In this paper we estimate the parameters of a multidimensional system from a record of noisy output measurements by using a multiwavelet denoising technique. In this output-only identification scheme, we extend wavelet denoising methods to the multiwavelet case. After the noise has been removed from the output records by wavelet methods, either full model identification or deterministic subspace identification can be performed. In the former case, full information on the system such as modal values and shapes becomes available by postprocessing. In the latter case, the observable modal values of the system as well as modal shapes at the sensor locations can be extracted from the identified parameters. Additionally, we discuss the requirements on the measuring devices to be compatible with wavelet transforms of a particular type. The validity and merit of the developed scheme are illustrated by examples of numerically simulated and experimentally measured signals, including comparisons with stochastic identification methods.
Subspace identification – a Markov parameter approach
