This study presents an innovative performance prediction and optimization method for prefabricated bridge structures, which integrates the strengths of Physical Information Neural Network (PINN) and Building Information Modeling (BIM). By combining the computational power of PINN for solving structural mechanics problems with BIM’s data management and modeling capabilities in construction engineering, a multi-layer, multi-dimensional intelligent framework for prediction and optimization is developed. BIM models provide crucial data, including geometric details, material properties, and construction process information, which are used to create a digital twin of the bridge. Leveraging PINN’s robust fitting ability, physical laws such as stress, strain, and equilibrium equations are incorporated into the neural network to predict the bridge’s performance under various operational conditions. The integration of physical constraints during PINN training enhances both the accuracy and reliability of the predictions. Subsequently, optimization algorithms refine the bridge design based on these predictions, ensuring improvements in performance while meeting key criteria such as safety, cost-effectiveness, and environmental impact. This method offers a scientifically rigorous and highly accurate approach to supporting the design and construction of prefabricated bridges, advancing intelligent construction and sustainable building practices. Experimental results demonstrate that the integrated PINN-BIM model outperforms traditional methods, exhibiting superior accuracy, generalization capability, and offering a novel solution for efficient bridge engineering construction and performance optimization.

Large-scale irregular cracks developed on the second-floor roof slabs of an industrial plant shortly after formwork removal. To identify the cracking mechanism and optimize the strengthening scheme, field inspection, X-ray diffraction (XRD) phase analysis, restrained-shrinkage stress calculation, and nonlinear finite element (FE) simulation were combined in a closed-loop study. The concrete strength, reinforcement arrangement, slab thickness, and flexural capacity all met the specification requirements, and XRD detected no crystalline sulfate-bearing phases within the 1–2 wt.% detection limit, ruling out structural overloading and sulfate-type chemical attack. The cracking was attributed mainly to excessive restrained drying-shrinkage stress caused by mix-proportion defects (a high water-to-binder ratio, excessive fly ash, and a high mud content in the fine aggregate); the calculated maximum restrained shrinkage stress of 3.19 MPa exceeded the concrete tensile strength. Preferential microcrack initiation in the aggregate–mortar interfacial transition zone (ITZ) is presented as a proposed mesoscale mechanism, because direct microstructural (SEM) observation was not performed. Under identical total CFRP consumption, a four-dimensional evaluation (bearing capacity, material utilization, crack suppression, and relative interfacial slip) showed that the grid-shaped layout outperformed the cross- and hash-shaped layouts, and a ±10% sensitivity analysis kept all predicted crack-width variations within ±5%. A composite remediation of low-pressure grouting plus grid-shaped CFRP was then applied; after 12 months of monitoring, the maximum crack width decreased from 0.32 to 0.04 mm, demonstrating the synergistic effect of the composite system rather than CFRP alone. This case provides a reference for the diagnosis and treatment of shrinkage cracking in comparable industrial roof slabs.

Deep subway station excavation in dense urban areas can induce rapid municipal pipeline deformation under complex, time-varying conditions. Existing statistical models often struggle to capture nonlinear spatiotemporal deformation patterns, while numerical simulation approaches remain computationally demanding for ultra-short-term, high-frequency forecasting. In addition, current Building Information Modeling (BIM) applications are still used mainly for static visualization and are not well integrated with real-time monitoring data. To address these limitations, this study proposes a BIM-integrated intelligent prediction framework driven by an improved spatiotemporal Swin-U-Net (ST-SwinUnet) model for ultra-short-term pipeline deformation forecasting. A four-dimensional BIM/GIS-based digital twin is constructed by integrating geometric semantics, geological information, construction schedules, and hourly monitoring data into a unified spatiotemporal data cube. On this basis, prediction-oriented feature engineering is designed to represent spatial relationships, construction sequences, and time-dependent deformation effects. The ST-SwinUnet model further incorporates hierarchical working-condition embedding and a memory-enhanced attention mechanism to improve adaptability under complex operating scenarios. The framework was evaluated using an eight-month monitoring dataset from a deep subway station excavation project under a chronological train/validation/test design. On the 1–6 hour forecasting task, ST-SwinUnet achieved an average RMSE of 3.14 mm, corresponding to reductions of 46.9%, 37.1%, and 18.7% relative to LSTM, ConvLSTM, and the original Swin-U-Net, respectively. Additional multi-seed experiments further showed that the proposed model maintained stable predictive performance across repeated runs. These findings support the proposed framework as a reproducible project-scale proof-of-concept for ultra-short-term deformation assessment and rolling early warning, while broader cross-project validation remains necessary in future work.