A machine-learning-based fault diagnosis approach for intelligent condition monitoring

We propose a machine-learning-based fault diagnosis approach for condition monitoring on the constant-speed rotating machines via vibration signals. There are mainly five phases in our approach, i.e., vibration signal measurement, discrete-wavelet-transformation-based preprocessing, feature extraction, base-line encoding, and fuzzy neural network. The advantage of this approach can identify the condition and faults of machine without sufficient diagnosis knowledge. Experimental results have demonstrated this approach is a useful tool for condition monitoring application.