Fully convolutional networks for structural health monitoring through multivariate time series classification

Abstract

We propose a novel approach to structural health monitoring (SHM), aiming at the automatic identification of damage-sensitive features from data acquired through pervasive sensor systems. Damage detection and localization are formulated as classification problems, and tackled through fully convolutional networks (FCNs). A supervised training of the proposed network architecture is performed on data extracted from numerical simulations of a physics-based model (playing the role of digital twin of the structure to be monitored) accounting for different damage scenarios. By relying on this simplified model of the structure, several load conditions are considered during the training phase of the FCN, whose architecture has been designed to deal with time series of different length. The training of the neural network is done before the monitoring system starts operating, thus enabling a real time damage classification. The numerical performances of the proposed strategy are assessed on a numerical benchmark case consisting of an eight-story shear building subjected to two load types, one of which modeling random vibrations due to low-energy seismicity. Measurement noise has been added to the responses of the structure to mimic the outputs of a real monitoring system. Extremely good classification capacities are shown: among the nine possible alternatives (represented by the healthy state and by a damage at any floor), damage is correctly classified in up to $$95 \%$$ 95 % of cases, thus showing the strong potential of the proposed approach in view of the application to real-life cases.

An Autoencoder-Based Deep Learning Approach for Load Identification in Structural Dynamics

“…A discussion on how to cope with the effects of the mentioned degradation processes within a SHM procedure like the one here proposed, is beyond the scope of this work. Anyhow, it is worth stressing that our procedure will be, in principle, able to address this issue by including in the generative factors the stiffness reduction of the structural members, hence by modelling the effect of damage on the building response [ 24 , 25 , 45 ].…”

“…Convolutional layers allow to detect both local correlations within a TS, and correlations among different TS. This latter aspect is very important for the SHM of civil infrastructures, as it provides a means to implicitly recognize the shape of the vibration modes [ 24 , 25 ]. By stacking convolutional layers, more complex correlation problems in time can be handled, allowing to detect the aforementioned modes of the structure, still keeping the number of the weights in rather limited.…”

Light-Weight CNN Enabled Edge-Based Framework for Machine Health Diagnosis

“…Further work is needed to improve the generalization capability of the model without compromising the requirement of edge implementation. This requires optimization on both algorithmic as well as architectural front, and we aim to explore methods such as fully convolutional networks [43], [44], transfer learning techniques for domain adaptation [45], [46], and layer optimization techniques such as depth separable convolution [47], [48]. Generation of spectrograms as input features from time-series signals consumes a significant percentage of the processing time and thus exploration of alternate features is another area we seek to explore.…”

Application of Bayesian Fe Model Selection for SHM of a Monumental Masonry Palace Through Semi-Supervised Learning

“…The non-destructive nature and the relatively low economic impact of the technology make SHM particularly suitable for the detection of structural damages suffered by monumental structures. Nowadays, the most common procedure in civil SHM concerns the data acquisition and feature extraction (system eigenfrequencies and modal shapes) through signal processing [20], while damage identification problems can be classified as a five-level hierarchical approachs [9]: (i) detection; (ii) localization; (iii) classification ; (iv) assessment; (v) prediction. On the one hand, variations over time (novelty detection) in the dynamic response of an healthy structure can be associated to material's damage and deterioration [26], with a special attention to proper techniques able to remove the environmental effects such as temperature and humidity [14] and the different sources of uncertainties related to SHM can be accounted for by means of Bayesian statistical frameworks [28,1,27].…”