Research And Application Of Quality Regression Algorithm For Precision Grinding Of Inner Ring Bore Of Spherical Roller Bearings

Currently,with the growing market demand for high-end bearings,there is an urgent need for bearing manufacturers to increase technological innovation,improve product quality and performance,and enhance independent innovation and core competitiveness.Compared with other types of bearings,spherical roller bearings are widely used in steel,paper,shipbuilding,electric power and other industries with the advantage of large radial conforming capacity.In the process of bearing processing,grinding is an important process link,which directly affects the processing quality of bearing products.In this paper,the correlation model between grinding process parameters and grinding quality is constructed for the semi-finish grinding process and finish grinding process of inner ring inner diameter grinding of spherical roller bearings,and the prediction model of surface quality of grinding process is completed.It provides an important theoretical guidance for improving the quality and processing efficiency of bearing grinding and reducing its manufacturing cost.In response to the problems of high processing quality inspection requirements and poor correlation between processes in the grinding process of spherical roller bearings,a solution was proposed to construct a correlation model between bearing processing surface quality and processing process parameters.The grinding test platform and machining surface quality inspection platform of spherical roller bearings were built;the semi-finish grinding process and finish grinding test of the inner diameter of bearing inner ring were carried out,and the machining surface roughness and roundness were detected,which provided an experimental system for building the prediction of the machining surface quality of bearings.In response to the lack of mathematical description of dynamic key factors in grinding processing and the difficulty of measuring dynamic factors,this paper uses the data of machining process parameters to construct a machine learning model to predict the grinding processing quality.The collected machining quality was visually analyzed,and each machining quality basically conformed to a normal distribution,and then the process-derived variables were constructed by subtracting the grinding wheel dressing speed before and after two times in the process parameters,and the XGBoost model was used for feature selection.In response to the limitations of a single prediction model,the Stacking model that incorporates SVR,RF,GBDT,Adaboost and BP neural network has better model performance.In the finish grinding process,the Stacking integrated model was constructed by comparing the machining quality after incorporating the semi-finish grinding process and verifying that the machining quality of the semi-finish grinding process has an effect on the surface roughness and roundness of the finish grinding process.For the problem that it is difficult to detect the machining quality of all ID grinding workpieces of spherical roller bearings,an active learning regression algorithm is designed to select more representative samples for the detection of surface roughness and roundness,and the four active learning algorithms and the EMCM algorithm integrated with Stacking are compared and experimentally verified,and the results show that the EMCM algorithm based on Stacking proposed in this paper has better results.The algorithm of EMCM has better results.In order to verify the effectiveness of the constructed spherical roller bearing inner ring bore quality prediction model and the designed active learning regression algorithm,test data of 20 spherical roller bearing inner ring bore semi-finishing grinding process and finishing grinding process were collected.In the semi-finish grinding process,the RMSEs of surface roughness and roundness predicted using the constructed stacking integrated model were 0.03985 μm and 0.2935 μm,and the MAEs were 0.03577 μm and 0.2617 μm.In the finish grinding process,the RMSEs of surface roughness and roundness predicted using the constructed stacking integrated model based on active learning were 0.03985 μm and 0.2935μm,and the MAEs were 0.03577 μm and 0.2617 μm.The RMSE of the active learning algorithm was 0.0367 μm and 0.2616 μm,and the MAE was 0.0289 μm and 0.2429 μm.The prediction performance of the stacked integrated model was better in the semi-finish and finish grinding processes.