Research on Damage Identification for Steel Frames Based on Convolutional Autoencoder and Correlation Function

Aiming at the problem that the damage recognition accuracy of the deep learning model decreases under the influence of data sample imbalance and noise, a structural damage recognition method based on the combination of convolutional autoencoder and correlation function was proposed. Taking the stand scale test of Qatar University as an example, the sample data of healthy working conditions were input into the convolutional autoencoder model for learning, and the constructed convolutional autoencoder model was used to reconstruct the data samples of the healthy structural data samples, and the maximum error of data reconstruction was used as a threshold to judge whether the structure was damaged. Then, Gaussian white noise with different signal-to-noise ratios was added to the data set containing healthy working condition samples and damaged working condition samples, and the data samples added with noise were preprocessed by autocorrelation function. The data set processed by the autocorrelation function was input into the model for training and prediction respectively, and the prediction results of the model were compared and analyzed. The results showed that the method could accurately identify structural damage when the data samples without noise were not balanced, and the identification accuracy could reach 100%. However, under the influence of adding noise, the original characteristics of the data could be highlighted after being processed by the autocorrelation function. Under the noise condition with a signal-to-noise ratio of 5, the recognition accuracy could still reach 100%, which proved that the method could effectively improve the accuracy of damage recognition of the autoencoder model and had better noise resistance.