Modeling of rotating machinery: A novel frequency sweep system identification approach

Abstract

Abstract In this study, the dynamic modeling of rotating machinery, which is a harmonic excitation system, is investigated based on a nonlinear autoregressive (NARX) model with external inputs. Generally, NARX model-based techniques require Gaussian (white) noise, and thus these methods are not suitable for rotating machinery. Although there have been some reports on the modeling of harmonic excitation systems, the existing methods cannot establish a single-input single-output (SISO) NARX model to represent the rotating machinery over a wide range of rotational speeds. An improved modeling method called the frequency sweep system identification approach is proposed in this study to solve this issue. A discrete-time Fourier transform (DTFT) is performed on the system input and output datasets over a wide speed range to obtain the resulting spectra, and the amplitudes corresponding to the rotational frequencies are extracted and spliced together to convert multiple time-domain signals into one input data set and one output data set composed of frequency-domain data. Then, the modeling process can be carried out using the orthogonal forward search algorithm. Moreover, the effect of the data sequence on the identification results is discussed theoretically. A key feature of the proposed method is that the model structure detection and coefficient calculation are conducted with spliced frequency-domain vectors. The feasibility of the proposed modeling approach is validated through numerical and experimental cases. This work is a supplement to existing modeling methods based on the NARX model and provides a modeling basis for the analysis and design of rotating machinery in combination with the NARX model.