Accuracy in the Identification of a Generator Thermal Bow

Model-based diagnostic techniques can be used to identify the faults that affect rotating machines. In general, the most important faults and malfunctions can be modelled by means of a suitable set of equivalent excitations that are applied to the nodes of a finite element model of the machine rotor train. Weighted least-squares error methods can be used to identify the set of forces and moments that minimize the error between experimental transient vibrations and the rotor system response obtained with the simulating model. However, the accuracy of the fault identification can be significantly influenced by the adequacy of the fault model as well as by the accuracy of the rotating machine model. This paper shows some methods, based on basic statistics, which enable the accuracy of the fault identification to be evaluated. The capabilities of the proposed methods have been proved by the results of the identification of the shaft thermalbow of a power unit generator that exhibited abnormal vibrations during machine coastdowns. The accuracy with which the location and the severity of the fault have been estimated is analyzed. In addition to this, the results obtained by the identification of the actual faults have been compared with those provided by the identification of false faults that cause similar symptoms.