Condition Monitoring Insight Using Bayesian Inference and Rotor Dynamics Modelling for Rotating Machinery

Abstract

Rotor dynamics modelling can be used to predict vibration levels for given inputs, such as unbalance levels and location, which may be of interest for condition monitoring or diagnosis. However, given measured vibration, using rotor dynamics models to find the corresponding root cause inputs is not straightforward. In the method presented in this paper, Gaussian Process models are developed as surrogates for the rotor dynamics finite element models, and are used with Bayesian Inference to determine the probability distributions of model inputs for a given vibration response. This method allows parameters describing the machine condition, such as unbalance location and magnitude, and bearing clearances, to be determined as well as the confidence in these predictions. The method is demonstrated by simulating the vibration response of a compressor rotor, adding noise to it, and then using the technique to accurately infer useful information such as the unbalance magnitude and location, and the clearance in each bearing. This technique can be applied as a risk-based approach to condition monitoring of rotating machinery. Further development of this approach as part of a digital twin which uses in-service measurements would provide operators with insight into the likelihood of different root causes of vibration, and the corresponding machine condition.