Analyzing Multiple Modal Interactions Induced by Internal Resonances in Cable-Stayed Structures Using Machine Learning Techniques

Nonlinear oscillations induced by modal interactions are commonly observed in cable-stayed bridges, which are complex to investigate through numerical simulations. Given this challenge, this paper conducted a series of laboratory experiments and investigated the mechanism of multiple modal interactions (MMIs) using machine learning techniques. To account for the global modal shape effects, an improved modal effective factor (MEF) was proposed to label the modal features within the experimental data set. Subsequently, a predictive model for cable responses in multiple internal resonances (MIRs) was developed using the Random Forest algorithm and validated against the experimental testing set. Analysis of the predictive outputs identified MEF-composed modal features as key contributors to MIR responses, demonstrating a strong correlation with the weight of resonance energy transfer. Through the interaction analysis among modal features, the complex mechanism of MMI was revealed from the perspective of resonance energy transfer. Furthermore, the global sensitivity analysis, derived from modal feature interactions, is expected to enhance risk assessment and inform safety strategies for managing nonlinear modal-coupling oscillations in cable-stayed bridges.