Analysis of Aeolian vibration failure mechanisms in transmission line fittings using machine learning algorithms

The integration of advanced machine learning algorithms in structural engineering has become a transformative approach for addressing complex failure mechanisms in critical infrastructure. This study leverages machine learning to develop a hybrid predictive framework for analyzing failure mechanisms in transmission line fittings subjected to aeolian vibration. The proposed framework combines physics-informed neural networks (PINNs), which embed domain-specific physical principles for theoretical consistency and interpretability, with gradient boosting machines (GBMs), which excel at identifying nonlinear interactions in high-dimensional datasets. These machine learning algorithms work in tandem to predict critical failure points and material-specific fatigue behaviors with unprecedented accuracy. A comprehensive dataset, including vibration frequencies, stress amplitudes, and environmental conditions, was collected through laboratory and field experiments to train and validate the hybrid machine learning model. The framework achieved superior performance metrics, with a mean squared error (MSE) of 0.008 and a prediction accuracy of 96.5%, significantly outperforming traditional empirical methods and finite element analysis (FEA). Key machine learning-driven insights include the accurate identification of critical vibration frequencies (10 Hz to 30 Hz) and stress accumulation patterns, which align closely with experimental observations. The use of SHapley Additive exPlanations (SHAP) further enhances the interpretability of the machine learning model, providing actionable insights into the contributions of vibration frequency, stress amplitude, and environmental factors to failure risks.