Deep Learning-Aided Dull Grading Scheme for Drill Bits

Abstract

Abstract Drill bit design plays a key role in drilling operations. A suitable choice of the drill bit parameters (such as the profile shape, the orientation, and the distribution of cutters, etc.) will significantly improve the durability and stability of the drilling process, increase the rate of penetration (ROP) and enhance the drilling efficiency. Traditional strategies of the drill bit design usually involve a trial-and-error process to adjust the parameters, which is usually time-consuming and contains an element of subjectivity. Alternatively, machine learning techniques provide powerful tools to make full use of the collected data, and analyze the relation between the design parameters and the performance of the drill bits on specified rock mechanics data. In the present work, we aim at building a machine learning-aided scheme to help design the parameters of the drill bits. A machine learning-aided scheme is proposed to perform the drill bit design. First, a dull grading system is established to evaluate the wear extent of the drill bits from images. The grading system includes a cutter localization model to separate the cutters from drill bit images, a reconstruction model to regenerate the 3D geometries of the cutters from several 2D images, a classifier to evaluate the wear extent of the cutters, and an aggregative algorithm to grade the worn drill bits. The grading system is then applied to build the data set of the worn drill bits. Finally, an individual deep learning model is adopted to learn the relation between the design parameters and the wear extent, and a sensitivity analysis is performed to optimize the design parameters. The research results showed that: (1) the proposed machine learning-aided scheme could produce instructive guides on the drill bit design, reduce the analysis cost and provide accurate prediction on the impact of the design parameters. (2) The 3D reconstruction algorithm could efficiently extract the geometry feature of the drill bits from 2D data, and provide detailed geometry parameters of each cutter. The performance of the 3D reconstruction model is mainly related to the number of corresponding 2D images of a given drill bit and the coincidence degree among them. (3) The classifier based on convolutional neural networks (CNN) could provide accurate wear extent assessments of drill bits. In this paper a novel machine learning-aided drill bit design scheme is proposed to effectively enhance the efficiency of the evaluation and optimization of the design parameters. A 3D reconstruction method is established to extract the 3D geometry parameters of the drill bits from several 2D images. The deep learning-based dull grading model is proposed to accurately evaluate the wear extent of the drill bits.