Bayesian Prediction of Pre-Stressed Concrete Bridge Deflection Using Finite Element Analysis

Area of Science:

• Civil Engineering
• Structural Engineering
• Computational Mechanics

Background:

• Vertical deflection is a critical safety indicator for railway bridges.
• Existing physics-based models for predicting time-dependent deflection face limitations due to uncertainties in material properties, creep, and temperature.
• Accurate prediction is challenging for bridges with insufficient data, such as during the construction stage.

Purpose of the Study:

• To develop a novel Bayesian method for accurate, time-dependent deflection prediction in railway bridges.
• To address limitations of physics-based models and data scarcity by integrating finite element analysis (FEA) and actual measurement data.
• To create a customizable and updatable probabilistic prediction model for specific bridges.

Main Methods:

• A Bayesian approach is proposed, integrating finite element model (FEM) results with actual measurement data.
• Gaussian process regression (GPR) is modified to incorporate both FEA outcomes and measured data, overcoming FEM imperfections and data shortages.
• The probabilistic prediction model is designed for continuous updating with new measurement data.

Main Results:

• Probabilistic prediction models were successfully derived for a pre-stressed concrete railway bridge in South Korea.
• The developed models demonstrated agreement with actual measurements, accurately capturing large downward deflections during the construction phase.
• The study confirmed the efficacy of the proposed method even with limited initial measurement data.

Conclusions:

• The proposed Bayesian method, utilizing FEA and GPR, provides a robust approach for predicting time-dependent bridge deflection.
• The developed probabilistic models are effective for specific bridges and can be updated as more data becomes available.
• This method enhances railway bridge safety management by offering more reliable deflection predictions.