Deep learning for data anomaly detection and data compression of a long-span suspension bridge

As intelligent sensing and sensor network systems have made progress and low-cost online structural health monitoring has become possible and widely implemented, large quantities of highly heterogeneous data can be acquired during the monitoring. This has resulted in exceeding the capacity of traditional data analytics techniques, especially in monitoring large-scale or critical civil structures. In particular, data storage has become a big challenge, hence, resulting in the emergence of data compression and reconstruction as a new area in structural health monitoring (SHM) of large infrastructure systems. SHM data generally include anomalies that can disturb structural analysis and assessment. The fundamental reasons for the abnormality of data are extremely complex. Therefore, reconstruction of the abnormal data is generally difficult and poses serious challenges to achieve high-accuracy after data has been compressed. Considering these significant challenges, in this paper, a novel deep-learning-enabled data compression and reconstruction framework is proposed that can be divided into two phases: (a) a one-dimensional Convolutional Neural Network (CNN) that extracts features directly from the input signals is designed to detect abnormal data with validated high accuracy; (b) a new SHM data compression and reconstruction method based on Autoencoder structure is further developed, which can recover the data with high-accuracy under such a low compression ratio. To validate the proposed approach, acceleration data from the SHM system of a long-span bridge in China are employed. In the abnormal data detection phase, the results show that the proposed method can detect anomaly with high accuracy. Subsequently, smaller reconstruction errors can be achieved even by using only 10% compression ratio for the normal data.