Physics-Based AI-Driven Surrogate Modeling for Structural Displacement Prediction in Mechanical Systems With Limited Sensor Data

This study introduces a machine learning (ML)-based surrogate model for finite element analysis, designed to predict structural strain distributions using a minimal number of strategically placed virtual sensors. The proposed approach eliminates the dependency on external force measurements, leveraging local strain measurements to infer global strain responses across a two-dimensional truss structure. Various regression algorithms, including decision trees and deep neural networks (DNNs), were evaluated, with DNNs achieving superior accuracy ( $\text {R}^{2}$ > 0.996, MAE < 4%, RMSE < 5.5%). The methodology significantly reduces sensor requirements and computational overhead, offering a practical, scalable solution for structural health monitoring (SHM) in complex mechanical systems. The results underscore the potential of ML-based surrogate models to enhance the efficiency and accuracy of continuous monitoring and dynamic analysis in large-scale infrastructure applications, setting the stage for future advancements in sensor optimization and three-dimensional system extensions.