Automatic taxonomy and modal analysis for vibration diagnosis

The authors give a case study concentrating on the task of providing automatic aids to both diagnosis and prognosis by recognition of machine condition in the routinely monitored vibration signatures of turbine generator rotors. The target population of rotors is the two-pole, 660 MW turbine generator rotors designed and manufactured by the GEC Alsthom group. The signatures provided by GEC Alsthom are multi-sensor run-up and run-down signatures taken using tachometer pulse and tracking filter techniques. They consist of a number of phasor trajectories, amplitude and phase of transverse displacement at about 300 speed steps. The authors describe the automatic modal analysis of the signatures using the Kennedy-Pancu method and then discuss cluster-based reasoning. The problems encountered with real signatures and the results of simulated signatures are discussed