LAUSR

Abstract The direct detection method of tool wear using industrial cameras is an important guarantee for product quality evaluation and optimization of machining parameters. Due to the limitations of camera resolution and the cost of machine vision systems, researchers use zoom lenses to achieve high-precision monitoring of tool wear in small fields-of-view. This leads to an increase in the installation cost of machine vision systems and the time for tool wear measurement. Regarding the issue above, the aim of this paper was to implement on-line milling cutter wear monitoring in a wide field-of-view camera. We perform tool wear monitoring through “location, segmentation and measurement” of tool wear area. Firstly, homomorphic filtering is used to preprocess tool images to enhance the wear edges. Then, we use histogram contrast to define saliency values for image pixels to locate the tool wear region. Secondly, a GrabCut model based on the visual attention mechanism was designed for precise segmentation of tool wear area, with a strong anti-noise advantage. Finally, the main cutting edge was fitted by the least square method to measure the flank wear. In the experiment of accelerating milling cutter life and different failure phenomena, the average location accuracy of the tool wear area by combining homomorphic filtering and histogram contrast is 97.06%. Compared with watershed segmentation and OTSU method, the F1-measure, precision and recall of the improved GrabCut model for tool wear region segmentation is 95.28%, 96% and 94.73%, respectively. The estimated accuracy of the flank wear based on the least square method linear fitting is more than 91%.