Practical approach for data-efficient metamodeling and real-time modeling of monopiles using physics-informed multifidelity data fusion

This paper proposes a practical approach for data-efficient metamodeling and real-time modeling of laterally loaded monopiles using physics-informed multifidelity data fusion. The proposed approach fuses information from one-dimensional (1D) beam-column model analysis, three-dimensional (3D) finite element analysis, and field measurements (in order of increasing fidelity) for enhanced accuracy. It uses an interpretable scale factor–based data fusion architecture within a deep learning framework and incorporates physics-based constraints for robust predictions with limited data. The proposed approach is demonstrated for modeling monopile lateral load–displacement behavior using data from a real-world case study. Results show that the approach provides significantly more accurate predictions compared to a single-fidelity metamodel and a widely used multifidelity data fusion model. The model’s interpretability and data efficiency make it suitable for practical applications.