Structural Health Monitoring using raw dynamic measurements is the subject of several studies aimed at identifying structural modifications or, more specifically, focused on damage assessment. Traditional damage detection methods associate structural modal deviations to damage. Nevertheless, the process used to determine modal characteristics can influence the results of such methods, which could lead to additional uncertainties. Thus, techniques combining machine learning and statistical analysis applied directly to raw measurements are being discussed in recent researches. The purpose of this paper is to investigate statistical indicators, little explored in damage identification methods, to characterize acceleration measurements directly in the time domain. Hence, the present work compares two machine learning algorithms to identify structural changes using statistics obtained from raw dynamic data. The algorithms are based on Artificial Neural Networks and Support Vector Machines. They are initially evaluated through numerical simulations using a simply supported beam model. Then, they are assessed through experimental tests performed on a laboratory beam structure and an actual railway bridge, in France. For all cases, different damage scenarios were considered. The obtained results encourage the development of computational tools using statistical indicators of acceleration measurements for structural alteration assessment.

Structural health monitoring (SHM) ensures the safety and reliability of civil infrastructure. Autoencoders, as unsupervised learning models, offer promise for SHM by learning data features and reducing dimensionality. However, comprehensive studies comparing autoencoder models in SHM are scarce. This study investigates the effectiveness of four autoencoder-based methodologies, combined with Hotelling’s $T^2$ statistical tool, to detect and quantify structural changes in three civil engineering structures. The methodologies are evaluated based on computational costs and their abilities to identify structural anomalies accurately. Signals from the structures, collected by accelerometers, feed the autoencoders for unsupervised classification. The latent layer values of the autoencoders are used as parameters in Hotelling’s $T^2$, and results are compared between classes to assess structural changes. Average execution times of each model were calculated for computational efficiency. Despite variations, computational cost did not hinder any methodology. The study demonstrates that the best fitting model, VAE-$T^2$, outperforms its counterparts in identifying and quantifying structural changes. While the AE, SAE, and CAE models showed limitations in quantifying changes, they remain relevant for detecting anomalies. Continuous application and development of these techniques contribute to SHM advancements, enabling the increased safety, cost-effectiveness, and long-term durability of civil engineering structures.

The present work evaluates the deep learning algorithm called Sparse Auto-Encoder (SAE) when applied to the characterization of structural anomalies. This study explores the SAE’s performance in a supervised damage detection approach to consolidate its application in the Structural Health Monitoring (SHM) field, especially when dealing with real-case structures. The main idea is to use the SAE to extract relevant features from the monitored signals and the well-known Support Vector Machine (SVM) to classify such characteristics within the context of an SHM problem. Vibration data from a numerical beam model and a highway viaduct in Brazil are considered to assess the proposed approach. In both analyzed examples, the efficiency of the implemented methodology achieved more than 99% of correct damage structural classifications, supporting the conclusion that SAE can extract relevant characteristics from dynamic signals that are useful for SHM applications.

This work investigates the effectiveness of using convolutional neural networks (CNNs) and instantaneous displacement measurements for damage identification in beams. The study involves subjecting laboratory beams to eight distinct damage scenarios and capturing the vertical positions of 60 points along the beam length during free-vibration tests using a high-speed camera. The data obtained was subsequently used to train a CNN in a supervised manner to estimate the level of damage at each point. Results showed that the CNN models were able to correctly localize and quantify the damage levels when trained on data from all damage scenarios. The soundness of the proposed methodology was demonstrated in a robustness assessment, where all eight damage scenarios were correctly identified even when two of them were excluded from the training dataset.