Uncertainty propagation from vision-based tool wear measurements to cutting loads
Litsenziya: cc-by
Annotatsiya
This study presents a computational approach to predicting cutting loads while accounting for uncertainty in machine-vision-based tool-wear measurements during computer numerical control machining. The proposed methodology integrates image segmentation, quantitative assessment of tool flank wear, and mathematical modeling of cutting force and spindle torque as functions of tool wear. Three segmentation methods are considered: Otsu thresholding with morphological operations, Canny edge detection with morphological processing, and a convolutional neural network based on the U-Net architecture. The performance of these methods was evaluated under different image-quality conditions using pixel accuracy, intersection over union (IoU), the Dice coefficient, and relative wear-width measurement error. When the image-quality index decreased from 1.00 to 0.40, the IoU of the neural-network-based method decreased from 0.866 to 0.626, while the Dice coefficient decreased from 0.927 to 0.759. Within the considered computational model, increasing tool wear was accompanied by a 100.6% increase in radial cutting force and a 66.3% increase in spindle torque. The results demonstrate that high pixel accuracy alone is not a sufficient criterion for reliable quantitative tool-wear measurement. A methodological basis was established for assessing the propagation of uncertainty from tool-wear measurements to predicted mechanical loads. The obtained results may be used for further development of tool-condition monitoring systems and cutting-load prediction in CNC machining. Additional quantitative and experimental validation is required to confirm the practical applicability of the proposed approach.
Hali tarjima qilinmagan