Estimating turbogenerator foundation parameters: model selection and regularization

Estimating a model of the foundation of large machines, such as turbogenerators, is vital for tasks such as fault diagnosis and modal balancing. Unfortunately, it is rarely possible to perform a modal test on the foundations without the rotor present. This paper considers a method to estimate a foundation model using response data from a run–down, using the inherent unbalance of the rotor to excite the foundations. The method requires an accurate model of the rotor and an approximate model of the bearings. Of critical importance is the quality of the model, which may be defined as the ability of the model to predict the response to unbalance excitations that are different from that present for the identification. It is shown in this paper that correct model selection and regularization are vital to produce a foundation model that meets this predictability criterion. The method is validated using simulated data and also experimental data from a test rig with a 4 m long rotor with four fluid film bearings.