Recently, there has been a focus on the use of inverse problem techniques in order to monitor rotor unbalance, and obtain a balancing solution, from vibration measurements on the casing and prior knowledge of the rotor-casing structure. In certain practical configurations that use nonlinear bearings like the squeeze-film damper (SFD) bearing, an inverse model of the bearing is an important part of the unbalance identification process. The inverse bearing model is used to estimate the journal vibration from casing vibration measurements, thus acting as a substitute for internal instrumentation in applications where the rotor is not accessible under operational conditions. Previous research has shown that an inverse bearing model can be identified using a trained Recurrent Neural Network (RNN) from experimental input/output data. However, the RNN was both trained and validated under simulated rotational conditions, wherein the motion was driven by two orthogonally-phased perpendicular shakers.

In this paper, a RNN of an inverse bearing model that is identified from experimental data generated under simulated rotational conditions is validated under actual rotational (i.e. unbalance-driven) vibration conditions. The necessary modifications to the test rig are presented, together with the identification/training procedure. The results of the validation tests show that the RNN is capable of predicting the frequency spectrum of the dynamic nonlinear response of the journal with reasonable accuracy. This inverse SFD bearing model can be thus used in a future work to identify rotor unbalance.

This content is only available via PDF.
You do not currently have access to this content.