Existing building management systems face critical limitations in real-time data integration, primarily relying on static models that lack dynamic updates from IoT sensors. To address this gap, this study proposes a novel system integrating MQTT over WebSocket with Three.js visualization, enabling real-time sensor-data binding to Building Information Models (BIM). The architecture leverages MQTT’s lightweight publish-subscribe protocol for efficient communication and employs a TCP-based retransmission mechanism to ensure 99.5% data reliability in unstable networks. A dynamic topic-matching algorithm is introduced to automate sensor-BIM associations, reducing manual configuration time by 60%. The system’s frontend, powered by Three.js, achieves browser-based 3D visualization with sub-second updates (280–550 ms latency), while the backend utilizes SpringBoot for scalable service orchestration. Experimental evaluations across diverse environments—including high-rise offices, industrial plants, and residential complexes—demonstrate the system’s robustness: Real-time monitoring: Fire alarms triggered within 2.1 s (22% faster than legacy systems). Network resilience: 98.2% availability under 30% packet loss. User efficiency: 4.6/5 satisfaction score from facility managers. This work advances intelligent building management by bridging IoT data with interactive 3D models, offering a scalable solution for emergency response, energy optimization, and predictive maintenance in smart cities.

Accurate identification of rock mass fractures is essential for evaluating tunnel stability and ensuring construction safety. However, existing deep learning-based approaches frequently exhibit limited performance in challenging tunnel environments and often lack integration with key engineering classification standards, thereby offering insufficient direct support for engineering decision-making. To address these issues, this study introduces an intelligent framework that integrates an improved YOLOv8-seg model with a standardized rock mass classification system. The improved model incorporates an Efficient Channel Attention (ECA) mechanism, which substantially enhances the detection of fracture features under complex conditions. Furthermore, the framework automatically extracts key fracture parameters, including fracture width and filling state. These parameters are directly correlated with the rock mass classification criteria specified in the Chinese National Standard GB 50487-2008. Experimental results demonstrate that the proposed method achieves a notable improvement in detection accuracy, evidenced by a 1.80% increase in mAP@0.5. At the same time, it retains real-time processing capabilities. The automated classification outcomes exhibit strong consistency with manual expert assessments, providing a reliable and efficient tool for supporting engineering decisions in tunnel construction.

The staggered grid finite difference method has emerged as one of the most commonly used approaches in finite difference methodologies due to its high computational accuracy and stability. Inevitably, discretizing over time and space domains in finite difference methods leads to numerical artifacts. This paper introduces a novel approach that combines the widely used Taylor series expansion with the least squares method to effectively suppress numerical dispersion. We have derived the coefficients for the staggered grid finite difference method by integrating Taylor series expansions with the least squares method. To validate the effectiveness of our approach, we conducted analyses on accuracy, dispersion, and stability, alongside simple and complex numerical examples. The results indicate that our method not only inherits the capabilities of the original Taylor series and least squares methods in suppressing numerical dispersion across small and medium wavenumber ranges but also surpasses the original methods. Moreover, it demonstrates robust dispersion suppression capabilities at high wavenumber ranges.