Fault diagnosis for valve train based on EMD, SVD and Mahalanobis distance
Xu Wang · Modern Machinery · 2008
By setting the different exhaust valve clearance and simulated minor leak with new valves, the four common working conditions of the valve train have been constructed. And against the non-stationary cylinder head vibration signals, a new method to improve non-stationary signals is introduced. The non-stationary signal is decomposed by empirical mode decomposition in Hilbert-Huang transform to reduce the non-stationarity in the signals. Based on this method and Singular Value Decomposition, some parameters extracted from cylinder head vibration signals are used for diagnosis. And then four common patterns vectors are received from small amount of training samples. Finally, MAHALANOBIS distance is used to identify the fault of the valve train. The results show that the model could finished with small amount of training samples, and once the training completed, the speed and recognition rate of unknown samples were both high. And it is easy to realize the on-line monitoring and diagnosis of valve train faults.