Improved frequency sweep modeling method based on model prediction output error for rub-impact rotor system

This study considers the identification of the Nonlinear Auto-Regressive with eXogenous Inputs (NARX) models of the rotor systems, in which the identified NARX model may sometimes fail. By incorporating rotor response signals at different speeds into the modeling process using an Extended Forward Orthogonal Regression (EFOR) algorithm. This way, the data information involved in identification is more abundant, but this is still not enough. To completely solve the issue, an identification method based on Model Prediction Output (MPO) error is proposed in this paper. Which can filter all possible identification results during the identification process based on MPO error, thereby avoiding potential failed identification results and obtaining the optimal solution. The proposed method improves the accuracy of NARX model identification and reduces the need for professional knowledge. The simulation and experimental cases of a rub-impact rotor system are presented to illustrate the application and the effectiveness of the new identification approach.