Multisensor Data Fusion Based on Independent Component Analysis for Fault Diagnosis of Rotor

Independent Component Analysis is applied to multi-channel vibration measurements to fuse the information of several sensors, and provide a robust and reliable fault diagnosis routine. Independent components are obtained from the measurement data set with FastICA algorithm, and their AR modeling estimates are calculated with BURG method. A probabilistic neural network is applied to the AR modeling parameters to perform the fault classification. Similar classification is applied directly to vibration measurements. Based on the results with real measurement data from the rotor test rig, it is shown that data fusion with ICA enhances the fault diagnostics routine.