Structural Vibration Data Anomaly Detection Based on Multiple Feature Information Using CNN-LSTM Model

Structural health monitoring (SHM) system has been operating for a long time in harsh environment, resulting various abnormalities the collected structural vibration data. Detecting these abnormal data not only requires user interaction but also is quite time-consuming. Inspired by manual recognition process, anomaly detection method based on combined model of convolutional neural network (CNN) and short-term memory (LSTM) proposed this paper. This simulates intelligent human decision making two steps. First, original are reconstructed feature sequences with higher universality smaller size. In domain, residual signal extracted from upper lower peak envelopes to characterize symmetry frequency power spectral density sequence interpretability Second, CNN-LSTM constructed trained which utilizes CNN extract local high-level features input inputs new continuous representations into LSTM learn global long-term dependencies features. For verification, was applied automatic classification 42 days long-span bridge, average accuracy results exceeded 94% 78 minutes. Compared existing methods, can detect more accurately efficiently stronger generalization ability.