The evaluation of Nonlinear Output Frequency Response Functions based on tailored data-driven modelling for rotor condition monitoring

Poorly assembled or faulty rotors often behave nonlinearly. The accurate extraction of weak nonlinear features is a vital issue. This paper proposes to extract the partial dynamic properties of concern for nonlinear feature based rotor condition monitoring. The realization of this concept relies on a novel tailored data-driven NARX (Nonlinear Auto Regressive with eXogenous input) modelling approach. The tailored NARX model represents part of the system's dynamic properties of concern. The details are first to use the harmonic product spectrum to determine the rotational speed of the inspected rotor system. After that, the least mean squares method is used to extract the harmonic components from the measured vibration signal. The low-order harmonics of concern are then reconstructed for the tailored NARX modelling. Then the Nonlinear Output Frequency Response Functions are evaluated from the tailored NARX model as features for rotor condition monitoring. Finally, the proposed method is validated by simulation and experiment cases. The results show that the obtained features based on the proposed tailored NARX modelling approach are more robust and more accurate than conventional methods when assessing rotor faults. The proposed approach provides a reference for the inspection of the assembly quality of components such as bearings and the on-line health monitoring of rotating machinery.

Automated summary

This paper proposes a method for rotor condition monitoring based on nonlinear features, which can accurately extract weak nonlinear features. This method uses a novel tailored data-driven NARX modeling approach to extract the partial dynamic properties of concern and reconstructs the low-order harmonics of concern using the harmonic product spectrum and least mean squares method. Nonlinear Output Frequency Response Functions are then calculated from the tailored NARX model as features for rotor condition monitoring. Experimental results show that the obtained features based on the proposed tailored NARX modeling approach are more robust and accurate than conventional methods when assessing rotor faults. The proposed approach provides a reference for inspecting the assembly quality of components such as bearings and the online health monitoring of rotating machinery.