Empirical identification of the inverse model of a squeeze-film damper bearing using neural networks and its application to a nonlinear inverse problem

The identification of nonlinear squeeze-film damper (SFD) bearings, typically used in aero-engines, has so far focused on their forward model (i.e. displacement input/force output). The contributions of this paper are the non-parametric identification of the inverse model of the SFD bearing (force input/displacement output) from empirical data, and its application to a nonlinear inverse rotor-bearing problem. This work is motivated by the need for a reliable substitute for internal instrumentation, to enable the identification of rotor unbalance using vibration data from externally mounted sensors, in applications where the rotor is inaccessible under operating conditions and there is no adequate linear connection between rotor and casing. The identification of the inverse model is fundamentally different from that of the forward model due to the need to account for system memory. A suitably trained Recurrent Neural network (RNN) is shown to be capable of identifying the inverse model of an actual SFD through two validation studies. In the first study, the RNN model satisfactorily predicted the SFD journal’s displacement time histories for given periodic time histories of the Cartesian SFD forces, although it could not predict the user-applied static offset in the SFD since it was not trained to do so. This was no limitation for the second study where, for both centred and non-centred SFD conditions, the RNN proved to be a reliable substitute for actual instrumentation as part of the inverse problem solution process for identifying the amplitudes and phases of the external excitation forces on a simple test rig.