Structural damage identification based on fast S-transform and convolutional neural networks

Abstract

Measured signals from installed sensors on structures cannot solely provide beneficial information about the presence of damage. Many signal processing methods have been developed to extract useful data from raw signals. However, each method has its advantages and disadvantages. In the present study, the fast S-transform signal processing method has been used for the damage detection of structures. Fast S-transform outperforms traditional S-Transform since it filters out unusable data using a scaling method. The method’s performance in obtaining correct natural frequencies of a structure has been evaluated using a 2 Degrees of Freedom (DOF) mass-spring dynamic system. Additionally, the ability of the method to detect damages in structural systems has been demonstrated using a 6 DOF structure subjected to Northridge earthquake record. The results of the mentioned numerical studies show that the fast S-transform can extract the accurate resonance frequencies of a structure, and also can detect the presence and the time of damage occurrence. In the experimental study, a 3 story plexi frame has been excited by a chirp sinusoidal signal. Furthermore, four different damage scenarios have been defined in the experimental study. To classify the four damage types, Convolutional Neural Network (CNN) has been used. The results of the experimental study show that fast S-transform can detect damages in real-time monitoring of structures, and the classification using CNN can identify the severity of damage.