A Damage Recognition Method for Civil Engineering Structures Based on MTL-1DCNN

Abstract

Within the discipline of civil engineering, real-time and accurate identification of building structure damage is an important means to ensure the service life and safety of civil engineering. To reduce the occurrence of civil engineering accidents and enhance the health, usability and integrity of building structures, this research uses deep learning to build detection models. The model mainly combines a onedimensional hollow CNN with a multi-task learning method to construct a multi-task learning onedimensional convolution network (MTL-1DCNN) model and use it for automatic feature extraction of structural damage signals. The model has a good performance in the position information judgment and damage degree diagnosis of structural damage. In the experiment of the two-story building model, the research proposed that the MTL-1DCNN model correctly identified 2031 locations out of 2048 samples, with a recognition accuracy of up to 99%, and the MSE of the calculation result was 3.47×10−5, demonstrating that the suggested strategy could mine effective and reliable damage information from the structural vibration response signal, which was of significant importance and engineering value to the good development of damage identification technology and real-time intelligent monitoring of the health status of engineering structures.