Application Research On End Mill Breakage Detection Based On Mask R-CNN And Multi-level Feature Fusion

Damage is one of the main forms of end mill failure during the milling process.It will not only make the end mill lose its milling ability,but also damage the processing process and product quality,serious when even will cause the workpiece scrap,damage to the machine tool,and the whole process to stop.Therefore,it is of great significance to obtain the damage state of the end milling cutter in time to improve production efficiency and product quality.In this paper,with the help of image processing and deep learning methods,the Mask RCNN network is researched and improved,and its application in the image damage segmentation detection of end mills is explored.Based on the target detection model,the network introduces a new mask branch to achieve target segmentation,and the integration of target detection and segmentation is realized by sharing convolution features.The main work of this paper is as follows:(1)The data set of breakage end milling cutters is constructed.In order to preserve the damage characteristics of the end mill as much as possible,a 5 million pixels industrial camera was used to complete the damage image acquisition according to the designed end mill image acquisition scheme,and the side window filter technology was used to preprocess the image for noise reduction.In order to enhance the robustness and generalization performance of the model,data amplification was realized by means of random rotation and brightness change,etc.Finished labeling the breakage end mill data set,and analyzed in detail the working principle and training strategy of Mask R-CNN.(2)Mask R-CNN was used to segment and detect the end milling cutter damage.Based on the Py Torch deep learning framework,the Res Net-50 network was established as the backbone network for end milling cutter damage feature extraction.The experimental results show that the model has 79.39% and 89.31% accuracy rates for the detection of chipping and breakage respectively.Through analysis,it is found that,Although Mask R-CNN can achieve breakage end mill detection,it has the phenomenon of missing detection and misdetection for the small size and close interval of chipped edge and part of breakage cutter.(3)Improved Mask R-CNN.In order to further improve the segmentation accuracy of end mill breakage detection,an improved Mask R-CNN model is proposed.Based on the combination of multi-level feature fusion,a bottom-up path is added to extract smaller-scale chipping features.The linear weighting method is used to improve the original NMS algorithm,and some correct bounding boxes are retained to reduce the missed detection rate of chipping.The experimental results of the model before and after improvement show that the accuracy of the improved Mask R-CNN model on the chipping detection is 18.76% higher than that before improvement,and the number of broken detections and missed detections of fragmentation is also significantly reduced.The m AP of the improved Mask R-CNN model is 88.32%,which is4.06% higher than that of the previous algorithm,which fully indicates that the improved model is helpful to improve the detection performance.In order to display the experimental results of network training more conveniently and concisely,this paper uses the Qt framework to write an interactive interface for end mill damage segmentation detection.