1D in-situ convolution system based on vibration signal for real-time structural health monitoring

Abstract

Structural health monitoring (SHM) of bridges and other building structures has been widely concerned by researchers and engineers. SHM achieved a breakthrough by using convolutional neural networks (CNNs) in deep learning. However, the traditional vision based SHM can not monitor the invisible structural damage. The traditional vibration based SHM needs to convert 1D signals into 2D images, which is easy to lose information in the conversion process, and requires a large scale of hardware circuit. In this paper, a 1D in-situ convolution system (ICS) based on vibration signal is proposed. The SHM is realized by 1D convolution calculation of the structural vibration signal detected by a high-sensitivity triboelectric nanogenerator (TENG) through one in-situ convolution transistor. Based on the natural adaptability of TENG to 1D convolution calculation, and combined with the ability of a single in-situ convolution device to replace more than eight multi conductance devices, ICS can reduce the required SHM computing array area by over 87.5%. Finally, the real-time identification of different positions and different damage levels of the steel bridge was realized by simulating the vibration response of the steel bridge.