Surface morphology segmentation and evaluation of diamond lapping pad based on improved Mask R-CNN

Objectives:Diamond lapping pads have significant characteristics such as fast cutting speed, high machining accuracy, wide application range, and certain self-sharpening. They are widely used in the grinding of various hard and brittle materials. However, during processing, the lapping pad is prone to wear, and the surface morphology changes after wear. Its surface morphology has a direct impact on the quality of the workpiece and the lapping performance of the lapping pad. This article explores the use of deep learning methods to accurately and efficiently detect the surface morphology of the lapping pad, and evaluates the effectiveness of the detection method.


Methods:An improved Mask R-CNN model segmentation method is proposed based on the characteristic of improving perceptual field of view through dilated convolution, which identifies and segments the diamond abrasives and pores in the surface images of the lapping pad. The model is trained and verified using a dataset of diamond lapping pads with magnesium oxychloride binder after grinding sapphire; In order to verify the difference between the diamond abrasives and pores segmented using this method and the actual results, three parameters of target number recognition accuracy, target segmentation area accuracy and target position error are proposed to evaluate the segmentation effect.


Results:Through training and verification of the improved Mask R-CNN segmentation model, the results show that this method can realize the recognition and segmentation of diamond abrasives and pores in the surface images of the lapping pad, and the average accuracy is 78.2 %. By comparing the segmentation results of the model with the manually annotated results, and calculating the three segmentation evaluation parameters, the results show that the number of diamond abrasives and pores obtained by the improved Mask R-CNN model recognition segmentation method does not differ significantly from the actual number of diamond abrasives and pores. However, due to the complex background of the surface image of the lapping pads and the unclear contrast between abrasive particles, pores and binder, there are certain missed or false detections when using this method to detect the surface morphology of the lapping pads, the recognition accuracy of the number of diamond abrasives and pores is 82.1 % and 93.4 %, respectively; the method has a good segmentation effect on the identified targets, with small differences between the segmented abrasive and pore areas and the actual areas, and a high degree of agreement, the segmentation area accuracy for diamond abrasives and pores is 89.9 % and 95.3 %, respectively; the contour of the diamond abrasives and pores obtained by this method is slightly different from the actual contour, but the centroid position error is small, the position error for diamond abrasives and pores is 3.8 % and 2.8 %, respectively.


Conclusions:The comparison between the improved Mask R-CNN model segmentation image and the manually annotated image, and the calculation of three evaluation parameters, fully demonstrate that the use of the improved Mask R-CNN segmentation model has a good effect on the segmentation of diamond abrasives and pores on the surface of lapping pads, proving the effectiveness of the segmentation method.