Diagnosis of rotating machines by utilizing a backpropagation neural net

The authors utilize a backpropagation neural net for the diagnosis of rotating machines. The abnormal vibrations due to imbalances, axis misalignments, and bolt-loosening have different spectra. Similar to a pattern recognition technique, the spectra of abnormal vibrations is used in obtaining characteristic feature vectors. For an experiment, a vibration test bench was constructed in such a way that artificial faults could be realized easily. The feature vectors of abnormalities obtained from the test bench were used for training the neural net. The performance of the trained neural net was tested in recognizing the causes of vibrations.<<ETX>>