Fault diagnosis of railway assets has drawn the interest of both the scholarly and engineering communities. Federated learning (FL) enables training models across distributed assets to preserve data privacy and reduce high data transfer costs, which has been applied in fault diagnosis. However, the imperfect data transmission problem due to communication errors easily results in low accuracy of FL-based fault diagnosis in the railway system. To solve the problem, a sequential and asynchronous federated learning framework is proposed for fault diagnosis of railway point machines (RPMs) in this work. First, a dual-branch network is proposed as the global model in asynchronous FL for reducing parameters, while maintaining high accuracy. Second, a time cycle mechanism based on sequential Kalman filtering is proposed for reducing the negative impact of data communication errors. Finally, experimental results demonstrates that the proposed method enhances the applicability of online RPM fault diagnosis training in real deployment scenarios.

The Passive Keyless Entry and Start system (PKES) has become an essential element of vehicle systems since it allows owners to lock or unlock their properties without having to take out the keys. However, the system suffers from a potential and serious security problem because of a relay attack. This paper introduces dynamic biometrics to resolve such potential problems. Precisely, it opts for hand-gesture and proposes a Multi-Task Siamese Feature-pyramid network with Bidirectional Gated Recurrent Units (MTSFBet) against relay attack. MTSFBet is made up of a Dynamic Biometrics Extraction Module (DBEM) and Multi-Task Operation Module (MTOM). DBEM extracts trajectory and kinematic features from the gesture data of owners, whereas MTOM performs gesture classification and identity authentication using these features. Moreover, we designed a loss function for the identity authentication part. Eventually, based on the dataset collected by mobile devices, we constructed comparative experiments and ablation studies to demonstrate the effectiveness of the method. Our comprehensive model achieves an accuracy of 0.86 in gesture classification and an accuracy of 0.73 in identity authentication. The results showed that our MTSFBet significantly outperforms other comparison methods. The MTSFBet can also be used in any scenario associated with identity authentication based on hand gestures.

This paper describes an efficient fire detection approach using color and texture information. Our approach consists of Fire Pixel Based Multi-rules Detection and Fire Texture Based Classification. We segment the fire candidate region in HIS and RGB color space using multi rules based on statistic color model, and then employ LBP to extract the fire texture features from those candidate regions. Finally, a SVM classifier is trained to determine the real fire region by these LBP features. To improve the predicting precision, we conduct cross validation to select the best parameters for SVM training model. The performance of the proposed approach is tested on various types of fire scene videos. Testing results show that it is effective on fire alarm and can give an early detection on video surveillance.

Background: Pembrolizumab monotherapy was approved for first-line (1L) treatment of metastatic non-small cell lung cancer (NSCLC) with PD-L1 tumor proportion score (TPS) ≥50% in October 2016 in the US, based on the findings from KEYNOTE-024 trial. A recent update of the trial with 25.2 months median follow-up demonstrated median pembrolizumab treatment duration of 7.9 months. The objective of our study is to estimate real-world time on treatment (rwToT) for 1L pembrolizumab monotherapy in a clinically matched real-world NSCLC database.

Intelligent machine networks provide wireless connectivity for automated guided vehicles (AGVs) in flexible final-assembly workshops. AGVs switch between assembly and non-assembly tasks with different closed-loop interaction periods, which changes packet-generation rates and creates time-varying uplink and downlink traffic demands. Such traffic asymmetry makes fixed or periodic time-division duplexing (TDD) configurations difficult to adapt to task variations. This paper proposes a task-aware closed-loop communication delay analysis and dynamic TDD resource allocation framework for multi-AGV networks. First, the relationship between AGV task states, interaction periods, and uplink/downlink packet arrival rates is established. Finite-blocklength transmission theory and M/M/1 queueing approximations are then employed to characterize transmission and queueing delays and obtain the overall closed-loop communication delay. Based on the task-dependent traffic demands, a reference TDD slot ratio is determined to reflect the required uplink and downlink resources. An event-triggered allocation method is further developed under a limited TDD reconfiguration budget, where the slot ratio is updated only when its deviation from the reference ratio exceeds an offline-selected threshold. To further enhance resource allocation adaptability, a PPO-based method is introduced to dynamically adjust the uplink–downlink slot ratio according to traffic demands, queue conditions, and the remaining reconfiguration budget. Numerical results demonstrate that both proposed methods reduce the average closed-loop communication delay compared with offline-optimized fixed-ratio and periodic allocation. At $N=21$, PPO achieves 1.04770 ms, compared with 1.112 ms for the additional advantage actor–critic (A2C) baseline, corresponding to a 5.78% reduction. PPO also reduces the delay by 19.65% relative to Optimization-Fixed. These results show that adapting TDD resources to task-induced traffic variations can improve multi-AGV closed-loop communication performance under the evaluated settings and limited reconfigurations.