Multi-level damage detection using a combination of deep neural networks

Abstract

In recent years, bridge damage identification using a convolutional neural network (CNN) has become a hot research topic and received much attention in the field of civil engineering. Although CNN is capable of categorizing damaged and undamaged states from the measured data, the level of accuracy for damage diagnosis is still insufficient due to the tendency of CNN to ignore the temporal dependency between data points. To address this problem, this paper introduces a novel hybrid damage detection method based on the combination of CNN and Long Short-Term Memory (LSTM) to classify and quantify different levels of damage in the bridge structure. In this method, the CNN model will be used to extract the spatial damage features, which will be combined with the temporal features obtained from Long Short-Term Memory (LSTM) model to create the enhanced damage features. The combination successfully strengthened the damage detection capability of the neural network. Moreover, deep learning is also improved in this paper to process the acceleration-time data, which has a different amplitude at short intervals and the same amplitude at long intervals. The empirical result on the Vang bridge shows that our hybrid CNN-LSTM can detect structural damage with a high level of accuracy.

Keywords

• structural health monitoring
• damage detection
• time series analysis
• deep neural networks