A multi-model identification method for the fiber stretching process based on the EM algorithm

Abstract

In the fiber production process, the stretching process plays a key role in the quality of the final fiber product. Due to the fiber stretching process with inborn nonlinearity, the performance of a single controller and an optimizer may be compromised or even unsatisfactory. Thus, we consider a multi-model identification method for the fiber stretching process. The dynamic transitions among different operating points are achieved by the change of the operating conditions in the fiber stretching process. To excite all of the nonlinearity character in the fiber stretching process, the transitions among different operating conditions is achieved. The structure of each sub-models, operating points, operating range are assumed. Based on the input output data of the process, a linear parameter varying (LPV) model is built by applying a probability identification method. To achieve the smoothly connected among the different operating conditions, an exponential function is used. Then a global LPV model is constructed by synthesizing the local models. Simulated results show that the LPV method has the effectiveness in solving the inherent nonlinearity of the fiber stretching process.