Unsupervised learning-based framework for indirect structural health monitoring using adversarial autoencoder. (28th April 2023)

• Record Type: Journal Article
• Authors: Calderon Hurtado, A.; Kaur, K.; Makki Alamdari, M.; Atroshchenko, E.; Chang, K.C.; Kim, C.W.
• Is Part Of: Journal of sound and vibration. Volume 550(2023)
• Publication Date: 2023-04-28
• Subjects: Adversarial autoencoder -- Deep learning -- Indirect SHM -- Bridge monitoring -- Damage; Sound -- Periodicals; Vibration -- Periodicals; Son -- Périodiques; Vibration -- Périodiques; Sound; Vibration; Periodicals; Electronic journals; 620.205
• DOI: 10.1016/j.jsv.2023.117598
• Languages: English
• ISSNs: 0022-460X

Abstract

Abstract: This paper studies the problem of bridge health monitoring in an unsupervised manner utilizing only the measured responses from a vehicle passing over a bridge. A half-car model along with a simply supported beam is adopted for numerically simulating vehicle–bridge interaction (VBI). Multiple bridge states including healthy bridge and damaged bridge with varying severity at different locations are considered. A data-driven approach based on adversarial autoencoder (AAE) by incorporating the generative capabilities of adversarial autoencoder and pre-processing techniques including frequency filtering and signal averaging is proposed. Vehicle acceleration responses associated only with the healthy bridge state are used for model training. Reconstruction error estimated by the proposed model is adopted as a damage detection index. Along with the detection of the damage in the bridge, the proposed framework is also able to estimate the severity of the damage. The proposed framework also overcomes the limitations of other unsupervised learning approaches such as principal component analysis and autoencoders due to its better representation of data in the latent sub-space with an additional prior distribution constraint. Further, the proposed framework is validated through experimentally obtained data from a laboratory scale bridge model. The contribution of this work is three-fold: First, an adversarial autoencoder-based unsupervised learning framework supplemented by appropriate pre-processing techniques is proposed for drive-by bridge monitoring for the first time, and its implementation is extensively investigated. Second, the superior performance of the proposed AAE framework compared to the competing techniques is demonstrated. Finally, this paper presents one of the early successful attempts of drive-by bridge inspection for monitoring the progressive change in the structure of a bridge. Research presented in this work can potentially open up new opportunities for condition monitoring of bridge networks. Highlights: An adversarial autoencoder-based framework is proposed for drive-by bridge inspection. The framework is based on unsupervised learning requiring no data from damaged states. Its superior performance compared to competing techniques is demonstrated.