Abstract
Bad hole cleaning can lead to severe drilling incidents such as stuck pipes, pack-offs, losses, well instabilities, and gas kicks. This paper presents a non-intrusive hybrid approach to detecting bad hole cleaning. In the hybrid approach, we combine a physics-based model solving multi-phase flow equations for cuttings transport with advanced data-driven and machine learning models in a non-intrusive way. We also demonstrate how data-driven modeling can enhance the accuracy of a physic-based model and vice versa in the context of drilling operations with actual data. We also demonstrate how the hybrid approach can be employed to integrate new data sources from the drilling process, which the pure physics-based models were not designed to incorporate. In particular, we studied the use of pressure sensors in the annulus along the drill string. These sensors are uniquely well suited to detect drilling problems such as bad hole cleaning, as they provide continuous data along a spatial dimension. This is a significant advantage over the sensors on the topside or the bottom-hole assembly. However, bad hole cleaning has a complicated and noisy signature with significant spatio-temporal variations. It is therefore, necessary to integrate the best of data analysis, machine learning and traditional modelling, to deliver a hybrid modelling framework (in this case — a non intrusive hybrid method) that can simulate the real-world hole cleaning drilling operations more accurately and enable better decision making. The results of the hybrid method application on two challenging drilling operation cases are presented here that demonstrates its potential, and conclusions are drawn for further improvement of this hybrid methodology.
