Guangzhou Building Materials Institute

Aiming at critical bottlenecks in the whole process of intelligent concrete production, such as the low security of synchronous transmission between quality traceability data and image carriers, the frequent failure of physical tags, and the high risk of information loss in harsh environments, this paper proposes an advanced high-capacity implicit embedding and lossless traceable image method based on an Improved Prediction Error Expansion (IPEE) framework. The proposed method performs meticulous structural processing on multi-dimensional quality parameters—including mix proportions and environmental sensor data—and utilizes Logistic chaotic mapping for bit-level scrambling to enhance data security significantly. To leverage the inherent high-frequency texture characteristics typical of concrete surfaces, a local-variance-driven adaptive prediction model is developed. This model optimizes the prediction error histogram (PEH) modification by intelligently steering pixel modifications toward high-correlation regions, thereby enabling high-capacity embedding while maintaining superior perceptual transparency and structural fidelity. Experimental evaluations demonstrate that at a standard payload of 0.5 bpp, the PSNR consistently exceeds 48 dB. Remarkably, even when the capacity scales to a high payload of 1.2 bpp, the PSNR remains as high as 41.5 dB, representing a 3.5 dB improvement over traditional PEE by effectively leveraging the high statistical redundancy in fine-grained concrete textures. This method ensures that critical parameters are bound in real-time, achieving pixel-level lossless restoration at the traceability end. Consequently, it provides a complete technical closed-loop for reliable quality evidence storage, supporting high-precision post-analysis and automated recognition tasks on digital construction sites.

Construction sites are characterized by complex environments, high personnel mobility, and resource-constrained conditions like occlusion and uneven lighting. This paper introduces LR-YOLO, a lightweight detection method optimized for real-time hazard identification on edge devices. To address the challenge of environmental disturbances, we implement a robust training strategy incorporating advanced data augmentation and evaluate performance on a specially curated "Adverse Scenario" test set. First, we introduce a structural re-parameterization backbone to reduce memory access cost (MAC) while preserving feature extraction capability. Second, an Efficient Channel Attention (ECA) mechanism is integrated into the neck to enhance sensitivity under complex visual conditions. To bridge the gap between laboratory training and harsh field conditions, a robust training strategy with targeted data augmentation is implemented. Based on the proposed model, an edge-side early-warning logic is designed to suppress false alarms. Experimental results on an NVIDIA Jetson Nano show that LR-YOLO delivers 35.6 FPS at 84.2% mAP@0.5 overall. Notably, LR-YOLO maintains 80.1% mAP on the Adverse Scenario Test Set (ASTS), outperforming the baseline by 5.9%. This confirms its superior reliability and optimal accuracy-speed balance in harsh construction environments.

To meet the requirements for high adaptability and stability in autonomous robotic welding of watertight patch plates within complex shipbuilding environments, this paper presents an integrated approach for weld seam extraction and motion planning based on a global depth camera. The proposed method adopts a multi-stage processing pipeline. First, the raw 3D point cloud acquired by the depth camera is preprocessed and enhanced through robust feature extraction to improve data quality and ensure reliable weld region detection. Subsequently, the weld centerlines and their topological nodes are extracted from the point cloud skeleton, enabling segmentation of the weld seams. A welding sequence is then planned under process constraints, and the torch poses along each seam are estimated based on local geometric characteristics of the weld path. Fi-nally, an initial trajectory is generated to guide subsequent high-precision welding operations. The proposed approach effectively addresses key challenges in watertight patch plate welding, including complex weld distributions, significant depth noise, and spatial constraints, demonstrating strong robustness and environmental adaptability. Experimental results verify that the method can stably generate continuous and feasible welding guidance paths under representative watertight patch plate scenarios. This not only provides high-quality initial trajectories for structured-light-based fine tracking but also establishes a reliable foundation for global perception and motion planning in autonomous robotic welding.

Modular steel buildings represent a structural system distinguished by rapid construction and environmental sustainability. The modular units and steel components of modular steel structures can be recycled, making this approach an important technology for sustainable development. Glass curtain walls, commonly used as facade systems in modern architecture, have recently appeared in several modular steel buildings. In this study, a seven-story model steel building is designed with a geometric scale factor of 1/9 to investigate its global and local safety in terms of seismic responses. Two glass curtain walls are installed on the seventh story of the model structure. A series of shaking table tests is conducted under varying seismic intensity levels (PGA = 0.035 g, 0.1 g, 0.22 g, 0.31 g). The results show the acceleration responses at the top story are predominantly governed by the fundamental translational modes (first mode and second mode). A slight stiffness degradation of a ratio less than 8.0% appears after the tests. The modular steel structure exhibits a significant acceleration amplification effect under almost all examined load cases. The measured peak structural accelerations (PSAs) notably exceed the limitations specified in current codes. The finite element simulation has validated such amplification. In addition, compared to these global responses, the glass curtain walls exhibit even higher PSAs, making them more vulnerable than the main steel frame. Therefore, the unfavorable seismic performance of modular steel buildings is manifested, and more attention needs to be paid to their design principles.

ABSTRACT Industrial Internet of Things (IIoT) systems generate massive behavioral data, demanding efficient analytics frameworks for real‐time monitoring. This study proposes a multi‐dimensional feature fusion framework integrating XGBoost, tailored for IIoT‐driven behavioral pattern recognition. A four‐dimensional architecture is constructed to analyze critical attributes across contact degree, status, duration, and social relations, leveraging edge‐computed IIoT footprints (e.g., mobile signaling, network interaction data). The framework defines three behavioral modes and achieves 98.89% precision, 98.85% recall, and 98.85% F1‐score via XGBoost. Feature importance analysis identifies key indicators such as mobile number status and interaction frequency. This work demonstrates the potential of harmonizing AI with IIoT data fusion, providing a scalable solution for real‐time monitoring in Industrial Internet and future network architectures.