Vibration-based damage detection for a population of nominally identical structures: Unsupervised Multiple Model (MM) statistical time series type methods. (October 2018)

• Record Type: Journal Article
• Authors: Vamvoudakis-Stefanou, K.J.; Sakellariou, J.S.; Fassois, S.D.
• Is Part Of: Mechanical systems and signal processing. Volume 111(2018)
• Page Start: 149
• Page End: 171
• Publication Date: 2018-10
• Subjects: Damage detection -- Population of structures -- Vibration-based methods -- Uncertainty -- Variable operating conditions -- Statistical time series type methods -- Unsupervised methods -- Structural Health Monitoring; Structural dynamics -- Periodicals; Vibration -- Periodicals; Constructions -- Dynamique -- Périodiques; Vibration -- Périodiques; Structural dynamics; Vibration; Periodicals; 621
• DOI: 10.1016/j.ymssp.2018.03.054
• Languages: English
• ISSNs: 0888-3270

Abstract

Highlights: Damage detection for a population of nominally identical structures is considered. Nominally identical structures exhibit variability in their dynamics. Proper Multiple Model (MM) based damage detection methods are postulated. The methods are systematically assessed and compared with 'conventional' ones. MM based methods achieve significantly improved damage detection performance. Abstract: The problem of vibration-based damage detection for a population of nominally identical structures is considered via unsupervised statistical time series type methods. For this purpose a population sample comprising 31 nominally identical composite beams with significant beam-to-beam variability in the dynamics is employed, with impact-induced damage at various positions and two distinct energy levels. Two Multiple Model, MM, based statistical time series type methods are postulated, assessed, and compared with two 'conventional' methods. The assessment is based on a comprehensive and systematic procedure, making use of thousands of test cases via a 'rotation' procedure, with the results presented in the form of Receiver Operating Characteristic, ROC, curves. These indicate that 'conventional' methods are mostly ineffective, especially with low impact energy damages. On the other hand, the postulated Multiple Model parameter based methods achieve significantly improved performance, characterized as very good and providing overall correct damage detection rates approaching 100 % for false alarm rates at or above 5 % .