Abstract: Based on analysis of the principle of typical faults and the basic charateristics of coal crusher, the vibration signal generated by crusher in operation can first be decomposed into different wave bands by analysis and then inputted into bp neural network in forms of eigenvectors after energy normalization process. The network-trained neural network can be used for identification of any actual fault of the crusher. As evidenced by test result, the technology is high in accuracy and capable of effectively identifying and early warning of any kinds of faults of crusher in speration.