Improved Drill Bit Forensics Through Hybrid Analysis of Smartphone-Captured Videos and Operational Time Series Data

Abstract Drill bit photos and videos can be used as primary data sources for bit forensic analysis, i.e., determination of the origins and causes of bit damage. However, photos and videos only provide information about the final state of the drill bit after it has been pulled to surface. There is often a high degree of uncertainty about how the damaged state was reached while the bit was deployed in the hole. The purpose of this work is to automate drill bit damage assessment using a video recording of the bit, which can be easily captured on a smartphone, and combining this with time series data of the drilling operation. This then allows a timeline of bit damage progression to be recreated. Through the combination of 3D mesh building and custom-trained damage classification models, an algorithm is proposed to accurately classify drill bit damage. It allows for precise capturing the cutter's shape and consistently recognizing the cutter's position through the different video frames. This information is then combined with time-dependent damage evaluation models making use of the time series data captured for the interval drilled by the drill bit in question to arrive at its damage timeline. This study therefore introduces an integration of two different techniques for conducting drill bit forensics, one using sophisticated image analysis using videos or photos, and the other using operational time-series data analysis. The video analysis provides a visual evaluation of cutter damage achieved while relying on a 3D point cloud, which enables precise identification of the shape of bit elements. The analysis of time series data helps to understand the underlying mechanism of cutter damage. The main focus regarding the root cause of bit damage is to correlate it with harmful drill string vibrations. The "hybrid" approach of combining the two techniques improves both the accuracy of bit dull grading assessment as well as the robustness of bit forensic analysis.