Drillstring Washout Diagnosis using Friction Estimation and Statistical Change Detection

Abstract

In oil and gas drilling, corrosion or tensile stress can give small holes in the drillstring, which can cause leakage and prevent sufficient flow of drilling fluid. If such \emph{washout} remains undetected and develops, the consequence can be a complete twist-off of the drillstring. Aiming at early washout diagnosis, this paper employs an adaptive observer to estimate friction parameters in the nonlinear process. Non-Gaussian noise is a nuisance in the parameter estimates, and dedicated generalized likelihood tests are developed to make efficient washout detection with the multivariate $t$-distribution encountered in data. Change detection methods are developed using logged sensor data from a horizontal 1400 m managed pressure drilling test rig. Detection scheme design is conducted using probabilities for false alarm and detection to determine thresholds in hypothesis tests. A multivariate approach is demonstrated to have superior diagnostic properties and is able to diagnose a washout at very low levels. The paper demonstrates the feasibility of fault diagnosis technology in oil and gas drilling.