An LSTM-based detection model for breathing cracks with multiple damage positions and degrees of beam-like bridges under vehicle loads

The cracks of bridges are not always completely open during vibrations but undergo an open and closed cycle mode under moving vehicles. Therefore, this study proposes a long short-term memory (LSTM)-based detection model for breathing cracks with multiple damage positions and degrees of beam-like bridges for the first time. The contact point displacement variation (CPDV), serving as an evaluation indicator of the bridge damage, is calculated by analyzing the acceleration history data of the vehicle that has passed over a randomly damaged bridge. At the same time, CPDV is also used as the input variable of an LSTM neural network to monitor the damage location and degree of the bridge. By employing the finite element simulation of the bridge half-vehicle model, a dataset was created for the training and prediction of LSTM. The numerical results show that the proposed method can accurately identify complex breathing crack information at different vehicle speeds, and it is also robust to road roughness. In addition, a laboratory experiment was conducted to verify the proposed method. The experimental results show that CPDV is sensitive to bridge cracks and LSTM can realize higher damage prediction.