The absence of global navigation satellite system (GNSS) signals in shield tunnels presents a challenge for robotic automation. Visual simultaneous localization and mapping (V-SLAM) is widely used for robot localization in GNSS-denied scenarios, but the uniform geometry of shield tunnels hinders visual feature extraction. This study proposes a point-line feature fusion V-SLAM method that combines oriented FAST and rotated BRIEF (ORB) feature points with line segments detected by the line segment detector (LSD) algorithm and matched using the line band descriptor (LBD) algorithm. These features are integrated into a map for autonomous localization. Data from a metro shield tunnel project validate the effectiveness of this method. The results show that compared with the point feature approach, most metrics improved by 30.00%, with a maximum improvement of 49.39%. The optimization also enhanced the robustness against image degradation and improved the performance as the mileage increased, with some cases showing reduced error accumulation over longer distances.
This study systematically elaborates on the core connotation and implementation pathways of the intelligent construction system from the contractors' perspective. It analyzes a digitally-driven intelligent construction framework and proposes replicable and scalable systematic solutions for intelligent construction in practical engineering projects. Furthermore, by integrating theory with practice, it validates the significant effectiveness of intelligent construction systems in enhancing efficiency, optimizing management, and mitigating risks. It can provide a reference for industry transformation and upgrading, promote the deep integration of intelligent construction technologies with engineering practices, and ultimately support the achievement of high-quality project delivery and whole-lifecycle value enhancement.
The safety hazards associated with hydropower construction are diverse and often occur in complex and variable spatial contexts. Human visual assessments based on experience are prone to cognitive and psychological biases. Existing studies face key limitations, including unimodal models failing to capture cross-modal hazard features, the limited generalization ability of small models, and the poor domain adaptability of general-purpose pre-trained models. To address these challenges, this study establishes the first multimodal image-text dataset for safety hazards in hydropower engineering. By leveraging the Qwen2.5-VL model, we implement efficient domain adaptation through LoRA and instruction tuning. A multimodal large model fine-tuned for the intelligent recognition of safety hazards in hydropower construction was proposed. Comparative experiments across hazard types, modalities, and model architectures reveal that: (1) data imbalance has a limited impact on performance differences across hazard types, (2) textual descriptions generally convey more critical hazard-related information than visual features, and (3) domain-specific fine-tuning and effective multimodal fusion are identified as key factors in enhancing hazard recognition performance.
Accurate detection of cracks in concrete structures is essential for ensuring civil engineering safety. Traditional manual inspection and machine vision methods suffer from high subjectivity, low efficiency, and poor adaptability to complex environments. In recent years, machine learning (ML) has significantly improved the precision and automation of crack detection. This review provides a systematic examination of ML-based crack detection through a unique four-stage pipeline-data acquisition, preprocessing, model training, and evaluation. Additionally, the integration of multimodal data and unmanned aerial vehicle (UAV)-based inspection has expanded the scope and applicability of detection technologies. Despite these advancements, challenges remain, including limited accuracy for fine crack detection, immature dynamic monitoring techniques, and lack of industry standards. Future research should focus on optimizing model design, incorporating temporal analysis and uncertainty quantification, and developing intelligent structural health assessment systems. This review offers a technical reference for applying machine learning in crack detection and highlights future directions for interdisciplinary integration and standardization.
Residential construction projects are characterized by high complexity, fragmented communication, and vulnerability to delays and budget overruns due to inefficient manual coordination. This paper presents Civil2PM, an adaptive, on-premises multi-agent project lifecycle management (PLM) system built to autonomously oversee the entire life cycle of residential construction projects. The system is designed to ensure robust cyber-security, maintain cost-efficiency, and seamlessly integrate with existing building information modeling (BIM) and enterprise resource planning (ERP) infrastructure. It employs compact yet high-performance large language models (LLMs), Phi-4-14B, and Qwen3-30B-A3B, running locally to ensure both speed and data sovereignty. These models are orchestrated through the LangChain framework, enhanced with agentic retrieval-augmented generation (ARAG), and supported by a dual-protocol architecture comprising the agent-to-agent (A2A) and model context protocol (MCP). The A2A protocol enables secure, structured communication among specialized artificial intelligence (AI) agents (initiation, tracking, and reporting), while MCP provides standardized and isolated access to enterprise data sources. Civil2PM was trained and contextualized on 3500 real-world residential construction project cards, enabling it to autonomously generate project plans, track progress, issue communications, and update databases. By automating these processes, the system significantly reduces human errors in project coordination. A multi-phase evaluation with 11 medium and large construction enterprises demonstrated the system's capability to reduce coordination latency, improve schedule adherence, and eliminate dependence on costly cloud application programming interface (APIs). This work contributes a novel agentic-AI-driven architecture for residential construction sector, merging compact LLMs, multi-agent coordination, and secure local deployment to address pressing economic and operational challenges.
