Failure identification of gear systems using Hilbert-Huang transform and artificial neural networks

The purpose of this research effort is to propose a new technique for predicting the gearbox condition through the vibration signals based on the Hilbert-Huang transformation (HHT) method. Vibration signals originating from the gearbox are considered to be highly non-stationary and nonlinear in nature. HHT has proven to be very successful for the analysis of such signals as a time-frequency representation method. However, its use as a feature extractor for pattern recognition has been barely explored. Experiments are conducted to record the vibration signals corresponding to the normal and abnormal conditions of a gearbox at various loading and frequency levels. The vibration signals are analyzed by HHT in the Matlab 7.0 programming environment to extract features as indicators of the gearbox condition. These features are later tested under various statistical and neural pattern recognition techniques. Results indicate a 100% classification rate for identifying the gearbox conditions with Artificial Neural Networks.