Applications of Computer Vision and Deep Learning in Visual Features Extraction of Drill Bits

Abstract Drill bit dull grading provides basic information for bit performance evaluation, drilling parameters optimization and BHA optimization. Traditional dull grading standard from IADC depends on visual inspection, which is subjective, low accuracy & efficiency and of lower information dimension. Computer vision is one of the most mature applications of artificial intelligence. In petroleum engineering, researchers started to use image processing to evaluate bit damage. But there are technical challenges for wider usage of this technology due to accuracy of object detection and image classification in complex environments, integrative description of 3D objects, various bit profiles, cutter structures and dull shapes, etc. More than 3000 images of bits before and after drilling are collected and pre-processed for the research. A workflow to detect objects (bit, blades, cutters, nozzles, etc.) in images are designed, and the performance of several front-edge algorithms are compared, such as Faster RCNN, Yolo, etc. and several methods are used to improve detection performance in complex light environment. Multi view images fusion was conducted for an integrative view of bit picture for integrative bit dull grading report. Structured light camera is tested to improve the accuracy of 3D cutting features. It's shown that the accuracy of object detection in wellsite environment is over 90% and Faster RCNN performs with more robustness. It's possible to generate 3D structures with multi view images or videos and the image resolution is critical to the definition of 3D images. Compared with common camera in mobile phone, structured light imaging features with higher accuracy in 3D structure mapping. The results helped to push the application of computer vision and AI in PDC bits evaluation at the well sites.