An Unsupervised Learning Approach for Early Damage Detection by Time Series Analysis and Deep Neural Network to Deal with Output-Only (Big) Data

Abstract

Dealing with complex engineering problems characterized by Big Data, particularly in structural engineering, has recently received considerable attention due to its high societal importance. Data-driven structural health monitoring (SHM) methods aim at assessing the structural state and detecting any adverse change caused by damage, so as to guarantee structural safety and serviceability. These methods rely on statistical pattern recognition, which provides opportunities to implement a long-term SHM strategy by processing measured vibration data. However, the successful implementation of the data-driven SHM strategies when Big Data are to be processed is still a challenging issue, since the procedures of feature extraction and/or feature classification may end up being time-consuming and complex. To enhance the current damage detection procedures, in this work we propose an unsupervised learning method based on time series analysis, deep learning and the Mahalanobis distance metric for feature extraction, dimensionality reduction and classification. The main novelty of this strategy is the simultaneous dealing with the significant issue of Big Data analytics for damage detection, and distinguishing damage states from the undamaged one in an unsupervised learning manner. Large-scale datasets relevant to a cable-stayed bridge have been handled to validate the effectiveness of the proposed data-driven approach. Results have shown that the approach is highly successful in detecting early damage, even when Big Data are to be processed.

Artificial Intelligence and Structural Health Monitoring of Bridges: A Review of the State-of-the-Art

“…A traditional CNN structure for classification is shown in Fig 5 [155]. The high capability and efficiency of deep learning networks in adapting to various issues and complexities in SHM of bridges, as well as their ability to function properly in learning from large amounts of data, has led to valuable studies in this field in recent years [156][157][158][159][160][161][162][163][164]. For bridge damage detection, Fernandez-Navamuel et al, (2022) developed a supervised deep learning strategy that incorporates Finite Element models to enhance the training phase of a deep neural network.…”

Detection of Partially Structural Collapse Using Long-Term Small Displacement Data from Satellite Images

“…Since the proposed TSL method consists of two ANNs, it is necessary to determine the number of neurons of the seven hidden layers of the DNN and one single layer of the SLNN. Inspired by Entezami et al [42], who developed a hyperparameter selection algorithm for deep autoencoders, another approach is proposed here by using the Akaike…”

“…Since the proposed TSL method consists of two ANNs, it is necessary to determine the number of neurons of the seven hidden layers of the DNN and one single layer of the SLNN. Inspired by Entezami et al [42], who developed a hyperparameter selection algorithm for deep autoencoders, another approach is proposed here by using the Akaike information criterion (AIC) under sample neuron sizes depicted in Figure 5. Accordingly, it only suffices to run the TSL algorithm and compute the sum of residual samples.…”

Health Monitoring of Large-Scale Civil Structures: An Approach Based on Data Partitioning and Classical Multidimensional Scaling

“…The AANN, thus, represents a smart algorithm for filtering out noise, outliers, and any type of variations in the data due to environmental and/or operational variability [ 43 ]. As far as the network hyperparameters are concerned, the number of neurons in each hidden layer was set according to the approach described in [ 44 ] and based on the final prediction error. It has turned out that the said number of neurons for the mapping, bottleneck and de-mapping layers has to be, respectively, set to 22, 3, and 22.…”