Validation and Recovery of Vibration Data in Electromachine Systems Using Neural Network Software

Analysis of vibration data from rotating electro machine systems is an important means of machine diagnostics and preventive maintenance. Vibration signals, which represent the lateral displacement of a shaft, are obtained from proximity probes and transmitted to a data acquisition system. Invariably, data distortions frequently occur because of the high sensitivity of the probes and transmission channels, imperfections of the shaft surface under the probes, and several other factors. To screen out corrupt data, a tedious and time-consuming visual inspection must precede further analysis. This paper describes a successful endeavor of using neural network software for validation and recovery of distorted vibration data. Structure of the developed networks, preparatory and training procedures, and experimental results are presented.