Tendon Stress Estimation from Strain Data of a Bridge Girder Using Machine Learning-Based Surrogate Model

Area of Science:

• Civil Engineering
• Structural Engineering
• Materials Science

Background:

• Prestressed concrete girders enable long spans and reduced cracking but require precise tensioning force control.
• Accurate design and monitoring of tendon force are crucial to prevent excessive creep and ensure structural integrity.
• Estimating prestressing tendon stress is challenging due to limited accessibility.

Purpose of the Study:

• To develop and validate a strain-based machine learning method for real-time estimation of applied tendon stress in prestressed girders.
• To assess the accuracy and feasibility of machine learning models in predicting tendon force.

Main Methods:

• A dataset was generated using finite element method (FEM) analysis on a 45 m girder with varied tendon stress.
• Machine learning network models were trained and tested using strain data to predict tendon stress.
• The model with the lowest Root Mean Square Error (RMSE) was selected for stress prediction.

Main Results:

• The developed machine learning models achieved prediction errors of less than 10% for tendon stress.
• The chosen model accurately estimated real-time tendon stress, enabling tensioning force adjustments.
• The study identified optimal girder locations and strain sensor configurations.

Conclusions:

• Machine learning, utilizing strain data, is a feasible method for instant tendon force estimation in prestressed girders.
• This approach offers a practical solution for real-time monitoring and control of prestressing forces.
• The findings contribute to improved design and maintenance strategies for prestressed concrete structures.