The Application Of Morphological Neural Network In Gearbox Fault Diagnosis

Modern equipment is developing towards high efficiency, high precision, automation and large-scale, equipment is more and more expensive. The fault of gear drive system would cause grave losses of life and property. Therefore, improving the reliability of gearbox is of great significance to reduce or eliminate gearbox fault and to ensure safe operation of equipment. Fault diagnosis technology is an important means to improve the reliability of the gear transmission system. The fault signal feature extraction and the design of classifier is the key of the gearbox fault diagnosis.On the basis of studying modulations theory of vibration signals of gearbox, this paper has analyzed the dynamic model of gear pair and bearing in gearbox, and learned correspondence between common faults and vibration signal characteristics. Then, a simulation model has been established which can responses the gearbox fault feature, that is a non- linear non-stationary signals containing periodic pulses. The capacity to extract gearbox fault feature has been researched using this simulation model and kinds of morphological filters. Results show that morphological difference filters have a better ability to extract periodic pulse feature with little calculation, and the weighted multi- scale morphological filters have the best ability to do that but based on many computations.In this paper, the signal parameters extracted above have been classified by using Morphological Neural Network(MNN). On the basis of learning lattice theory, this paper has studied on MNN model, deeply analyzed the deficiencies of the training algorithm and classification algorithm, brought forward concepts of local optimum of MNN and redundancy of hyper-cube cluster, and given the solutions. At last, practical examples of calculation shows feasibility of the gearbox fault classification by using MNN.