Anomaly Detection with Universal Representation of Modal Testing Response Data

  • Chao Jiang
  • , Haoyu Wang
  • , Xuan Han
  • , Liang Yu
  • , Yingjun Deng

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

In structural testing, modal analysis of the response data reflects the specifications of a structure. This paper proposes a data-driven anomaly detection method using the universal representation of modal testing response data. High-frequency time series data are preprocessed and transformed into truncated frequency response signals. Subsequently, the universal representation learning of frequency response signals is realized through contrastive learning based on TS2Vec, and distance-based anomaly detection is performed using the K-nearest neighbor algorithm with semi-supervised learning. For a limited excitation experimental dataset consisting of 32 samples, the proposed method achieves a detection rate of 80.0%. This result demonstrates the validity of the universal representation of modal testing response data.

Original languageEnglish
Title of host publicationProceedings - 2023 14th International Conference on Reliability, Maintainability and Safety, ICRMS 2023
EditorsLiming Ren, W. Eric Wong, Hailong Cheng, Xiaopeng Li, Shu Wang, Kanglun Liu, Ruifeng Li
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages105-110
Number of pages6
ISBN (Electronic)9798350329988
DOIs
StatePublished - 2023
Externally publishedYes
Event14th International Conference on Reliability, Maintainability and Safety, ICRMS 2023 - Urumqi, China
Duration: 26 Aug 202329 Aug 2023

Publication series

NameProceedings - 2023 14th International Conference on Reliability, Maintainability and Safety, ICRMS 2023

Conference

Conference14th International Conference on Reliability, Maintainability and Safety, ICRMS 2023
Country/TerritoryChina
CityUrumqi
Period26/08/2329/08/23

Keywords

  • anomaly detection
  • frequency response
  • modal test
  • rotating machinery
  • signal

Fingerprint

Dive into the research topics of 'Anomaly Detection with Universal Representation of Modal Testing Response Data'. Together they form a unique fingerprint.

Cite this