Non-parametric identification of non-linear oscillating systems

The problem of system identification from a time series of measurements is solved by using non-parametric additive models. Having only few structural information about the system, a non-parametric approach may be more appropriate than a parametric one for which detailed prior knowledge is needed. Based on non-parametric regression, the functions in the additive models are estimated by a penalized least-squares approach using backfitting. The optimal smoothing parameters are determined via generalized cross-validation, making this approach completely adaptive to the data. The procedure is applied to identify the non-linear restoring force of vibrationally excited helical wire rope isolators.