Fault Diagnosis Method Of Gear Box Based On The Time-frequency Feature Extraction And BP Neural Network

With the improvement of modern science and technology progress of industrialtechnology, continuous and automation have become the development direction of mechanicalequipment, the connection between devices becomes more closely. The gear box hasadvantages of large driving torque, high transmission accuracy and the compact structure, isvery important general parts in mechanical equipments. Vibration signal contains all theinformation on the running state of a gearbox, the vibration signal characteristics will changewhen the running state of a gear box is changed. Study on vibration characteristics of gear box,then extracting useful information from the vibration signal, has important significance for thereal-time monitoring and fault diagnosis of gear box, it can not only reduce the cost ofequipment repair, but also can effectively improve the reliability and stability of gear box inthe work process, which is of great significance for people’s life and property protection.Artificial neural network is a new cross-disciplinary, which is developed since the1940s.Different from the traditional data processing method, artificial neural network has powerfullearning ability and nonlinear, so it obtained the rapid development in the field of faultdiagnosis.This paper takes gearbox as the research object, established gear vibration model basedon the study of the gearbox’s internal structure, analyzed the mechanism of the gearboxvibration in detail. On this basis, a fault diagnosis test-bed of gear box was built to simulatefive kinds of common faults, then the time and frequency domain analysis method was usedto extract the feature of vibration signal, then used the normalized characteristic parameters totrain the BP network under different learning rate. After a series of training and tests, thefollowing conclusions were obtained: when other parameters are in right set, the network’s correct rate is more than86%in fault diagnosis, for the two or more than two kinds ofcombination faults, the network is still applicable.