Vibration fault detection of large turbogenerators using neural networks

Vibration analysis for machine condition monitoring is a well established area where specific signal processing techniques are used to determine the operating condition of the machine. This paper reports on the application of a supervised backpropagation neural network to classify the vibration measured from several large 120 MW turbogenerators with journal bearings having four typical operating conditions consisting of acceptable condition, imbalance, resonance and severe preload. The network was successfully trained using preprocessed positive and negative frequency spectra and tested with all four conditions. It was found that the network was also able to detect the severity of imbalance of the rotor as well as a combination of small imbalance and acceptable condition at the same time