Affiliation:
1. The University of Texas at Austin
Abstract
Abstract
New high-frequency downhole recorders and high-bandwidth real-time data transmission tools (such as wired drillpipe) are heralding the era of big data in drilling. Nevertheless, high-frequency data is not yet used to its full potential, as the industry is only just beginning to make sense out of the many gigabytes of recorded data. Analysis of high-frequency data appears to be particularly useful to better characterize and understand vibration events, which are prominent technical limiters of drilling performance. The wealth of information provided by high-frequency vibration patterns, which are not present in low-frequency surface data, offers the possibility to significantly improve vibration mitigation methods. This, in turn, provides opportunities for step-changes in drilling performance improvement.
A simple kinematic model was developed to study the expected high-frequency acceleration measurement output under whirl and stick-slip vibrational dysfunctions. The vibration patterns predicted by the model matched the patterns observed in high-frequency field data very well, both in the time and frequency domain. The comparison reveals essential details of the downhole kinematics of vibrational modes, particularly regarding the nature of whirl, which may eventually lead to the development and application of novel whirl mitigation methods. Using the kinematic model, sensor measurements can be reproduced and thus fully understood, leading to sensor parameter and placement optimization. The match between modeled and field data also seems to show that the downhole high-frequency vibration patterns are apparently largely independent from specific operational parameters as well as borehole geometry. Moreover, vibration patterns appear to be more useful to classify and quantify vibration than absolute vibration parameter values.
Using the premise that vibration patterns hold the key to vibration identification and ultimately mitigation, a new methodology was developed and applied to extract and analyze essential features from high-frequency vibration data for identification purposes. Using a Bayesian approach with training supervised learning algorithms, vibrations were automatically classified with a success rate in excess of 90% when applied to high-frequency field datasets.
Cited by
13 articles.
订阅此论文施引文献
订阅此论文施引文献,注册后可以免费订阅5篇论文的施引文献,订阅后可以查看论文全部施引文献