Accurately estimating the directional tendency of a drill string is important to ensure tracking of the planned trajectory and to prevent over/undershooting targets. Predicting this tendency is often achieved through calibration of a physical model of wellbore propagation using previous directional drilling data. However, the amount of relevant prior information to use and the uncertainty in future predictions is not always clear.

The goal of the work reported here was to determine the amount of prior directional drilling data to use for calibration, and the frequency at which recalibration should occur to ensure accurate predictions of wellbore propagation. A procedure was designed to provide this information for a given wellbore propagation model in a given field. The particular model used in this study is a quaternion-based wellbore propagation model that describes the motion of the bit as a rigid body in space which moves based on fixed angular velocities and drilling inputs given in the form of a traditional slide sheet. However, the procedure can be used for any wellbore propagation model that requires calibration on prior directional drilling data.

The procedure consists of two tests. The first test (hold-out test) shows how the prediction error varies over an increasing amount of calibration data while holding the testing-set size constant. The second test (prediction test) shows how the prediction error varies over an increasing testing-set size while holding the amount of calibration data constant. These variations in prediction error can then be examined by looking at the spread of data over these tests.

The test methodology was applied on field data obtained from North American wells where bent sub motors were used for deviation control. The first test showed diminishing returns when incorporating more actions into the model calibration. In cases where bottom-hole assembly (BHA) changes were observed, calibrating on a larger number of prior actions slightly increased the test error. The second test allowed us to approximate the error propagation rate of our model predictions in the current field. The procedure described in this paper allows for such information to be quantified and visualized easily to improve the directional driller's understanding of the needs and reliability of the wellbore propagation model being used.

You can access this article if you purchase or spend a download.