An Investigation Into Two Alternative Approaches for the Identification of SFD Bearings for Aeroengine Analysis

Abstract

Identification techniques provide a means of efficiently implementing complex nonlinear bearing models in practical turbomachinery applications. This paper considers both identification from an advanced numerical model and identification from experimental tests. Identification from numerical models is essential at the design stage, where rapid simulation of the dynamic performance of a variety of designs is required. Experimental identification is useful to capture effects that are difficult to model (e.g. geometric imperfections, compressibility and its effect on cavitation). With regard to identification from a numerical model, it was shown in a previous paper that the numerical solution of the incompressible Reynolds equation may be replicated using Chebyshev polynomial fits. Tests were performed on a simple rotor-bearing configuration incorporating an advanced numerical bearing model. The identified model was found to be able to match the accuracy of the numerical solution to the Reynolds equation while requiring a fraction of the computation time. In the present work the SFD identification scheme is applied to a realistically-sized representative whole-aeroengine model. It is shown that using recently introduced nonlinear solvers combined with the identified high accuracy bearing models it is possible to run full engine rotor-dynamic simulations, in both the time and frequency domains, at a fraction of the previous computational cost. One major drawback of the Chebyshev technique is that it is not amenable to experimental identification of actual bearings. For this reason, a second identification approach, involving the use of neural networks, is considered in this paper. A test rig that enables empirical identification of SFD forces has been constructed and details of the building and operation of the test rig is presented. The method used to ascertain the training data required by the neural network identification scheme, is also described.