Data-Driven Surrogate Modeling for Efficient Dynamic Response Prediction of Modular Buildings under Uncertainty

Modular building systems are increasingly adopted due to their construction efficiency and scalability; however, their dynamic response is strongly influenced by uncertainties in material properties, connection behavior, and operational loading. Accurate prediction of such responses typically requires computationally intensive stochastic simulations, particularly when Monte Carlo-based approaches are employed, leading to significant challenges in large-scale or real-time applications. This study proposes a data-driven surrogate modeling framework for efficient prediction of the dynamic response of modular buildings under uncertainty. The approach integrates high-fidelity numerical simulations with machine learning-based surrogate models to approximate the structural response across a wide range of uncertain parameters. The surrogate is trained using a limited set of simulations and subsequently employed to perform rapid uncertainty propagation, significantly reducing the computational burden associated with conventional double-loop stochastic analysis. A representative modular building system is considered to demonstrate the effectiveness of the proposed framework. Key response quantities, including peak displacements and vibration characteristics, are evaluated under stochastic loading scenarios. The results indicate that the surrogate model achieves high accuracy in capturing the probabilistic response while reducing computational cost by orders of magnitude compared to direct simulations. The proposed methodology provides an efficient and scalable tool for uncertainty quantification and performance assessment of modular structural systems, with potential applications in reliability analysis, design optimization, and real-time structural health monitoring.