期刊
APPLIED SCIENCES-BASEL
卷 11, 期 4, 页码 -出版社
MDPI
DOI: 10.3390/app11041716
关键词
Structural Health Monitoring; machine learning; damage detection; extreme function theory; non-destructive testing; extreme value theory; generalised extreme distribution
This paper presents an application of Extreme Function Theory (EFT) and Gaussian Processes (GPs) for mode shape-based damage detection and Structural Health Monitoring (SHM), expanding on the concepts of Extreme Value Theory (EVT) and utilizing Gaussian Process Regression (GPR) for outlier detection in various engineering examples.
Featured Application This paper presents an application of Extreme Function Theory (EFT) and Gaussian Processes (GPs) to perform mode shape-based damage detection and Structural Health Monitoring (SHM). Any damaged condition is a rare occurrence for mechanical systems, as it is very unlikely to be observed. Thus, it represents an extreme deviation from the median of its probability distribution. It is, therefore, necessary to apply proper statistical solutions, i.e., Rare Event Modelling (REM). The classic tool for this aim is the Extreme Value Theory (EVT), which deals with uni- or multivariate scalar values. The Extreme Function Theory (EFT), on the other hand, is defined by enlarging the fundamental EVT concepts to whole functions. When combined with Gaussian Process Regression (GPR), the EFT is perfectly suited for mode shape-based outlier detection. In fact, it is possible to investigate the structure's normal modes as a whole rather than focusing on their constituent data points, with quantifiable advantages. This provides a useful tool for Structural Health Monitoring, especially to reduce false alarms. This recently proposed methodology is here tested and validated both numerically and experimentally for different examples coming from Civil and Aerospace Engineering applications. One-dimensional beamlike elements with several boundary conditions are considered, as well as a two-dimensional plate-like spar and a frame structure.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据