A Review of Vibration-Based Damage Detection in Civil Structures: From Traditional Methods to Machine Learning and Deep Learning Applications (2004.04373v1)

Abstract: Monitoring structural damage is extremely important for sustaining and preserving the service life of civil structures. While successful monitoring provides resolute and staunch information on the health, serviceability, integrity and safety of structures; maintaining continuous performance of a structure depends highly on monitoring the occurrence, formation and propagation of damage. Damage may accumulate on structures due to different environmental and human-induced factors. Numerous monitoring and detection approaches have been developed to provide practical means for early warning against structural damage or any type of anomaly. Considerable effort has been put into vibration-based methods, which utilize the vibration response of the monitored structure to assess its condition and identify structural damage. Meanwhile, with emerging computing power and sensing technology in the last decade, Machine Learning (ML) and especially Deep Learning (DL) algorithms have become more feasible and extensively used in vibration-based structural damage detection with elegant performance and often with rigorous accuracy. While there have been multiple review studies published on vibration-based structural damage detection, there has not been a study where the transition from traditional methods to ML and DL methods are described and discussed. This paper aims to fulfill this gap by presenting the highlights of the traditional methods and provide a comprehensive review of the most recent applications of ML and DL algorithms utilized for vibration-based structural damage detection in civil structures.

• The paper presents a comprehensive review of the evolution from traditional vibration-based methods to advanced ML and DL techniques for structural damage detection.
• It details methodologies including nonparametric and parametric approaches, highlighting their strengths and limitations in accurately assessing damage.
• The review underscores the success of CNN-based methods for real-time monitoring and points to future directions like unsupervised learning and IoT integration.

A Comprehensive Review of Vibration-Based Damage Detection in Civil Structures: From Traditional to Emerging Machine Learning and Deep Learning Techniques

The advancement of damage detection methods in civil structures is a critical component of maintaining the structural integrity, safety, and operational life of infrastructure assets. Monitoring structural damage is crucial for early intervention and prolonging the service life of civil structures. This paper by Avci et al. provides an extensive review of the evolution from traditional vibration-based structural damage detection (SDD) methods to contemporary applications utilizing Machine Learning (ML) and Deep Learning (DL) algorithms.

Traditional Vibration-Based Damage Detection Methods

Traditional methods for SDD primarily focus on altering the vibration response of structures to identify damage. These techniques are broadly categorized into nonparametric and parametric approaches.

Nonparametric Methods: These techniques utilize statistical metrics derived directly from vibration signals without requiring modal parameter estimation. Methods such as Auto-Regressive Moving Average (ARMA) modeling, Gaussian Mixture Models (GMM), and Auto-Regressive with eXogenous input (ARX) models are prevalent. These methods offer significant advantages in terms of computational efficiency and practicality but can be limited by their sensitivity to operational and environmental variability.

Parametric Methods: In contrast, parametric methods involve estimating the modal parameters such as natural frequencies, mode shapes, and damping ratios to detect changes indicative of damage. Techniques like Complex Mode Indicator Functions (CMIF), Frequency Domain Decomposition (FDD), and Stochastic Subspace Identification (SSI) are common. The main limitation of parametric methods lies in their dependency on accurate system identification, which can be computationally intensive and sensitive to external factors.

Machine Learning Techniques in SDD

The emergence of ML has introduced robust and adaptive methods for SDD. ML algorithms such as Artificial Neural Networks (ANNs), Support Vector Machines (SVMs), and Probabilistic Neural Networks (PNNs) have been extensively applied in both parametric and nonparametric damage detection.

Nonparametric ML-Based Methods: These methods bypass the need for explicit feature extraction by employing time-series modeling directly on the vibration response. Examples include the use of AR coefficients, Principle Component Analysis (PCA), and Singular Value Decomposition (SVD) as feature extractors, which are then classified using ML algorithms. For instance, Figueiredo et al. demonstrated effective damage detection using Auto-Associative Neural Networks (AANN) combined with AR modeling and PCA, achieving superior performance compared to traditional methods.

Parametric ML-Based Methods: These approaches involve training ML models on extracted modal parameters to identify and localize damage. Several studies have successfully employed ANNs to model the relationship between modal characteristics and damage scenarios. Notably, Mehrjoo et al. used modal frequencies and mode shapes as inputs to a multi-layer perceptron (MLP) to detect and quantify damage in truss structures. Similarly, Yuen and Lam applied Bayesian-designed ANNs to optimize the network's complexity for effective damage localization.

Deep Learning in Vibration-Based SDD

DL has revolutionized SDD by enabling automated feature extraction directly from raw data, thus eliminating the dependency on hand-crafted features. Among the various DL architectures, Convolutional Neural Networks (CNNs) have shown remarkable success in processing 1D vibration signals.

1D CNNs for SDD: Compact 1D CNNs are particularly advantageous due to their lower computational complexity and ability to process 1D data directly. Avci et al. demonstrated the effectiveness of 1D CNNs in real-time structural damage detection in a wireless sensor network. The method significantly outperformed traditional approaches in detecting and localizing damage under varying environmental conditions.

2D CNNs for SDD: For applications requiring spatial relationships, 2D CNNs have been employed by transforming vibration signals into 2D representations. Studies by Khodabandehlou et al. and Yu et al. utilized deep 2D CNNs to classify damage states in structures, achieving high accuracy in both damage identification and quantification.

Implications and Future Directions

The transition to ML and DL methods in SDD signifies a substantial shift towards more automated, accurate, and real-time damage monitoring systems. While traditional methods laid the groundwork, the adaptive capabilities and robustness of ML and DL techniques provide comprehensive solutions to the challenges posed by varying operational and environmental conditions.

Future developments in this field should focus on:

1. Unsupervised and Semi-Supervised Methods: Many current approaches rely on supervised learning, which requires labeled data. Developing unsupervised or semi-supervised algorithms could enable effective damage detection without extensive labeled datasets.
2. Transfer Learning: Exploring transfer learning to adapt pre-trained models on related problems can significantly reduce the need for extensive training datasets.
3. Integration with IoT and Real-Time Systems: Enhancing the integration of ML and DL-based SDD methods with Internet of Things (IoT) devices for continuous and real-time monitoring.

In summary, the reviewed paper underscores the vital role of ML and DL in advancing SDD techniques, highlighting their potential to transform structural health monitoring into a more resilient and intelligent system.