Damage detection in steel beams using natural frequency analysis and machine learning techniques

Recently, machine learning (ML) algorithms have shown strong potential in predicting structural damage. However, structural health monitoring data mostly represents normal operating conditions or minor deviations, rarely capturing severe or hazardous states. This makes it challenging to create realistic datasets for training ML models in structural damage detection. If such data were available, it would likely involve variables like the stress intensity factor range and stress ratio, which are challenging to measure directly in real-world structures. In this study, ML models, including Artificial Neural Network (ANN), Extreme Gradient Boosting (XGB), and Random Forest (RF), were developed to predict the locations, widths, and depths of saw-cuts in steel beams, using changes in natural frequencies as predictive indicators. The natural frequencies under various damage conditions were calculated using the Finite Element Method (FEM). To validate the results, the natural frequencies of both damaged and undamaged beams were calculated using the Rational Fraction Polynomial (RFP) method and compared with the outcomes from the Finite Element Method (FEM) to evaluate their consistency. The study seeks to establish conclusions regarding the most effective machine learning models and the integration of FEM, RFP, and machine learning for enhanced structural health monitoring.