SVM-assisted damage identification in cantilever steel beam using vibration-based method

Structures naturally degrade over time, with the degree of deterioration influenced by factors such as location and construction quality. Additionally, unexpected loads can exacerbate this process, leading to varying degrees of failure. This study aims to improve the early detection of structural damage by integrating vibration-based methods with machine learning techniques, offering greater resilience compared to current approaches. Frequency-based analysis has its limitations, particularly in detecting damage farther from boundaries. To address this gap the study examines twenty-one cantilever beams with differing damage intensities and locations, alongside an undamaged beam, employing both experimental and numerical analyses. Frequency analysis conducted has shown that damage severity and location will affect the natural frequencies, especially revealing that while damage typically decreases fundamental frequency when it is nearer to the boundary and slowly frequency increases, while the damage moves farther from the boundary and for all the scenarios there exists one location where damage frequency is almost equal to undamaged. So, frequency analysis alone may sometimes fail to accurately detect damage. To overcome this limitation support vector machine algorithms are employed for the data obtained from the experimental analysis and found to achieve an accuracy of 85% in predicting the presence of damage, irrespective of its severity.