Drill Bit Dull Grading Automated Damage Assessment Using U-Net Semantic Segmentation

Abstract The drill bit is the bottom hole assembly (BHA) component that destroys the rock in a drilling operation. This action exposes the drill bit to high wear. Once the bit run finishes, the drill bit is pulled out to the surface and dull graded. The drill bit is evaluated using the International Association of Drilling Contractors (IADC) dull grading standard; however, the visual inspection is usually performed by untrained personnel and is highly exposed to subjectivity. As a result, the drill bit dull grades are frequently inconsistent and unreliable. The correct drill bit selection is a critical factor in successful techno-economical operation. The dull grading provides essential information for improving drill bit selection, design, and BHA optimization. With the recent advances in computer vision technologies, a model can be trained to perform automated dull grading. Automation will expand the drill bits specialists’ expertise into a model providing consistency and more valuable data. To automate the dull grading assessment, it is necessary to segment the cutters and then perform area and contour measurements. Basic image processing techniques such as thresholding are not sufficient to be able to isolate the cutter. The task requires more advance and complex deep-leaning methodologies to train a model capable of segmenting the cutter from the rest of the image. Semantic segmentation can be performed to train a model capable of identifying and segmenting the cutters. Then to quantify the cutter damage and assign a dull grade, it is required to measure the area and contour. A semantic segmentation model based on U-Net was trained using a dataset consisting of multiple lateral blade images of used drill bits from multiple wells. The model exhibited an accuracy greater than 90% in detecting the cutters and isolating them from the rest of the image. A border (boundary) class was added to the cutter and background classes to improve segmentation. Then, OpenCV segmentation functionalities were used to measure the area and contour of the worn cutter, assigning a respective dull grade. This paper presents a semantic segmentation model using U-Net to segment all the cutter faces combined with OpenCV functionalities to measure the area and contour the cutters and automatically grades each cutter individually. Automated dull grade assessment eliminates subjectivity in the process, providing a better and more consistent evaluation to improve bit design and selection.