Reference-Free Damage Detection Using Instantaneous Baseline Measurements Steven R. Anton * and Daniel J. Inman † Virginia Polytechnic Institute and State University, Blacksburg, Virginia 24061 and Gyuhae Park ‡ Los Alamos National Laboratory, Los Alamos, New Mexico 87545 DOI: 10.2514/1.43252 A novel method of guided wave-based structural health monitoring is developed in which no direct baseline data are required to identify structural damage. Conventional wave propagation structural health monitoring techniques involve the comparison of structural response data to a prerecorded baseline or reference measurement taken while the structure is in pristine condition. The need to compare new data to a prerecorded baseline can present several complications, including data management issues and difficulty in accommodating the effects of varying environmental and operational conditions on the data. To address the complications associated with baseline comparison, this new method accomplishes reference-free damage detection by acquiring what is referred to as an instantaneous baseline measurement for analysis. The instantaneous baseline technique is validated through both analytical and experimental testing. Analytical tests show that the instantaneous baseline method is able to correctly identify simulated damage. It is found experimentally that nonpermanent damage in the form of removable putty as well as permanent damage in the form of corrosion and cuts are all identifiable in thin aluminum plate test structures without direct comparison to baseline data when implementing the instantaneous baseline method. Nomenclature A 0 , A 1 = antisymmetric Lamb wave modes a = peak cross-correlation value b = sum of original modified cross-correlation values with damaged path subtracted c = sum of scaled modified cross-correlation values d = damaged path index i = reference path index, also column index j = comparison path index, also column index k = row index MCC n = modified cross-correlation value after damaged path removal MCC s = scaled modified cross-correlation value n = total number of paths, also number of samples in each Lamb wave signal p = power of individual Lamb wave signals p m = mean power of all Lamb wave signals r = principal component analysis ratio S 0 , S 1 = symmetric Lamb wave modes V = matrix of eigenvectors of covariance matrix V = voltage v = eigenvector entry  v = mean of eigenvector entries X = matrix of Lamb wave data x = entries in matrix of Lamb wave data  = diagonal matrix of eigenvalues of covariance matrix  = eigenvalue  = covariance matrix I. Introduction S TRUCTURAL health monitoring (SHM) has gained a significant amount of attention in the research and industrial communities over the last two decades. The concept of actively monitoring structures for damage is of interest because it presents the ability to detect and locate damage in a structure before it can propagate and cause serious failure. The ability to know when and where damage has occurred in a structure can reduce the costs associated with scheduled inspections and the repair of failed structures, and also improve the overall safety of the structure. Several successful guided wave-based SHM techniques have been developed in the literature. Review articles summarizing the relevant work on wave propagation-based SHM are presented by Giurgiutiu and Cuc [1], and Raghavan and Cesnik [2]. More specifically, methods have been explored for detecting damage in a variety of structures including thin metal plates [3–5], aircraft panels [6–8], composite materials [9–11], and civil structures [12–15]. These methods rely on some knowledge of the structure in a healthy state to identify damage. Typically, baseline measurements are recorded when a structure is pristine and are stored for comparison to future data for damage detection. Changes in the newly recorded data when compared to the baseline are used to identify structural damage. One concern with the use of these baseline subtraction methods is the ability to discern structural changes from the effects of varying environmental and operational conditions when analyzing the struc- tural response of a system. Changes in environmental and operational conditions such as fluctuations in temperature, variation in surface moisture, and varying loading conditions can all cause the response of a structure to change significantly from the baseline measurement. The use of a standard baseline comparison method may falsely indicate damage or allow damage to go undetected in the presence of these varying conditions. Additionally, standard baseline compar- ison methods are unable to detect damage that exists in a structure before the installation of the health monitoring system, therefore, they are only useful for detecting future damage. Several researchers have investigated the development of SHM techniques that take into consideration the effects of varying envi- ronmental conditions, specifically temperature changes. Sohn et al. [16,17] present works in which neural networks are trained using features extracted from healthy baseline data and from data taken under various environmental conditions. The network is trained to be Received 15 January 2009; revision received 24 March 2009; accepted for publication 31 March 2009. Copyright © 2009 by Steven R. Anton. Published by the American Institute of Aeronautics and Astronautics, Inc., with permission. Copies of this paper may be made for personal or internal use, on condition that the copier pay the $10.00 per-copy fee to the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923; include the code 0001-1452/09 and $10.00 in correspondence with the CCC. * Ph.D. Candidate, Department of Mechanical Engineering, Center for Intelligent Materials Systems and Structures, 310 Durham Hall. † G.R. Goodson Professor and Director, Department of Mechanical Engineering, Center for Intelligent Materials Systems and Structures, 310 Durham Hall. Fellow AIAA. ‡ Engineering Institute. AIAA JOURNAL Vol. 47, No. 8, August 2009 1952