly optimizes the robot voltage and embeds steering-angle limits as linear-inequality voltage constraints inside the optimizer. This avoids cascade-induced error accumulation and extra phase lag in MPC+PID while guaranteeing actuator-level feasibility. A Simulink–CarSim co-simulation evaluates two scenarios: (1) double-lane change (DLC) at 70/40 km·h−1; and (2) straight-line tracking with/without sinusoidal crosswind modeled as an equivalent lateral force. Metrics include lateral-error RMS/Peak/P95 and real-time statistics (WCET, average per-update time, and utilization rate). The results show consistent gains: at 70 km·h−1, RMS/Peak/P95 decrease by 22.3%/18.0%/17.7%; and, at 40 km·h−1, by 17.0%/19.5%/18.9%. Real-time feasibility is met with T = 10 ms, average ≈ 1.7 ms, WCET ≈ 2.1~2.3 ms, utilization ratio ≈ 0.17. Under crosswind, robustness improves over the cascaded baseline by 9.7%/35.6%/30.8% on RMS/Peak/P95. The method provides tighter tracking, stronger disturbance rejection, and strict timing for safety-critical testing.

Pancreas segmentation faces two core challenges: anatomical heterogeneity (the head/body/tail volume ratio can vary by up to 2.3-4.7 times between individuals) and low contrast at tissue boundaries (the CT value difference with surrounding organs such as the duodenum is often less than 30 HU). To address the high false positive rate (≥12%) and missed segmentation of small structures in existing methods, this paper proposes the 3D-MERMA-Net model. Its innovations include: 1) a dual-encoder architecture that fuses the global semantic priors of a pre-trained 3D ResNet with the local features of a 3D UNet to improve generalization to small samples and complex boundaries; 2) a multi-path residual attention module that deeply integrates residual connections with multi-head self-attention to effectively model long-range dependencies within organs while mitigating high-level semantic loss; and 3) a multi-scale loss function that combines Tversky Loss, Dice Loss, and weighted cross entropy to optimize model convergence and detail preservation. Experimental results show that the Dice coefficient of the model reaches 78.798% and 77.189% on the MSD (pancreas) and NIH datasets, respectively.

The stable operation of the power system is of key significance to social production and residents’ lives. As an important means to ensure the security of the power network, anomaly detection urgently needs the support of high-precision and structure-aware methods. In view of the shortcomings of traditional methods in describing complex power grid topology and adjacent dependencies, this paper proposes a power network anomaly detection algorithm based on the Graph Attention Network (GAT). Firstly, a graph structure modeling method suitable for the power system is constructed, and nodes and edges are respectively given multi-dimensional physical properties; then a multi-layer and multi-head GAT model structure is designed, and the selective aggregation of key neighborhood information is achieved through learnable attention weights; finally, comprehensive experiments are carried out on IEEE 33-node and 118-node systems. The results show that the proposed algorithm is superior to the comparative methods such as GCN, GraphSAGE, PCA+SVM in terms of F1-score, AUC and other indicators, and has good anomaly separability and structural robustness. This study provides an effective graph learning framework for intelligent perception and risk warning in power networks.

Aiming at the problems of data heterogeneity, abnormal pattern differences and collaborative detection difficulties in multi-regional power grids, this paper proposes a distributed anomaly detection mechanism based on federated transfer learning. First, by constructing abnormal characteristics and disturbance models under regional division, multi-type abnormal scenario simulation and data difference modeling are realized. Secondly, a federated migration framework including maximum mean difference (MMD) and personalized output structure is designed to effectively improve cross-regional model sharing and robustness. Finally, the comprehensive advantages of the proposed method in accuracy, communication efficiency and generalization ability are verified under the IEEE 118 node system, providing technical support for collaborative perception of smart grids.

With the rapid development of UAV technology, UAV has been widely used in agriculture, logistics, surveying and mapping, public safety and other fields. However, in the actual flight process, the UAV may encounter various faults, which not only affect the completion of the flight mission, but also may lead to safety hazards. The purpose of this paper is to discuss the common failure types of UAV in flight, including signal loss, battery failure, mechanical failure, software failure and environmental factors. According to each fault type, this paper provides corresponding emergency treatment strategies to help pilots respond quickly in case of emergency and ensure flight safety. Through the analysis of practical cases, this paper summarizes the experience and lessons of flying hands in fault handling, and emphasizes the importance of pre-flight preparation and real-time monitoring in flight. Finally, this paper looks forward to the future development trend of UAV fault handling, pointing out that with the progress of technology, the intelligence and automation of fault handling will become the development direction of the industry, providing a safer and more efficient flight experience for flying hands.

ABSTRACTThe objective of this study was to analyse the difference in sensory flavour and volatile compounds among the three processing methods of soy whey, as well as to establish an evaluation model representing the correlation between sensory flavour and volatile compounds. The results showed that the soy whey produced by the boiling‐to‐filtering method (BFM) exhibited a higher total soluble solid (TSS) content and total flavonoid content (TFC) compared to that of the repeated boiling‐to‐filtering method (RBFM) and the filtering‐to‐boiling method (FBM), while RBFM soy whey showed a higher total protein content (TPC). The sensory analysis showed that the FBM soy whey exhibited an intense beany flavour and grease odour. The SPME/GC–MS results showed that FBM soy whey exhibited more categories and higher contents of flavour compounds, particularly for alcohols and aldehydes. 1‐Octen‐3‐ol, n‐hexanol, hexanal and nonanal were the main volatile compounds influencing soy whey with different processing methods. The evaluation model revealed that the beany flavour was close to FBM soy whey and highly correlated with 1‐octen‐3‐ol, n‐hexanol, hexanal, heptanal and 2‐ethyl furan. (E, E)‐2,4‐Heptadienal, (E, E)‐2,4‐nonadienal and octanal were close to FBM and highly correlated with grease and bean fragrance. Decanal, ethyl caproate and ethyl caprylate were highly correlated with the floral aroma and close to FBM soy whey. Sweet and fruity aromas were associated with RBFM soy whey and highly correlated with 3‐methyl butanal, (E)‐2‐nonenal and nonanal. The milk aroma was close to BFM soy whey and highly correlated with 2‐ethylfuran, decanal, ethyl caprylate and ethyl caproate. Overall, RBFM and BFM soy whey contained higher levels of protein and flavonoids and exhibited less beany flavour, potentially offering greater nutritional value. In contrast, the FBM soy whey exhibited a higher content of volatile flavour compounds, especially 1‐octen‐3‐ol, n‐hexanol, hexanal and 2‐ethyl furan.