J. Decuyper, J. P. Noël, T. De Troyer, M. C. Runacres, J. Schoukens
DOI Number: N/A
Conference number: IFASD-2017-053
Polynomial nonlinear state-space (PNLSS) models have proven to be very useful in modelling highly nonlinear systems, encountered over a variety of engineering applications. In this work, we focus on modelling the kinematics of an oscillating circular cylinder, submerged in a low Reynolds number flow. Such a set up is typically used to study vortex shedding phenomena and their related forces. The power of the PNLSS model class comes from its large flexibility in candidate nonlinear basis functions. Flexibility, however, comes at a price. The number of parameters generally grows large, hampering the identification process and leading to a loss of insight in the nonlinear functions. The objective of this work is to investigate how prior knowledge of the nonlinearity can be introduced in the basis functions of these nonlinear models and how this affects the accuracy of the estimated model. In particular, the usage of polynomial functions in terms of states and the input is compared to nonlinear functions in terms of the output variable. An improved model was obtained when a deliberate choice of basis functions was chosen based on prior knowledge of the nonlinearity. In addition, promising results were obtained from using dedicated nonlinear basis functions in terms of the output on a system closely related to the vortex shedding system.