Intelligent Gear Fault Diagnosis Based On Impact Features Analysis

In this paper,the factors that can influence the vibration response signal of gear transmission system are divided into fault and environmental state,and the data features caused by them are called fault features and environmental features.The training data of an intelligent gear fault diagnosis model usually comes from vibration experiments carried out under specific environmental state.Besides fault features,the training data also contains environmental features caused by specific environmental state.These environmental features cannot reflect the real health state of gears,but the diagnosis model cannot obtain this prior knowledge,and will be inevitably affected by these environmental features in its training process,what is more,the diagnosis model may even judge the health state of gears according to these environmental features,which is completely in violation of physical meaning,and it will cause that the diagnosis model cannot keep its effectiveness when environmental state is changed.The objective of fault diagnosis in this paper is localized faults of gear.Localized faults of gear will generate periodic impact excitation in gear transmission system,the data features of its response signal are affected by factors such as running speed and frequency response function of gear transmission system,running speed and frequency response function is part of environmental state.Through the analysis of impact features,it is found that: the amplitude spectrum features of localized fault vibration response signal are insensitive to the change of speed,while the time domain features are insensitive to the change of frequency response function.Therefore,in view of the change of running speed and frequency response function,a gear fault diagnosis method based on amplitude spectrum features and a method based on time domain impact features are proposed,and two gear fault diagnosis models based on convolution neural network(CNN)are proposed according to corresponding feature property.In addition to fault excitation,environmental state excitation also appears in gear transmission system,through the analysis of vibration characteristics,the fault response and the environmental state response in response signal can be distinguished to some extent,and then the environmental state response can be removed as far as possible in sample construction process,so as to reduce the influence of environment state on diagnosis model.In order to test the influence of environmental state change on proposed diagnosis models,the models are trained by vibration data of gearboxes collected under certain environmental state,and then are tested by vibration data collected under another environmental state,the results show that proposed models are robust to the change of running speed,frequency response function and other environmental state.