High-Fidelity Numerical Modeling for the Mechanical Characterization of a Full-Scale Test Bridge

This paper presents a high-fidelity, physics-based numerical model of a full-scale test bridge at the University of the Bundeswehr Munich, which was refined using Bayesian updating on experimental data to create a functional digital twin for accurately assessing structural performance under various environmental and damage conditions.

The Gist

The Big Picture: Building a "Digital Twin"

Imagine you have a real, physical bridge. Now, imagine you build an exact, invisible copy of that bridge inside a super-powerful computer. This computer copy isn't just a static picture; it's a "Digital Twin." It acts like a "digital shadow" that mimics the real bridge's behavior perfectly.

The goal of this research is to make that digital shadow so accurate that engineers can use it to figure out if the real bridge is sick or injured, even before they can see the damage with their eyes.

The Problem: Why Do We Need This?

Bridges are like the arteries of a city; if they fail, traffic and emergency services stop. But bridges are getting old, traffic is getting heavier, and extreme weather (like floods) is becoming more common.

One specific danger is scour. Imagine a river flowing fast and washing away the dirt under a bridge's legs (foundations). This is like pulling the rug out from under a table; the table (bridge) might collapse.

The problem for engineers is that they don't have enough data on what a bridge looks like when it is slightly damaged but not yet collapsed. It's hard to study this in the real world because you can't just break a real bridge to see what happens.

The Solution: The Test Bridge and the Experiment

To solve this, the researchers built a full-scale test bridge at the University of the Bundeswehr Munich.

• The Real Bridge: It's a 30-meter-long steel and concrete bridge. It has a special feature: its middle supports can be moved up or down slightly to simulate "foundation settlement" (like the bridge legs sinking into the mud).
• The Experiment: They monitored this bridge for almost a month. They put sensors on it to measure how much weight (force) the middle legs were carrying and how the temperature changed.

The Temperature Twist:
The researchers noticed something tricky. The bridge's legs carried different amounts of weight depending on the time of day. Why? Because the sun hits the south side of the bridge but leaves the north side in the shade. This thermal expansion (metal getting hot and expanding) made the bridge push harder on one leg than the other. It was like a thermometer disguised as a bridge.

The Method: Cleaning the Data and Updating the Model

The researchers had two main tools:

1. The Real Data: The messy, real-world numbers from the sensors (full of temperature noise).
2. The High-Fidelity Computer Model: A detailed 3D simulation of the bridge that knows exactly how steel and concrete behave.

Step 1: Removing the "Weather Noise"
First, they had to clean the data. They used a mathematical trick (like a noise-canceling headphone for data) to subtract the effect of the sun and temperature. This left them with the "pure" mechanical forces acting on the bridge.

Step 2: The "Sherlock Holmes" Update (Bayesian Updating)
Now, they compared the cleaned real-world data with their computer model.

• The Mystery: The real bridge was pushing down on its legs slightly differently than the computer model predicted.
• The Deduction: The researchers asked, "What tiny change in the bridge's setup would cause this difference?"
• The Answer: Using a statistical method called Bayesian updating, they solved an "inverse problem." They worked backward from the force measurements to guess the physical state of the bridge.

The Discovery:
The math revealed a hidden secret about the bridge's construction. When the bridge was built, the southern leg was accidentally propped up about 0.7 millimeters higher than it should have been (a "parasitic counter-mount"), while the northern leg sank slightly. It was a tiny error, invisible to the naked eye, but the digital twin and the math found it.

The Result: A Dynamic Shadow

By feeding the real-world data into the computer model, the "Digital Twin" was updated. It is no longer just a theoretical design; it is now a dynamic, data-driven shadow of the physical bridge.

• Before: The model was a guess based on blueprints.
• After: The model knows the exact, real-world quirks of that specific bridge, including its tiny, accidental unevenness.

Conclusion

This study proves that by combining a super-detailed computer model with real-world sensor data, we can create a "Digital Twin" that helps us understand the true health of a bridge. It allows engineers to detect tiny problems (like a leg sinking a fraction of a millimeter) that would otherwise go unnoticed, helping to keep bridges safe for longer.

What's Next?
The authors say this is just the beginning. In the future, they want to:

• Use even smarter math to handle complex weather effects.
• Test the system with bigger, more dramatic damage scenarios (not just tiny settlements).
• Eventually, use this system to monitor real bridges in the wild, not just test bridges in a lab.

Technical Summary

Technical Summary: High-Fidelity Numerical Modeling for the Mechanical Characterization of a Full-Scale Test Bridge

Problem Statement
The structural health monitoring (SHM) of bridges is increasingly critical due to aging infrastructure, rising traffic loads, and the intensifying impact of climate change, particularly regarding flood-induced scour and foundation settlement. A primary bottleneck in developing robust SHM strategies is the scarcity of high-quality experimental data featuring well-characterized damage states under realistic environmental conditions. While laboratory experiments exist, there is a need for intermediate steps bridging the gap between controlled lab settings and complex real-world monitoring. Specifically, understanding the mechanical response of bridge substructures to differential settlements requires data that accounts for confounding environmental factors, such as temperature-induced thermal deformations, which often mask true structural signals.

Methodology
This study presents a framework that integrates high-fidelity experimental data from a full-scale test bridge with a detailed finite element (FE) model to characterize structural behavior and quantify foundation settlements.

1. Experimental Setup and Data Acquisition:
   The research utilizes a 30-meter steel-concrete composite test bridge located at the University of the Bundeswehr Munich. The structure features two main HEB1000 steel girders and a segmented concrete deck. A long-term monitoring campaign (spanning nearly three weeks) was conducted using a mobile measurement system. The study focuses on reaction forces ($F_1$ and $F_2$) measured by load cells under the southern and northern central supports, alongside ambient air temperature data. The data reveals significant temperature-dependent variations, with the southern support (exposed to direct sunlight) carrying higher loads than the northern (shaded) support.

2. High-Fidelity Finite Element Modeling:
   A physics-based FE model was developed in PyAnsys to serve as a "digital shadow" of the physical structure. The model employs shell elements for primary components (girders, stiffeners, deck slabs, and braces) and distributed springs to model the interaction between segmented concrete slabs. While bolted connections are simplified using beam elements for computational efficiency, the model captures the essential composite action and load transfer. The model is linked to experimental data via a script that maps sensor readings to specific FE nodes.

3. Temperature Compensation:
   To isolate mechanical responses from thermal effects, a temperature compensation strategy was applied using penalized B-spline regression. This approach models the non-linear relationship between temperature and reaction forces ($y = g(t) + \epsilon$). A smoothing parameter ($\lambda$) was optimized to prevent over-fitting to asynchronous dynamics while capturing the global thermal trend. The regression function was subtracted from raw data to isolate the structural response attributable to mechanical loading.

4. Bayesian Model Updating:
   The core of the mechanical characterization involves solving a linear inverse problem to estimate foundation settlements. A surrogate model ($u = Ax + u_0$) was derived from the FE model to map foundation settlements ($x$) to reaction forces ($u$). Using Bayesian inference, the study updates the prior distribution of settlement values based on the temperature-compensated experimental data. The process assumes Gaussian priors and measurement errors, allowing for a closed-form solution to update the posterior distribution of the settlements without requiring computationally expensive Monte Carlo methods.

Key Contributions

• Integration of Data and Physics: The paper demonstrates a workflow that combines long-term ambient monitoring data with a high-fidelity FE model to refine the mechanical characterization of a full-scale structure.
• Temperature Mitigation: The application of penalized B-spline regression effectively removes thermal noise from reaction force data, revealing underlying structural behaviors that would otherwise be obscured.
• Quantification of Parasitic Settlements: Through Bayesian updating, the study successfully quantifies the uncertainty and magnitude of foundation settlements, identifying specific deviations from the ideal design state.
• Digital Shadow Concept: The work establishes a unidirectional data stream from the physical bridge to the numerical model, creating a dynamic, data-driven "digital shadow" that evolves with incoming measurements.

Results
The analysis of the temperature-compensated data and subsequent Bayesian updating yielded the following findings:

• Thermal Impact: The southern support exhibited a marked sensitivity to solar irradiance compared to the northern support, confirming the necessity of environmental compensation.
• Settlement Characterization: The model calibration revealed that during the placement of the central supports, a parasitic positive counter-mount of approximately 0.7 mm was assigned to the southern support, while a negative settlement of similar magnitude occurred at the northern temporary bearing.
• Statistical Correlation: Prior to data assimilation, the settlement parameters were assumed independent. The posterior distribution, however, revealed a statistical link (correlation) between the two support settlements, resulting in a bivariate normal distribution not aligned with the input parameter axes.
• Model Validation: The posterior mean of the settlement vector successfully intercepted the common locus of the model response and experimental readings, validating the inverse problem solution.

Significance and Outlook
The paper posits that this integrated approach provides a foundational framework for advancing SHM in real-world bridge networks. By treating the numerical model as a dynamic digital shadow, the study supports informed maintenance decisions and enhances the accuracy of condition assessments. The authors emphasize that this work represents a crucial intermediate step toward a full digital twin (DT), laying the groundwork for future bidirectional interaction between physical and virtual assets.

The study modestly outlines future directions, including:

• Testing more advanced algorithms, such as dynamic time warping, to better handle environmental effects.
• Expanding the dataset to include artificial, large-scale settlement scenarios that induce non-linear structural responses, potentially requiring generalized polynomial chaos expansion (gPCE).
• Integrating spaceborne monitoring capabilities and vibration-based scour detection methods.
• Extending the framework from individual bridges to a population-based perspective to improve critical infrastructure protection and resource prioritization.