Streaming variational inference-empowered Bayesian nonparametric clustering for online structural damage detection with transmissibility function

verfasst von: Ling Feng Mei, Wang Ji Yan, Ka Veng Yuen, Michael Beer
Abstract: Transmissibility function (TF) is widely applied in damage detection due to its sensitivity to damage and robustness to external excitations, but its application in online damage detection is rarely reported due to challenges in handling data streams. This study proposes a new TF-based online damage detection method that integrates a truncation-free variational inference-based full Dirichlet process Gaussian mixture model (VI-FDPGMM) within a streaming variational inference (SVI) paradigm. As an improved Bayesian nonparametric approach, the truncation-free VI-FDPGMM addresses the issue of truncating mixing components in traditional VI-DPGMM for online learning with increasing data by strategically setting the variational distributions of parameters for the components without assigned data (i.e., inactivated components) to their prior distributions based on the Bayesian viewpoint, which enables computing the probabilities to assign data points to these components and determining the creation of new components. As a result, the truncation-free VI-FDPGMM allows dynamically adding components to the mixture model, providing the flexibility to automatically adapt the number of components for arbitrary amounts of data. This characteristic enables its intuitive integration into the SVI paradigm featured as the variational posterior conditioned on the previous data as the prior when new data are observed, facilitating continuous refinement of the mixture model without repeatedly making inference of previous data. Therefore, the proposed method is highly efficient and well-suited for online damage detection. The proposed method is validated using two case studies, demonstrating its capability to dynamically generate new clusters as new data are available online to indicate the emergence of new damage patterns during the monitoring process, which enables it to perform structural anomaly detection tasks in a semi-supervised manner. Furthermore, the method outperforms some state-of-the-art methods due to its capability for continuous model refinement and robustness in interpreting and capturing uncertainties.
Organisationseinheit(en): Institut für Risiko und Zuverlässigkeit
Externe Organisation(en): The University of Liverpool
Tsinghua University
University of Macau
Typ: Artikel
Journal: Mechanical Systems and Signal Processing
Band: 222
ISSN: 0888-3270
Publikationsdatum: 01.01.2025
Publikationsstatus: Veröffentlicht
Peer-reviewed: Ja
ASJC Scopus Sachgebiete: Steuerungs- und Systemtechnik, Signalverarbeitung, Tief- und Ingenieurbau, Luft- und Raumfahrttechnik, Maschinenbau, Angewandte Informatik
Elektronische Version(en): https://doi.org/10.1016/j.ymssp.2024.111767 (Zugang: Geschlossen)

BibTeX

@article{513c05e7d10e4e149e4f4648fbbe2e0a,
title = "Streaming variational inference-empowered Bayesian nonparametric clustering for online structural damage detection with transmissibility function",
abstract = "Transmissibility function (TF) is widely applied in damage detection due to its sensitivity to damage and robustness to external excitations, but its application in online damage detection is rarely reported due to challenges in handling data streams. This study proposes a new TF-based online damage detection method that integrates a truncation-free variational inference-based full Dirichlet process Gaussian mixture model (VI-FDPGMM) within a streaming variational inference (SVI) paradigm. As an improved Bayesian nonparametric approach, the truncation-free VI-FDPGMM addresses the issue of truncating mixing components in traditional VI-DPGMM for online learning with increasing data by strategically setting the variational distributions of parameters for the components without assigned data (i.e., inactivated components) to their prior distributions based on the Bayesian viewpoint, which enables computing the probabilities to assign data points to these components and determining the creation of new components. As a result, the truncation-free VI-FDPGMM allows dynamically adding components to the mixture model, providing the flexibility to automatically adapt the number of components for arbitrary amounts of data. This characteristic enables its intuitive integration into the SVI paradigm featured as the variational posterior conditioned on the previous data as the prior when new data are observed, facilitating continuous refinement of the mixture model without repeatedly making inference of previous data. Therefore, the proposed method is highly efficient and well-suited for online damage detection. The proposed method is validated using two case studies, demonstrating its capability to dynamically generate new clusters as new data are available online to indicate the emergence of new damage patterns during the monitoring process, which enables it to perform structural anomaly detection tasks in a semi-supervised manner. Furthermore, the method outperforms some state-of-the-art methods due to its capability for continuous model refinement and robustness in interpreting and capturing uncertainties.",
keywords = "Bayesian nonparametric model, Damage detection, Streaming variational inference, Truncation-free full Dirichlet process Gaussian mixture model",
author = "Mei, {Ling Feng} and Yan, {Wang Ji} and Yuen, {Ka Veng} and Michael Beer",
year = "2025",
month = jan,
day = "1",
doi = "10.1016/j.ymssp.2024.111767",
language = "English",
volume = "222",
journal = "Mechanical Systems and Signal Processing",
issn = "0888-3270",
publisher = "Academic Press Inc.",
}