Data-Driven Bridge Digital Twin: Cable Tension Estimation Under Thermal Effects via Cloud and FEA APIs

Engineers from the CECI Facilities Technology Center and National Cheng Kung University presented a data-driven digital twin framework for a three-span, twin-pylon extradosed bridge in CECI Engineering Technology. The bridge has a total length of approximately 230 meters and features 24 external cables monitored by 4G and NB-IoT sensing networks recording tension, strain, displacement, acceleration, and temperature data.\n\nThe team created a Python graphical user interface connecting cloud Web APIs with MIDAS Civil NX APIs to automate data ingestion and FEA executions. To minimize daytime solar radiation and traffic disturbances, the framework adopts 04:00 daily as the baseline reference, extracting baseline tension and temperatures to iteratively calibrate initial cable pretension until convergence. Subsequent time steps automatically translate box-girder concrete and ambient temperature variations into thermal loads, executing numerical simulations across the day.\n\nEvaluation of cable ForC2-01 on July 14 and 15, 2026, demonstrated that the model captured the typical daily pattern: decreasing tension in the morning, hitting a minimum around noon, and recovering by evening. Daily estimation discrepancies remained within ±2%, with most intervals within ±1%. While short-term fluctuations from traffic and local disturbances are not fully reproduced, future plans include incorporating output-only system identification and genetic algorithm optimization.

Key takeaways

• A Python GUI integrates cloud monitoring Web APIs with MIDAS Civil NX APIs to automate sensor data retrieval, structural model updates, and numerical simulations.
• The workflow anchors a baseline model at 04:00 to calibrate pretension, calculating subsequent thermal loads based on concrete and ambient temperature deviations.
• Field validation across two consecutive days in July 2026 showed cable tension trends aligned well with measurements, with daily discrepancies within ±2%.
• Short-term fluctuations remain influenced by traffic and local gradients; the team plans to integrate output-only system identification and genetic algorithm optimization.