Nonlinear modeling of streaming electrification measured in swinging cylinder system

This paper is related to the phenomenon of streaming electrification. The selected results of nonlinear identification of the electrification process are presented. We described the signals used for identification and validation and then two structures of the models used: the nonlinear autoregressive model and the Hammerstein-Wiener model. We then conducted parametric identification and compared the obtained results. We used the Gauss function as the input signal. In the case of the autoregressive model, we used the sigmoid network to estimate the nonlinearity. In the case of the Hammerstein-Wiener model, we tested various types of input nonlinearities. The piecewise linear nonlinearity estimators turned out to be the most effective. Results for the most effective models of both structures: the results of the nonlinear ARX model and the Hammerstein-Wiener model were similar and had varying fit values in the range of 1 to 3 percent. They also had similar generalisation features. In all cases, we also obtained similar shapes of the processes of residues and correlation functions. They indicate a relatively faultless representation of the responses of models as well as certain differences, especially in the end part of the processes representing the given process.