Application of subspace methods to the identification of a winding process

In this paper, a linear model identification of a winding pilot plant for a tension control application is presented. Such a system is characterised by strong interactions between the tensions, frictions, distorsione, sliding and bending effects which make the tension control difficult to handle. Moreover, most of physical state variables as the radius are not measured. Herein, a black-box identification approach based on subspace methods is developed in order to estimate a nominal state-space model around a given operating point. A Linear Quadratic Gaussian (LQG) controller has been implemented to illustrate the applicability of the estimated model. While the pilot plant is non linear, results of this application show that such an ‘average’ state-space model is enough to obtain interesting tension control performances on the whole winding zone.