Learning to love computer vision

In oil and gas exploration, the drill bit industry constantly innovates, driven by the challenges of increasingly complex and difficult wells. For decades, drill bit product development has aimed to improve drilling performance, utilising iterative design changes to drive incremental gains in performance metrics such as rate of penetration (ROP). The pursuit of technological excellence has led to further refinement of methods. This article delves into the cutting-edge use of computer vision for polycrystalline diamond compact (PDC) cutter damage classification, demonstrating how advanced image analysis is transforming the field.

The importance of high-quality images

The process of PDC cutter damage classification starts with something seemingly simple yet crucial: taking high-quality images. However, it is not enough to just take good pictures in the traditional sense. Taurex Drill Bits established a process to take high-quality photos akin to rig dull photos and set out to establish the largest dataset of high-quality PDC cutter photographs. This massive dataset serves as the foundation for advanced analysis and ma-chine learning applications. High-quality images are the backbone of accurate cutter damage classification. The clarity and detail provided by these images are essential for both human analysts and machine learning algorithms to make precise assessments of cutter wear and damage. Traditional rig photos often lack the resolution and consistency needed for detailed analysis, which can lead to subjective evaluations and inconsistent results. Figure 1 shows a comparison between high-fidelity images captured by BitVision technology and standard rig photos. BitVision photos are clear and detailed, providing a basis for both human and machine analysis.

The challenges of PDC cutter damage classification

The key issues of cutter damage classification are deeply rooted in the classic tropes of PDC dull analysis: consistency and time investment. Traditionally, grading each cutter is a time-consuming process that requires meticulous attention to detail. The challenge is compounded when consistency is also a requisite. Human evaluators, despite expertise, can introduce variability in assessments, leading to inconsistencies that affect the reliability of the analysis. Moreover, the time investment required to grade every single cutter ac-curately can be substantial, delaying the feedback loop necessary for rapid design and operational improvements. One of the primary challenges in PDC cutter damage classification is the variability in human assessments. Different evaluators might grade the same cut-ter differently based on experience and perception, leading to inconsistencies in the data. This subjectivity makes it difficult to establish reliable benchmarks for cutter performance and wear patterns. Another significant challenge is the time required to grade each cutter. The traditional process involves visually inspecting each cutter, identifying dam-age modes, and recording the findings manually. This labour-intensive pro-cess is not scalable, especially when dealing with large datasets or aiming for high-frequency analysis. The delays in obtaining and processing this information can hinder timely decision-making and slow down the development of new cutter designs.

A new era of cutter analysis

BitVision technology has brought a significant advancement in cutter analysis. The technology captures high-fidelity photos of the entire bit, with each cut-ter photographed separately. This provides high-quality images for both human analysis and machine learning (ML) models. The ability to examine the damage occurring on a specific cutter, or group of cutters, helps to under-stand the type of wear that is occurring. This insight allows for the selection of cutters with different wear attributes for optimal placement in the bit. The high-resolution images enable detailed inspection of each cutter. Analysts can zoom in on specific areas to identify subtle wear patterns that might be missed with traditional photos. This level of detail is crucial for under-standing the mechanisms of cutter wear and for making informed decisions about cutter placement, design modifications, and operational changes to limit drilling dysfunction. In addition to enhancing human analysis, the imag-es provide the necessary data for training machine learning models. These models can learn to recognise different types of cutter damage, such as chip-ping, wear, and fractures, based on the detailed visual information captured. This automated approach to damage classification not only increases accuracy but also speeds up the analysis process significantly.

Leveraging machine learning for advanced damage classification

Using this ever-growing dataset, experts on PDC damage began to use computer vision tools to segment and label images of cutters to denote the dam-age modes present. These labelled images form the basis for the PDC damage mode machine learning models that are being actively used to improve feedback loops in applications engineering and cutter development. The application of machine learning in cutter damage classification represents a significant leap forward in the field. Machine learning algorithms, particularly convolutional neural networks (CNNs), have proven to be highly effective in image recognition tasks. By training these models on a large dataset of la-belled cutter images, the process of damage classification can be automated, making it faster and more consistent than manual grading. To effectively classify cutter damage, sophisticated machine learning models are employed, particularly focusing on CNNs, which are highly effective in image recognition tasks.

To access the full version of this article and get a free trial subscription to Oilfield Technology, sign up here!