Premium
Sparse representation of ultrasonic guided‐waves for robust damage detection in pipelines under varying environmental and operational conditions
Author(s) -
Eybpoosh Matineh,
Berges Mario,
Noh Hae Young
Publication year - 2016
Publication title -
structural control and health monitoring
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.587
H-Index - 62
eISSN - 1545-2263
pISSN - 1545-2255
DOI - 10.1002/stc.1776
Subject(s) - sensitivity (control systems) , piping , computer science , pipeline transport , representation (politics) , ultrasonic sensor , structural health monitoring , range (aeronautics) , real time computing , structural engineering , acoustics , engineering , electronic engineering , mechanical engineering , physics , politics , political science , law , aerospace engineering
Summary The challenges of guided‐wave based structural health monitoring can be discussed under three headings: (a) multiple modes, (b) multi‐path reflections, and (c) sensitivity to environmental and operational conditions (EOCs). The objective of this paper is to develop damage detection methods that simplify guided‐wave signals while retaining damage information and have low sensitivity to EOC variations. A supervised method is proposed for damage detection. The detection performance is maximized, by imposing a sparsity constraint on the signals. This paper reports a diverse set of laboratory and field experiments validating the extent to which EOC variations, as well as damage characteristics can influence the discriminatory power of the damage‐sensitive features. The laboratory setup includes an aluminum pipe with temperature varying between 24 and 38 ° C. The method is further validated using an operational hot water supply piping system of different size and configuration than the one used in the laboratory, which operates under noisy environment, with constantly varying flow rate, temperature, and inner pressure. Moreover, the proposed method is used to detect occurrence of consecutive actual damages, namely, a crack and a mass loss as small as 10% and 8% of the wall thickness, respectively. The validation results suggest that a simple binary‐labeled training data (i.e., undamaged/damaged), obtained under a limited range of EOCs, are sufficient for the proposed method. That is, the detection method does not require prior knowledge about the characteristics of the damage (e.g., size, type, and location), and/or a training dataset that is obtained from a wide range of EOCs. Copyright © 2015 John Wiley & Sons, Ltd.