Research On Image Recognition And Roughness Of Gear Surface In Honing

Gears are essential mechanical parts in automobiles,aerospace vehicles,robots and other equipment.Gear machining accuracy and microtopography of gear surface directly affect the performance,level and reliability of these equipment and hosts.Power gear honing technology,as a new finishing method,can effectively improve the gear tooth surface roughness,gear accuracy and fatigue resistance after honing,and improve the equipment transmission accuracy and noise.Under different process conditions,the micro-geometry generated by irregular super-hard abrasive micro-edge honing randomly distributed on the tooth surface of honed wheels is an important influencing factor on the roughness of the honed tooth surface.Based on the meshing principle,the paper deduces the equation for forming the tooth surface of the honed workpiece,establishes the roughness prediction model under different machining processes,analyzes the factors affecting the surface roughness after honing,builds a test platform,designs orthogonal experiments and collects surface roughness images.Using convolutional neural network to recognize and detect the image of tooth surface roughness after honing,analyze the effect of network model under different learning rates,realize the prediction of tooth surface roughness after honing and detection of tooth surface roughness after honing based on image recognition,and provide a new method to control the quality of tooth surface of honed tooth processing.The main research is as follows.（1）Establish the motion coordinate system and fixed coordinate system of honing wheel and workpiece gear,determine the coordinate conversion formula between coordinate systems,derive the spiral involute equation and contact line equation of the workpiece tooth surface after honing,and establish the theoretical formula of tooth surface roughness after honing based on meshing theory according to the classical theory of grinding.The regression analysis method was used to establish the prediction model of tooth surface roughness after honing with tooth surface roughness as the target and three processing parameters of honing wheel speed,axial feed speed and radial feed as variables.（2）To design the honing experiment by orthogonal test method,collect the images of tooth surface roughness after honing under different machining parameters and establish the corresponding data set.The least squares method was used to determine the roughness prediction equation based on the machining parameters,and the reliability of the prediction model was proved by comparing the real value and the predicted value.Combined with the experimental results and meshing principle,the effect of three machining parameters on roughness was studied by using three-factor ANOVA and extreme difference analysis.The results show that the honing wheel speed n has a large influence on the roughness and is negatively correlated,while the axial feed speed1)and radial feed1)have a small influence on the roughness and are positively correlated.（3）An efficientnet model based on convolutional neural network is established to detect the roughness of honed tooth surface,and the efficientnet model is trained,detected and analyzed under different learning rates.It is experimentally demonstrated that the detection results based on the efficientnet model are basically consistent with the results of the roughness prediction model based on machining parameters,and the training recognition accuracy is 95.8%at the learning rate of 0.001,and the experimental verification accuracy is 93.3%.