Enhanced structural-load forecasting: Fusion of image analysis and time series with physics-driven deep-learning models

Abstract

This study focuses on dynamic-load identification in structural-health monitoring, which is crucial for the safety and longevity of ocean structures. It introduces an innovative data-processing method that applies sliding-window techniques to transform time-series data into Gramian angular field (GAF) images and a two-dimensional series. This novel approach exploits the image-feature sensitivity of deep-learning networks and fuses three-dimensional images with two-dimensional time series to enhance the input-feature representation. Furthermore, the study tailors the loss function of the algorithm using the stiffness, mass, and damping matrices of specific structures, effectively combining mathematical and physical models to improve the structural load-identification accuracy. The proposed Physical Information-based Gramian angular field (PI-GAF) method is validated through numerical simulations and practical experiments, and the results demonstrate its superiority and generalizability in the field of ocean engineering, marking a significant advancement in the domain of structural-health monitoring.