Time-frequency Spectrum Segmentation Method for Separating Multimodal Guided Waves in Long Bones

Zhenggang Zhang,Dean Ta,Weiqi Wang

DOI: https://doi.org/10.1121/1.4708847

2012-01-01

Abstract:Ultrasonic guided waves have great potential for evaluating long bones. However, the measured signal often contains multiple wave modes because of the complicated characteristics of guided waves, which cause difficulties to further analysis. In this study, a hollow cylinder filled with a viscous liquid was used to model the long bone, and the multimodal signals were simulated using finite difference time-domain (FDTD) method. The time-frequency spectrum segmentation method was proposed to separate multiple modes. First, the Gabor time-frequency spectrum of multimodal signals was calculated. Second, a multilevel image segmentation algorithm, including the watershed and region growing, was used to find the corresponding area of each mode in the spectrum. Finally, time domain signals representing individual modes were reconstructed from these areas. The validations of this method were analyzed by simulated multimodal signals under different elastic modulus (EM) of long bones, with or without noise. The results showed that the Gabor time-frequency representations of the individual modes were in good agreement with the theoretical dispersion curves. This study suggests that time-frequency spectrum segmentation method can correctly separate multimodal guided waves, which provides a foundation for feature extraction of individual guided modes.