Offline and online detection of damage using autoregressive models and artificial neural networks

Piotr Omenzetter, Oliver De Lautour

Research output: Chapter in Book/Report/Conference proceedingChapter (peer-reviewed)

4 Downloads (Pure)

Abstract

Developed to study long, regularly sampled streams of data, time series analysis methods are being increasingly investigated for the use of Structural Health Monitoring. In this research, Autoregressive (AR) models are used in conjunction with Artificial Neural Networks (ANNs) for damage detection, localisation and severity assessment. In the first reported experimental exercise, AR models were used offline to fit the acceleration time histories of a 3-storey test structure in undamaged and various damaged states when excited by earthquake motion simulated on a shake table. Damage was introduced into the structure by replacing the columns with those of a thinner thickness. Analytical models of the structure in both damaged and undamaged states were also developed and updated using experimental data in order to determine structural stiffness. The coefficients of AR models were used as damage sensitive features and input into an ANN to build a relationship between them and the remaining structural stiffness. In the second, analytical exercise, a system with gradually progressing damage was numerically simulated and acceleration AR models with exogenous inputs were identified recursively. A trained ANN was then required to trace the structural stiffness online. The results for the offline and online approach showed the efficiency of using AR coefficient as damage sensitive features and good performance of the ANNs for damage detection, localization and quantification.
Original languageEnglish
Title of host publicationSensors and Smart Structures Technologies for Civil, Mechanical, and Aerospace Systems 2007
Subtitle of host publicationProceedings of SPIE
EditorsMasayoshi Tomizuka, Chung-Ban Yun, Victor Giurgiutiu
PublisherSPIE
Number of pages12
Volume6529
ISBN (Print)9780819466501
DOIs
Publication statusPublished - 30 Apr 2007

Fingerprint

Neural networks
Damage detection
Stiffness
Time series analysis
Structural health monitoring
Excited states
Analytical models
Earthquakes

Cite this

Omenzetter, P., & De Lautour, O. (2007). Offline and online detection of damage using autoregressive models and artificial neural networks. In M. Tomizuka, C-B. Yun, & V. Giurgiutiu (Eds.), Sensors and Smart Structures Technologies for Civil, Mechanical, and Aerospace Systems 2007: Proceedings of SPIE (Vol. 6529). [65292N] SPIE. https://doi.org/10.1117/12.715853

Offline and online detection of damage using autoregressive models and artificial neural networks. / Omenzetter, Piotr; De Lautour, Oliver.

Sensors and Smart Structures Technologies for Civil, Mechanical, and Aerospace Systems 2007: Proceedings of SPIE. ed. / Masayoshi Tomizuka; Chung-Ban Yun; Victor Giurgiutiu. Vol. 6529 SPIE, 2007. 65292N.

Research output: Chapter in Book/Report/Conference proceedingChapter (peer-reviewed)

Omenzetter, P & De Lautour, O 2007, Offline and online detection of damage using autoregressive models and artificial neural networks. in M Tomizuka, C-B Yun & V Giurgiutiu (eds), Sensors and Smart Structures Technologies for Civil, Mechanical, and Aerospace Systems 2007: Proceedings of SPIE. vol. 6529, 65292N, SPIE. https://doi.org/10.1117/12.715853
Omenzetter P, De Lautour O. Offline and online detection of damage using autoregressive models and artificial neural networks. In Tomizuka M, Yun C-B, Giurgiutiu V, editors, Sensors and Smart Structures Technologies for Civil, Mechanical, and Aerospace Systems 2007: Proceedings of SPIE. Vol. 6529. SPIE. 2007. 65292N https://doi.org/10.1117/12.715853
Omenzetter, Piotr ; De Lautour, Oliver. / Offline and online detection of damage using autoregressive models and artificial neural networks. Sensors and Smart Structures Technologies for Civil, Mechanical, and Aerospace Systems 2007: Proceedings of SPIE. editor / Masayoshi Tomizuka ; Chung-Ban Yun ; Victor Giurgiutiu. Vol. 6529 SPIE, 2007.
@inbook{e26c76a44bb6468bab14dc6300a556fe,
title = "Offline and online detection of damage using autoregressive models and artificial neural networks",
abstract = "Developed to study long, regularly sampled streams of data, time series analysis methods are being increasingly investigated for the use of Structural Health Monitoring. In this research, Autoregressive (AR) models are used in conjunction with Artificial Neural Networks (ANNs) for damage detection, localisation and severity assessment. In the first reported experimental exercise, AR models were used offline to fit the acceleration time histories of a 3-storey test structure in undamaged and various damaged states when excited by earthquake motion simulated on a shake table. Damage was introduced into the structure by replacing the columns with those of a thinner thickness. Analytical models of the structure in both damaged and undamaged states were also developed and updated using experimental data in order to determine structural stiffness. The coefficients of AR models were used as damage sensitive features and input into an ANN to build a relationship between them and the remaining structural stiffness. In the second, analytical exercise, a system with gradually progressing damage was numerically simulated and acceleration AR models with exogenous inputs were identified recursively. A trained ANN was then required to trace the structural stiffness online. The results for the offline and online approach showed the efficiency of using AR coefficient as damage sensitive features and good performance of the ANNs for damage detection, localization and quantification.",
author = "Piotr Omenzetter and {De Lautour}, Oliver",
year = "2007",
month = "4",
day = "30",
doi = "10.1117/12.715853",
language = "English",
isbn = "9780819466501",
volume = "6529",
editor = "Masayoshi Tomizuka and Chung-Ban Yun and Victor Giurgiutiu",
booktitle = "Sensors and Smart Structures Technologies for Civil, Mechanical, and Aerospace Systems 2007",
publisher = "SPIE",

}

TY - CHAP

T1 - Offline and online detection of damage using autoregressive models and artificial neural networks

AU - Omenzetter, Piotr

AU - De Lautour, Oliver

PY - 2007/4/30

Y1 - 2007/4/30

N2 - Developed to study long, regularly sampled streams of data, time series analysis methods are being increasingly investigated for the use of Structural Health Monitoring. In this research, Autoregressive (AR) models are used in conjunction with Artificial Neural Networks (ANNs) for damage detection, localisation and severity assessment. In the first reported experimental exercise, AR models were used offline to fit the acceleration time histories of a 3-storey test structure in undamaged and various damaged states when excited by earthquake motion simulated on a shake table. Damage was introduced into the structure by replacing the columns with those of a thinner thickness. Analytical models of the structure in both damaged and undamaged states were also developed and updated using experimental data in order to determine structural stiffness. The coefficients of AR models were used as damage sensitive features and input into an ANN to build a relationship between them and the remaining structural stiffness. In the second, analytical exercise, a system with gradually progressing damage was numerically simulated and acceleration AR models with exogenous inputs were identified recursively. A trained ANN was then required to trace the structural stiffness online. The results for the offline and online approach showed the efficiency of using AR coefficient as damage sensitive features and good performance of the ANNs for damage detection, localization and quantification.

AB - Developed to study long, regularly sampled streams of data, time series analysis methods are being increasingly investigated for the use of Structural Health Monitoring. In this research, Autoregressive (AR) models are used in conjunction with Artificial Neural Networks (ANNs) for damage detection, localisation and severity assessment. In the first reported experimental exercise, AR models were used offline to fit the acceleration time histories of a 3-storey test structure in undamaged and various damaged states when excited by earthquake motion simulated on a shake table. Damage was introduced into the structure by replacing the columns with those of a thinner thickness. Analytical models of the structure in both damaged and undamaged states were also developed and updated using experimental data in order to determine structural stiffness. The coefficients of AR models were used as damage sensitive features and input into an ANN to build a relationship between them and the remaining structural stiffness. In the second, analytical exercise, a system with gradually progressing damage was numerically simulated and acceleration AR models with exogenous inputs were identified recursively. A trained ANN was then required to trace the structural stiffness online. The results for the offline and online approach showed the efficiency of using AR coefficient as damage sensitive features and good performance of the ANNs for damage detection, localization and quantification.

U2 - 10.1117/12.715853

DO - 10.1117/12.715853

M3 - Chapter (peer-reviewed)

SN - 9780819466501

VL - 6529

BT - Sensors and Smart Structures Technologies for Civil, Mechanical, and Aerospace Systems 2007

A2 - Tomizuka, Masayoshi

A2 - Yun, Chung-Ban

A2 - Giurgiutiu, Victor

PB - SPIE

ER -