Integrated system with digital twin technology for enhancing safety in crane lifting operations

Abstract

The construction industry drives economic growth in both developing and developed countries, leading to an increasing demand for construction projects and activities. Modern construction practices increasingly adopt off-site prefabrication of heavy modules due to limited onsite space, which requires transportation and crane-based installation. Traditionally, certified operators manage lifting operations based on manual judgment, which often leads to accidents caused by occlusion, collision, overloading, or unsafe practices. According to data from the Hong Kong Housing Authority, 38% of crane-related accidents occur during lifting operations, underscoring the urgent need for improved safety measures. In response, Hong Kong introduced the Smart Site Safety System (4S) in 2023. However, 4S relies on manual reporting and suffers from limitations in accuracy and adaptability, necessitating advanced technological solutions to enhance safety and reduce risks in lifting operations. This thesis addresses three types of hazards during crane lifting operations and covers all contributing factors associated with these risks. These risks include occlusions/collisions, overloading, and unsafe practices. First, for occlusions/collisions, existing Computer-Aided Lift Path Planning and Replanning (CALPP-RP) are using parallel computing systems, but there are several challenges in parallel computing systems, including computational inefficiency, limited exploration capability, instability due to suboptimal problem division, and communication latency in dynamic environments. To overcome these limitations, this thesis proposes an integrated computing system for occlusion-free and collision-free path planning and replanning of robotized cranes. The workspace is modeled in Unity 3D and utilizes hybrid algorithms, combining A* for Configuration Space (c-space) exploration and Genetic Algorithms (GA) for path optimization. The Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) is employed to select optimal and sub-optimal paths, while raycasting is used for occlusion and collision detection. The system integrates a Decision Support System (DSS) for proactive planning and a Path Re-Planner (PRP) for real-time updates. Pilot testing demonstrated improvements of 83.33% in computation speed and 33.32% in overall performance in high-dimensional environments compared to traditional sequential and parallel computing systems. CALPP-RP systems integrated with the proposed computing framework help avoid occlusion and collisions during lifting operations. Although cranes may operate for up to 60 years of Remaining Useful Life (RUL), they are subjected to cyclic and variable amplitude loads over time. These loads induce fatigue and stress in the crane structure, which accumulate and may eventually exceed critical thresholds, potentially leading to structural failure. Therefore, a system is needed to predict this degradation based on real-time sensor data. Second, aging tower cranes face structural degradation, yet their lifting capacity over time remains underexplored. This thesis presents a Digital Twin-Driven (DTD) model to predict the degraded lifting capacity of aging cranes. The DTD model integrates fatigue analysis, real-time data, and Machine Learning (ML) models, achieving high accuracy (Mean Squared Error (MSE) = 0.2253, coefficient of determination (R²) = 0.9973) in predicting degraded load charts over a 70-year lifespan. These predictions enhance safety monitoring, enabling operators to mitigate risks and prevent structural failures. Even after addressing occlusion, collision, and structural degradation, crane accidents occur due to unsafe practices. Therefore, there is a need for an automated safety risk assessment system that monitors all lifting activities and operations in real-time. Third, traditional safety monitoring relies heavily on manual reporting, often error-prone and limiting its effectiveness. To advance 4S technologies, a cascaded model for automated safety risk assessment is proposed. It integrates Super-Resolution Generative Adversarial Networks (SRGAN) for image preprocessing, Real-Time Detection Transformer-Large (RT-DETR-L) for crane detection, self-DIstillation with NO labels (DINOv2) for safety classification, and Vision Transformer (ViT) for activity recognition. Comprehensive risk values are calculated using probability matrices, triggering real-time warnings for high-risk scenarios. The model demonstrates superior performance, achieving 92.10% detection precision, 99.25% safety classification accuracy, and 99.47% activity classification accuracy, with an inference speed of 0.70 seconds, providing a reliable, real-time monitoring solution. This research significantly advances crane safety in construction through integrated computing, predictive maintenance, and automated risk assessment. These innovations enhance efficiency, accuracy, and safety, offering scalable solutions to address critical challenges in dynamic construction environments.