Using ANN for the recognition of vibration signals of off-shore equipment's failure

The studied offshore mechanical equipment are mainly divided into pumps, turbines and compressors. The first step in signal processing has been executed by traditional Kohonen self organizing maps for feature extraction from different vibration spectra. The network's varied input modes composted by definite frequency bands, overall evaluated the parameters and machine running temperature, and the analysis results show that the possible output categories represented almost 80% of typical failures and reached over 90% after some modification of input. In the second step with the accumulation of experience and cause-effect case studies, the modified backpropagation network has also been recommended for supplementing the network used. The synthesis of the two methods improved the reliability of whole survey.