Steam turbine vibration diagnostic system based on neural hologram

The problem of condition monitoring of large rotating machines is well known by technicians working in the field of Energy. To record and analyze the vibration data coming from sensors properly mounted on the bearings of the machine is a common practice. Those data are usually processed by FFT and get available as Amplitude and Phase values individually for the all significant harmonics, which are multiple values of the rotor speed. From the analysis of these data, various malfunctions and/or incoming mechanical problems can be recognized, even if there are two main problems connected with data processing and analysis: (1) this kind of operation continuously requires a very large amount of data: for instance, on a typical steam turbine set, 5 signals are read for every bearing; at least 3 harmonics are evaluated in amplitude and phase for each bearing every second; due to the fact that a typical turbine of this size can have at least 5 bearings continuously monitored, the amount of data to manage should be about 576,000 per hour; (2) so far there are no intelligent methods to automatically control the actual behavior of the machine and to compare it with some reference statuses, in ordermore » to ask the operator for some corrective actions and/or a more detailed analysis of the data. An innovative method, based on Neural Networks, has been developed jointly by Ansaldo Ricerche and Ansaldo Energia. On the basis of the different steady-state working points: (1) a database is implemented with the data recorded from the field (thanks to the implicit posterior observation) when the behavior of the machine and the vibration phenomena problems are recognized as normal; (2) starting from the data stored in the database, a neural multi-dimensional hologram is then synthesized; (3) an automatic system compares continuously the values read from the field with those evaluated by the neural estimation.« less