A principled multiresolution approach for signal decomposition.

One of the common issues with the Structural Health Monitoring (SHM) of civil infrastructure, is that signals measured from structures are nonstationary. The nonstationarity is the result of uncontrolled variations in the environmental and operational variations (EOVs) in the structure, which result from the fact that monitoring can only be carried out in situ. Unfortunately, the EOVs prove to be a problem for damage detection, as they can mask the onset of damage, itself a nonstationary process. The aim of the current paper is to demonstrate a principled, but simple, decomposition method, which can separate signals into stationary and nonstationary components; the stationary component can then be used as a feature for SHM. The method is based on a wavelet/multiresolution analysis of the signal of interest, followed by a sequence of wavelet level-by-level tests for stationary using the ADF (Augmented Dickey-Fuller) test statistic. The decomposition is illustrated using data from the famous SHM campaign on the Z24 Bridge. The paper should be of interest, not just to engineers, but to econometricians, or anyone concerned with nonstationary signal processing.