Detection of poor hole cleaning and stuck pipe

Our algorithm for early detection of poor hole cleaning and stuck pipe [5] use the “noise” in the real-time measurements to produce an early warning of these drilling problems. Similar to how the change in noise from a car can alert the driver to a mechanical problem, drilling problems such as stuck pipe can be diagnosed from “anomalies” in the real time data. A test on recorded data shows that the method can detect problems hours in advance, in time for the driller to rectify the situation. The algorithm is further shown to be more robust to bad data quality than similar noise analysis techniques The second method to be implemented was a method for predicting the pressure in the well. For the driller, this is a crucial parameter to estimate. The code can be found in STUCKPIPE_SITECOM.ZIP.

Pressure prediction during MPD operations
For the prediction of the pressure we use a fast and simple model suitable for integration into an MPD control system. We then tune this model using a so-called Ensemble Kalman Filter. Being a technique that has previously been employed in reservoir modelling, the Ensemble Kalman Filter allows us to take multiple possible scenarios into account simultaneously and manipulate the likely range of parameters and their uncertainties rather than give a precise but possibly wrong input to the prediction. The code can be found in ENKF_SITECOM.ZIP.

The two algorithms have been implemented as stand-alone code in Matlab, but also communicate with Kongsberg’s SiteCom system when accessing real-time drilling data. Being stand-alone code, it can be used and developed further by any IO Center industry partner, without being tied to any particular supplier. The optional link to the SiteCom system means the software is ready out-of-the-box for testing in live operations, by any IO Center industry partner which happens to use the SiteCom system. The accompanying report (Report Problem Detection.pdf) provides a technical manual on the use of the above implementations, as well as explaining the algorithms for anyone wishing to make their own implementations.