Parameter estimation by data reconstruction

This paper considers the estimation of the parameters of structured state-space models, based on experimental data. An efficient filtering procedure to reconstruct the system state and its derivative is proposed, using a special choice of non-causal moving average (MA) filter, which represents a least squares polynomial fit through the data, enabling the calculation of smooth higher order derivatives. This enables reliable parameter estimation, under some smoothness conditions on the true data, and assumptions about the measurement noise. Most mechanical systems satisfy these assumptions, by suitable experiment design. If the system model can be linearly parameterized, as is often the case with mechanical systems, the parameters can be efficiently computed by least squares methods. The proposed procedure is carried out successfully on an experimental hydraulic actuator setup.