Maximizing the value of information from high-frequency downhole dynamics data
Downhole drilling dynamics are poorly understood. Neither models nor experiments seem capable of fully describing the movements and forces of the drillstring during drilling. Downhole measurements could potentially hold the key to those missing insights, however data is not yet used to its full potential. This work addresses the barriers to obtaining value from downhole dynamics data and offers solutions to overcome them. A novel kinematic model was developed that fully accounts for sensor position and measurement design. It supports the hypothesis that lateral vibrations cause high-frequency fluctuations of tangential accelerations. Hence, against currently prevailing scientific opinion, “high-frequency torsional oscillations” (HFTO) are not actually a torsional phenomenon, but the consequence of a lateral vibration. A downhole measurement tool under off-center rotation captures particular high-frequency data patterns that can be considered a sensor artifact. If ignored, these artifacts can impact the calculations of RPM and other derived measurements from downhole data. An extensive set of downhole data was analyzed to improve downhole dynamics data collection schemes for detecting drilling dysfunctions. For each prominent type of dysfunction, minimum data collection frequencies are specified. Such guidelines assist in collecting downhole data at sampling rates that are high enough to draw meaningful conclusions, but low enough to not flood limited available bandwidth and memory capacities. Even though a sensor is set up to measure only a single parameter along a single axis, it captures a variety of downhole events, which may lead to misinterpretations. These events can still be differentiated based on their typical frequency ranges. It is further shown how ‘noisy’ frequency ranges can be detected and selectively removed by combining multiple downhole measurements. A lack of transparency and inefficient processes around sensor design, data collection, processing, and transfer cause misinterpretation and under-utilization of drilling downhole data. A review of tool design and sensor identifies sources of bad data quality. Eventually, defined data quality requirements will offer sustainable sensor data improvement. To work with downhole data generated under current circumstances, data processing techniques are developed and demonstrated. Algorithms that combine data, drilling processes, and physics automatically correct sensor errors. Further, a machine learning approach for automated vibration classification based on patterns is developed. A standardized structure to transfer downhole data from the service provider to the end user is suggested. The structure does not only define how the data should be shared, but also what additional data (metadata) is required. Specifications of such informational requirements improve transparency and comparability of measurements. Therefore, the proposed data format is a prerequisite for automated drilling data analysis.