Physics-Informed Data-Driven Prediction of 2D Normal Strain Field in Concrete Structures.

Area of Science:

• Civil Engineering
• Materials Science
• Computational Mechanics

Background:

• Concrete structures exhibit complex time-dependent behaviors like creep and shrinkage, challenging accurate long-term prediction.
• Existing empirical and numerical models often rely on limited laboratory data and struggle with site-specific conditions and differential effects.
• Current data-driven methods using structural health monitoring show promise but lack sensor integration and geometric considerations.

Purpose of the Study:

• To develop a novel physics-informed data-driven approach for predicting the 2D normal strain field in prestressed concrete structures.
• To overcome limitations of existing models by integrating structural geometry, loading history, and multiple sensor data.
• To accurately capture differential rheological effects in long-term concrete behavior.

Main Methods:

• A physics-informed data-driven framework combining a simplified analytical structural model and a data-driven temperature field model.
• Embedding neural networks within rheological time-functions to model time-dependent material behavior.
• Training the model simultaneously on data from multiple sensors to estimate strain evolution across the structure.

Main Results:

• The proposed method accurately predicts the long-term 2D normal strain field in prestressed concrete structures.
• The approach effectively captures differential rheological effects by leveraging multiple sensor inputs and structural geometry.
• Strain evolution can be estimated at any point of interest within the structure's longitudinal section.

Conclusions:

• The physics-informed data-driven approach offers a significant advancement in predicting long-term concrete behavior, outperforming traditional methods.
• This method provides a more accurate and holistic understanding of time-dependent strains in prestressed concrete structures.
• The integrated approach enables more reliable structural health monitoring and performance assessment over time.