Gaussian process modeling for damage detection in composite aerospace structures by using discrete strain measurements

Joham Alvarez-Montoya, M. A. Torres-Arredondo, Julián Sierra-Pérez

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

    3 Scopus citations

    Abstract

    The first approaches to damage detection in the context of Structural Health Monitoring (SHM) focused on identification techniques based on physical models, which imply the need of a model to describe the relationship between actions and responses in the structure. However, in the case of aerospace structures, where complex designs and materials are used to accomplish the demanding requirements, it is difficult to estimate accurate models to be used for diagnosis in aircraft Health and Usage Monitoring Systems (HUMS). This is even more evident with the increased use of composite materials in the aerospace industry. In this way, data-driven models which are based only on experimental data have been successfully implemented. Such techniques use machine learning algorithms to "learn" and evaluate the structural integrity with high accuracy. In previous work, the authors have developed a methodology based on Self-Organizing Maps (SOM) and Principal Component Analysis (PCA) for this kind of structures. Suitable results were obtained for strain data acquired from an Unmanned Aerial Vehicle (SMARP UAV) in flight tests by means of 20 Fiber Bragg Grating (FBG) sensors. However, there are different existing machine learning techniques that can be applied to develop novel methodologies for this kind of structures. One promising approach is Gaussian Process (GP) modeling, which can be seen as a generalization of a Gaussian distribution, adapted to classification problems for diagnosis. The main advantage of using this type of modeling is its well-founded approach to deal with the learning and model selection problem. Therefore, the aim of this work is to evaluate a methodology for damage detection in aerospace structures based on GP modeling by means of discrete strain measurements. The performance of the methodology is evaluated by using Receiver Operating Characteristic (ROC) curve analysis achieving a F1 score of about 0.944 in the best case.

    Original languageEnglish
    Title of host publicationProceedings of the 7th Asia-Pacific Workshop on Structural Health Monitoring, APWSHM 2018
    EditorsZhongqing Su, Shenfang Yuan, Hoon Sohn
    PublisherNDT.net
    Pages710-718
    Number of pages9
    ISBN (Electronic)9783000603594
    StatePublished - 2018
    Event7th Asia-Pacific Workshop on Structural Health Monitoring, APWSHM 2018 - Hong Kong, China
    Duration: 12 Nov 201815 Nov 2018

    Publication series

    NameProceedings of the 7th Asia-Pacific Workshop on Structural Health Monitoring, APWSHM 2018

    Conference

    Conference7th Asia-Pacific Workshop on Structural Health Monitoring, APWSHM 2018
    Country/TerritoryChina
    CityHong Kong
    Period12/11/1815/11/18

    Bibliographical note

    Publisher Copyright:
    © APWSHM 2018. All rights reserved.

    Keywords

    • Aerospace
    • Damage detection
    • Diagnosis
    • Gaussian processes
    • Machine learning

    Fingerprint

    Dive into the research topics of 'Gaussian process modeling for damage detection in composite aerospace structures by using discrete strain measurements'. Together they form a unique fingerprint.

    Cite this