On multi-site damage identification using single-site training data. (24th November 2017)

• Record Type: Journal Article
• Authors: Barthorpe, R.J.; Manson, G.; Worden, K.
• Is Part Of: Journal of sound and vibration. Volume 409(2017)
• Page Start: 43
• Page End: 64
• Publication Date: 2017-11-24
• Subjects: Structural health monitoring -- Support vector classification -- Multi-site damage identification -- Statistical pattern recognition; Sound -- Periodicals; Vibration -- Periodicals; Son -- Périodiques; Vibration -- Périodiques; Sound; Vibration; Periodicals; Electronic journals; 620.205
• DOI: 10.1016/j.jsv.2017.07.038
• Languages: English
• ISSNs: 0022-460X

Abstract

Abstract: This paper proposes a methodology for developing multi-site damage location systems for engineering structures that can be trained using single-site damaged state data only. The methodology involves training a sequence of binary classifiers based upon single-site damage data and combining the developed classifiers into a robust multi-class damage locator. In this way, the multi-site damage identification problem may be decomposed into a sequence of binary decisions. In this paper Support Vector Classifiers are adopted as the means of making these binary decisions. The proposed methodology represents an advancement on the state of the art in the field of multi-site damage identification which require either: (1) full damaged state data from single- and multi-site damage cases or (2) the development of a physics-based model to make multi-site model predictions. The potential benefit of the proposed methodology is that a significantly reduced number of recorded damage states may be required in order to train a multi-site damage locator without recourse to physics-based model predictions. In this paper it is first demonstrated that Support Vector Classification represents an appropriate approach to the multi-site damage location problem, with methods for combining binary classifiers discussed. Next, the proposed methodology is demonstrated and evaluated through application to a real engineering structure – a Piper Tomahawk trainer aircraft wing – with its performance compared to classifiers trained using the full damaged-state dataset.