A Method for Evaluating the Sensitivity of Signal Features in Pattern Recognition Based on Neural Network

In equipment monitoring and fault diagnosis, correctly evaluating and selecting the signal features contribute greatly to the effectiveness and accuracy of recognition result. Because it is difficult to create a criterion to evaluate the feature of measured signals in condition of small samples when we use traditional statistic pattern recognition theory to do this, this paper put forward a method for calculating the feature sensitivity via artificial neural network, and created a criterion function for evaluating the feature sensitivity. This criterion was applied in selecting the features of the diesel engine vibration.