The paper discusses possibilities, limitations and pitfalls in using advanced real time mathematical models during drilling operations, and in particular the real time calibration of such models vs. available sensor data. Examples from recent MPD operations in the North Sea, as supported by real-time model-based interpretation and automation are shown and discussed.
Real time mathematical models for drilling, including hydraulic and mechanical models at different levels of sophistication, are used to provide decision support and to automate parts of operations. This is often combined with managed pressure drilling (MPD) as the pressure margins are very small, operations are complex, and robust automation may help a lot to improve performance and robustness of the overall system.
An important part of successful use of real time mathematical models is robust calibration of models, in particularly when there is significant deviation between the model prediction and the measured data. Reasons for deviation include inaccurate data for fluid properties, string and well geometry, survey data, temperature profile etc. In addition real time data may be inaccurate or delayed, and the mathematical model itself may be inaccurate under the actual conditions. Whenever data deviate from model and are considered sufficiently reliable, the mathematical model should be adjusted to meet measurements; manually, automatically or a combination.
There are many pitfalls and challenges related to model calibration, especially when done automatically or from predefined procedures by an inexperienced operator, accordingly efforts to develop robust procedures or algorithms are important. The paper will contribute to this development by discussing challenges in detail, and experience from real operations will be described to illustrate the points. Focus will be on when and what parameters to be calibrated, rather than the calibration technique.
The work presented shows that models can add significant value to operations, but robust implementation and procedures are necessary to avoid the pitfalls, which a.o. include use of inaccurate configuration data, invalid real time data, misinterpretation and inferior action when model deviates from measured data.
The paper analyses use and calibration of advanced real time models for decision support and automation, and illustrates the discussion with field experience from a recent, successful North Sea operation. The paper is thus a useful reference and guidance for people working with model based decision support and automation, including experience and analysis not published earlier.