A review of ML for bridge vertical displacement

The vertical displacement of bridges is a crucial metric for assessing structural safety and operational performance. Accurate prediction of such displacement is indispensable for effective bridge management and timely risk warning. Nevertheless, traditional finite element methods (FEM) encounter significant limitations in practical applications, primarily due to intricate modeling processes and limited generalization capacity. With the accumulation of monitoring data and recent progress in artificial intelligence, machine learning (ML) approaches have demonstrated substantial potential in displacement prediction. This review synthesizes commonly adopted ML techniques for both non-temporal and temporal modeling, and provides a comparative analysis of their performance and applicability. It further classifies existing studies according to the types of input data—such as strain, temperature, traffic load, historical displacement, and multi-factor fusion—thereby elucidating how data characteristics shape model development. In addition, the review highlights emerging research directions, including explainable AI for bridge prediction models, spatial modeling of monitoring topology, and physics-data-driven modeling approaches. Collectively, these explorations establish a theoretical foundation and research agenda for developing more accurate and reliable predictive models.