Automated Modelling and Assessment of the Impact of Damage Applied on Suspension Bridges' Main Cables Using Natural Frequency Analysis

Abstract

This thesis describes the analysis of the effect of damage application to the main cables of the Lower Liard River Bridge (LLRB). A program in Python is created to generate custom suspension bridge finite element models in Abaqus CAE based on generalized parameters. While this study focuses strictly on the LLRB, the Python program is flexible to adapt to most suspension bridge designs. From there, over 2,000 test cases are defined with varying damage distributions along the main cables of the model. Damage is interpreted as corrosion based on inspection reports, and distributions are created to observe the effect of these variations on a few critical properties of the bridge. Key outputs studied are shifts in the natural frequencies of the modes of vibration of the deck and the vertical deck displacement under gravity. The Python program was enhanced to automate test execution and analysis. As part of this process, a rudimentary mode identification algorithm calibrated to the LLRB to recognize vibration modes in the deck from the outputs of Abaqus is created. Among these tested cases are a few simple wind response analyses of the bridge under dynamic wind loading. For this, an algorithm is defined to generate high-frequency wind velocity signals that are required to obtain a time history response of the bridge. The results obtained in this research confirm that, as demonstrated in previous studies, there are observable changes in the natural frequencies of the bridge when damage is applied. This research reveals how damage location and intensity affect bridge behaviour, offering insights for future developments. Key findings include the identification of vertical vibration modes of the deck as the clearest indicator of the applied damage parameters. In contrast, only a few torsional modes reveal any correlation. Specific behaviours are observed when damage is implemented at particular locations along the length of the main cables, such as near the anchor, towers, and midspan. Notably, a strong correlation is established between the shifts in natural frequencies of individual cases with single damage locations and the complementary case combining two damage locations in the same case. Findings suggest the potential to parameterize a bridge's behaviour based on single damage case results to estimate the damage applied from the measured dynamic properties. Further work should explore more damage patterns and integrate machine learning to advance this method.