Increasing the robustness of fault identification in rotor dynamics by means of M-estimators

One of the most common problems in rotordynamics is the identification of faults and model-based methods are often used for this purpose. In some applications, the least-squares (LS) estimate is used to find out the position and the severity of impending faults on the basis of experimental vibration data of rotating machinery. Anyhow LS are not very robust with respect to possible outliers (noise and gross errors) in the experimental data and to inaccuracies in the model. The introduction of weights in the LS algorithm has proven to be effective in increasing the robustness and successful experimental cases, both on test rigs and on real machines, are reported in literature. However, the arbitrary choice of the weights is normally based on operators’ experience. In this paper, an improvement is presented by introducing a method that is robust in itself, the M-estimate, which allows defining automatically the weights. This method is general and can be applied in every problem of regression or estimation, not necessarily related to rotordynamics. The fundamental theoretical aspects are introduced in the first part, while several experimental test cases are presented by means of faultidentification on a test rig and on a gas turbo generator in the second part of the paper. The obtained results highlight the increasing of the accuracy allowed by M-estimate.