Abstract
A new method for signal processing of non-linear and non-stationary data, the Hilbert Huang Transform (HHT), offers insight into signals that cannot be achieved using conventional methods such as the Fourier transform and wavelet decompositions. This study investigates HHT as a potential tool for damage detection, to be eventually incorporated into a realistic structural health monitoring system. After a review of the method and supporting research, an analytical study begins to offer insight into the behavior of HHT. First, HHT is performed on simple sinusoid signals to see how it separates frequency content, then on the same signal with noise added. Finally HHT is performed on data generated from a numerical 8-degree-of-freedom mass-spring system with random burst excitation, where the stiffness of one spring can be decreased to simulate damage. The study finds that there is considerable variability associated with the implementation of HHT. Many variables, such as sampling frequency and overlapping frequencies, can affect the output of HHT drastically, possibly detracting from the value of the result. The method may, however, automatically separate noise from a signal. Application of HHT to the mass-spring system showed that there was only a noticeable pattern to the effect of damage when that damage became severe. However, the implementation process used may need to be more refined before final assessment can be made. Generally, the variability of HHT needs to be better quantified and understood so that a damage index can be derived based on the unique information that the method provides from signals. If this is possible, HHT could be an important tool for structural damage detection and health monitoring.
