Separation of harmonic excitation responses from contaminated measurements based on ICA

A methodology based on independent component analysis (ICA) is presented for the separation of harmonic excitation responses from contaminated measurements. ICA is an algorithm recently used for blind source separation. It assumes that observed signals are linear mixtures of statistically independent original signals. The original signals and mixing matrix are both unknown, and ICA provides a way to identify both solely by use of observed signals. Here, we assume that a structure is affected by two input forces, harmonic excitation and microtremors, and try to separate the responses under the extremely strict conditions that the responses to harmonic excitation and microtremors are almost in the same dimension. The two inputs are considered as two original signals, the acceleration responses as observed signals. ICA then is introduced and separation carried out. The methodology is applied to a numerical simulation and its feasibility verified.