Research on steel structure damage detection based on TCD-CNN method

Abstract

With the development of social economy, steel structures are widely adopted in building structures. However, the structure is inevitably damaged during service due to environmental erosion, load action, and other reasons. Therefore, it is significant to detect the structural damage. This study propose a damage detection method of steel structure based on TCD-CNN. The method uses transmissibility change data(TCD) as the damage indicator and the convolutional neural network(CNN) model for damage detection of the structure. The benefit of this method is no need to measure the load response of the structure, which is more conducive to practical engineering applications. In this paper, the vibration experiments of the cantilever beam steel structure are performed, and three damage indicators, namely, time domain data(TDD), frequency domain data(FDD) and TCD, are proposed. The comparison of detection results of these three damage indicators shows that the use of TCD has higher detection accuracy and more stable detection results. Furthermore, 1D-TCD-CNN and 2D-TCD-CNN are used to detect the damage degree of structure respectively. The results illustrate that test accuracy with 2D-TCD-CNN is higher, even up to 100%, and with a shorter running time.