Doctor of Philosophy

dissertationGuided wave structural health monitoring is widely researched for remotely inspecting large structural areas and di cult to access locations to determine whether the structure is damaged. As guided waves propagate through the structure, wave interactions with defects cause re ections in received signals. To detect, locate, and characterize damage, guided wave methods often compare data to a baseline signal. A test is used to determine whether damage has occurred. Perturbations in the damage statistic cause errors in that decision. A cause of these perturbations is temperature variations. Temperature variations create large di erences between the baseline and collected measurements. Temperature compensation algorithms, such as baseline signal stretch and the scale transform, have been used to correct for temperature e ects. While these methods are e ective in some conditions, their performance is limited by large temperature variations, long propagation distances, and high frequencies. Dynamic time warping has been applied as a temperature compensation method to better align guided wave data and to overcome errors. Improvements in temperature compensation are evaluated using statistical analysis. This analysis inspects the range and ability of temperature compensation methods to remain sensitive to damage in an environment of changing temperature and noise. To improve the robustness of damage detection, a multimeasurement baseline extension framework is used. By leveraging the similarities between measurements, clustering of measurements representing similar structural states are revealed. Using model change to identify these events over time, the presence of multiple events is extracted. While damage statistics constructed from batch-based strategies o er advantages over single-signal strategies, care must be taken to ensure data quality is su cient to reveal damage. An evaluation is performed on how compositions of an examined data set a ects batch-based damage statistics. The Anderson statistic is introduced as a singular value decomposition based damage statistic that is robust to noise and has greater sensitivity to damage than common batch-based statistics