During well construction operations, when the drill cuttings are not transported efficiently out of the well, or if parts of the wellbore collapse on the drillstring because of an unstable formation, the well can begin to pack off. There are a lot of models to estimate the volume of drilling cuttings. But few literatures can be found for volume estimation of cavings from stress-induced wellbore damage.

Caving volume is predicted using a geomechanical caving model. Assuming the caving debris has the shape of a prism, caving volume can be estimated as a function of wellbore radius, caving depth, caving length, and the breakout angle. Breakout angle is estimated as a function of rock strength, geostress and wellbore pressure. The effects of wellbore inclination and azimuth are also taken into account for deviated wells. Caving depth is typically obtained from the caliper log, which is usually not available during drilling operations. Therefore, hybrid engineering and data-driven approach are employed to predict the caving depth.

Different data-driven models (e.g., Random Forest) were developed and tested for predicting caving depth h, assuming it is a function of rock geomechanical properties, geo-stresses, pore pressure, wellbore pressure, and gamma ray log. Caliper data log data of multiple wells are used to train and cross-validate those data-driven models. Cross-validation results indicate that three data-driven models including random forest, neural network, and Adaboost are performing very well.

The simulation method can be applied for real-time monitoring of pack-off risk during drilling operations. In the workflow, real-time survey data (measured depth, inclination, azimuth) and rig-state data, including measured depth (MD), gamma ray (GR), flow rate, rotational speed (RPM), rate of penetration (ROP), weight on bit (WOB), etc., are continuously received by the computer and buffered in caches. These buffered data are then consumed by the geomechanical caving width model and the data-driven caving depth model. As a result, real-time caving volume is calculated at every depth of interest.

Combing both real-time cuttings data and real-time caving data, a real-time packoff volume percent can be obtained and displayed in a continuously updating plot. The approach is based on synthesis of an engineering model and a data-driven model.

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