Use this URL to cite or link to this record in EThOS: https://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.801118
Title: Temperature-driven anomaly detection methods for structural health monitoring
Author: Zhu, Yanjie
ISNI:       0000 0004 8504 9953
Awarding Body: University of Warwick
Current Institution: University of Warwick
Date of Award: 2018
Availability of Full Text:
Access from EThOS:
Access from Institution:
Abstract:
Reported in this thesis is a data-driven anomaly detection method for structural health monitoring which is based on the utilization of temperature-induced variations. Structural anomaly detection should be able to identify meaningful changes in measurements which are due to structural abnormal behaviour. Because, the temperature-induced variations and structural abnormalities may produce significant misinterpretations, the development of solutions to identify a structural anomaly, accounting for temperature influence, from measurements, is a critical procedure to support structural maintenance. A temperature-driven anomaly detection method is proposed, that introduces the idea of blind source separation for extracting thermal response and for further anomaly detection. Two thermal feature extraction methods are employed corresponding to the classification of underdetermined and overdetermined methods. The underdetermined method has the three phases of: (a) mode decomposition by utilising Empirical Mode Decomposition or Ensemble Empirical Mode Decomposition; (b) data reduction by performing Principal Component Analysis (PCA); (c) blind separation by applying Independent Component Analysis (ICA). The overdetermined method has the two stages of the pre-indication according to PCA and the blind separation by the devotion of ICA. Based on the extracted thermal response, the temperature-driven anomaly detection method is later developed in combination with the four methodologies of: Moving Principal Component Analysis (MPCA); Robust Regression Analysis (RRA); One-Class Support Vector Machine (OCSVM); Artificial Neural Network (ANN). Therefore, the proposed temperature-driven anomaly detection methods are designed as Td-MPCA, Td-RRA, Td-OCSVM, and Td-ANN. The proposed thermal feature extraction methods and temperature-driven anomaly detection methods have been investigated in the context of three case studies. The first case is a numerical truss bridge with simulated material stiffness reduction to create levels of damage. The second case is a purpose constructed truss bridge in the Structures Lab at the University of Warwick. The third case study is Ricciolo curved viaduct in Switzerland. Two primary findings can be confirmed from the evaluation results of these three case studies. Firstly, temperature-induced variations can conceal damage information in measurements. Secondly, the detection abilities of temperature-driven methods, which are Td-MPCA, Td-RRA, Td-OCSVM, and Td-ANN, for disclosing slight anomalies in time are more efficient when compared with the current anomaly detection method, which are MPCA, RRA, OCSVM, and ANN. The unique features of the author’s proposed temperature-driven anomaly detection method can be highlighted as follows: (a) it is a data-driven method for extracting features from an unknown structural system. In another word, the prior knowledge of the structural in-service conditions and physical models are not necessary; (b) it is the first time that blind source separation approaches and relative algorithms have been successfully employed for extracting temperature-induced responses; (c) it is a new approach to reliably assess the capability of using temperature-induced responses for anomaly detection.
Supervisor: Not available Sponsor: University of Warwick ; Chinese Scholarship Council
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.801118  DOI: Not available
Keywords: QA Mathematics ; TA Engineering (General). Civil engineering (General)
Share: