Due to the irregularity and complexity of ground and non-ground objects, filtering non-ground data from airborne LiDAR point clouds to create Digital Elevation Models (DEMs) remains a longstanding and unresolved challenge. Recent advancements in deep learning have offered effective solutions for understanding three-dimensional semantic scenes. However, existing studies lack the capability to model global semantic relationships and fail to integrate global and local semantic information effectively, which are crucial for the ground filtering of point cloud data, especially for larger objects. This study focuses on ground filtering challenges in large scenes and introduces an elevation offset-attention (E-OA) module, which considers global semantic features and integrates them into existing network frameworks. The performance of this module has been validated on three classic benchmark models (RandLA-Net, point transformer, and PointMeta-L). It was compared with two traditional filtering methods and the advanced CDFormer model. Additionally, the E-OA module was compared with three state-of-the-art attention frameworks. Experiments were conducted on two distinct data sources. The results show that our proposed E-OA module improves the filtering performance of all three benchmark models across both data sources, with a maximum improvement of 6.15%. The performance of models was enhanced with the E-OA module, consistently exceeding that of traditional methods and all competing attention frameworks. The proposed E-OA module can serve as a plug-and-play component, compatible with existing networks featuring local feature extraction capabilities.

This study investigated the macro and meso mechanisms of void formation in graded aggregates within high-speed railway subgrades under train loads using a hybrid discrete element–finite difference method (DEM-FDM). First, a contact parameter inversion model based on a linear model (LM) was developed using extensive DEM simulations through angle of repose, drop, and inclined plate tests. The contact parameters for graded aggregates were further calibrated through physical and triaxial tests. Next, a refined hybrid DEM-FDM model was established to capture void formation behavior, characterized by the contact force chain ratio, and was validated against field measurements. Finally, simulations were conducted under different levels of void formation to explore the associated mechanisms based on dynamic response and meso-mechanical analysis. The results showed that the LM-based inversion model could accurately determine the contact parameters. The hybrid model’s predictions of dynamic displacement and acceleration under various train speeds fell within the range of the field data. When the fine particle loss ratio l_p was ≤3%, the dynamic displacement and acceleration remained below the standard limits of 0.22 mm and 10 m/s². As l_p increased, the contact between the roadbed and base weakened, and complete separation occurred at l_p ≥ 11%, preventing effective load transfer. These findings offer new insights into void formation in graded aggregates and support the safe operation of high-speed railways.

In railway operations and safety oversight, vast amounts of accident-related data are recorded in unstructured textual formats, posing challenges for efficient information extraction and analysis. To address this, we constructed a knowledge graph from railway accident profile texts and integrated it with a Graph Retrieval-Augmented Generation (GraphRAG)-enhanced retrieval framework to support both basic and composite queries on railway accident information. First, railway accident profile texts were preprocessed, and Easy Data Augmentation (EDA) was used to expand samples of different types, thereby constructing a railway accident profile text dataset for subsequent knowledge extraction. We then introduced a deep learning model, RoBERTa-CNN-BiLSTM-CRF (RCBC), for automatic extraction of seven categories of entities, including accident IDs and causes. To extract semantic relations, we designed a prompt-based template leveraging large language models (LLMs). To mitigate information loss during entity extraction, a GCN-Attention-LLM (GA-LLM) model was further designed for knowledge graph completion. Experimental results show that RCBC achieves MicroF scores above 85% across entity extraction tasks, while GA-LLM attains an average Hits@3 of 82.84% in knowledge completion. In tests on basic and composite questions, LLM-GraphRAG outperformed both LLM and LLM-RAG in faithfulness, semantic similarity, context precision, and context recall. The resulting knowledge graph contains 1493 entities and 1832 relations. Combined with the retrieval framework, the system enables access to key railway accident information and offers technical support for intelligent railway safety management.

Purpose: The indoor vibration compaction test (IVCT) was a key step in controlling the compaction quality for high-speed railway graded aggregate (HRGA), which currently had a research gap on the assessment indicators and compaction parameters.Design/methodology/approach: To address these issues, a novel multi-indicator IVCT method was proposed, including physical indicator dry density (ρ_d) and mechanical indicators dynamic stiffness (K_rb) and bearing capacity coefficient (K₂₀). Then, a series of IVCTs on HRGA under different compaction parameters were conducted with an improved vibration compactor, which could monitor the physical-mechanical indicators in real-time. Finally, the optimal vibration compaction parameters, including the moisture content (ω), the diameter-to-maximum particle size ratio (R_d), the thickness-to-maximum particle size ratio (R_h), the vibration frequency (f), the vibration mass (M_c) and the eccentric distance (r_e), were determined based on the evolution characteristics for the physical-mechanical indicators during compaction.Findings: All results indicated that the ρ_d gradually increased and then stabilized, and the K_rb initially increased and then decreased. Moreover, the inflection time of the K_rb was present as the optimal compaction time (T_lp) during compaction. Additionally, optimal compaction was achieved when ω was the water-holding content after mud pumping, R_d was 3.4, R_h was 3.5, f was the resonance frequency, and the ratio between the excitation force and the M_c was 1.8.Originality/value: The findings of this paper were significant for the quality control of HRGA compaction.

Purpose: This study aims to enhance the accuracy of key entity extraction from railway accident report texts and address challenges such as complex domain-specific semantics, data sparsity and strong inter-sentence semantic dependencies. A robust entity extraction method tailored for accident texts is proposed. Design/methodology/approach: This method is implemented through a dual-branch multi-task mutual learning model named R-MLP, which jointly performs entity recognition and accident phase classification. The model leverages a shared BERT encoder to extract contextual features and incorporates a sentence span indexing module to align feature granularity. A cross-task mutual learning mechanism is also introduced to strengthen semantic representation. Findings: R-MLP effectively mitigates the impact of semantic complexity and data sparsity in domain entities and enhances the model's ability to capture inter-sentence semantic dependencies. Experimental results show that R-MLP achieves a maximum F1-score of 0.736 in extracting six types of key railway accident entities, significantly outperforming baseline models such as RoBERTa and MacBERT. Originality/value: This demonstrates the proposed method's superior generalization and accuracy in domain-specific entity extraction tasks, confirming its effectiveness and practical value.