Early Kick Detection for Deepwater Drilling: New Probabilistic Methods Applied in the Field

Abstract

Abstract This paper documents the development and field test of a sensitive new kick detection system. The detector uses a Bayesian probabilistic framework to make good decisions based upon noisy drilling data. It works on any rig type, but brings particular improvements to kick monitoring for deepwater drilling, especially in high heave conditions. The approach offers sensitivity improvements over existing systems, while maintaining a low false alarm rate. Introduction The early detection of kicks is increasingly important with the growth in deepwater drilling, where late detection can have serious safety and financial implications. A variety of systems have been developed to detect kicks in real time by using flow data, and other surface and downhole measurements. The simpler existing systems do not have the sensitivity required for deepwater applications, or require special surface or downhole sensors. More sophisticated systems have been built which predict the flow data using a complex model of the mud path. Such systems compare actual and predicted flow and also require careful calibration and sensitivity adjustment. We describe and present field results of a new kick detection system, which uses a Bayesian probabilistic model-based approach. Kicks of various types and rates are modeled explicitly as time series of flow data, and no thresholds are involved. The Bayesian approach leads to automatic sensitivity adjustment for noise, eliminating the need for careful calibration. In medium and high noise data, the new system can give a tenfold improvement in sensitivity compared with conventional approaches, for the same false alarm rate. Other drilling events which often cause false alarms (such as connections and pipe movement) can also be modeled explicitly, further reducing false alarms due to ambiguous data. The system can also be easily extended to address losses. The system has been deployed with good results on a number of rigs, both fixed and floating, conventional and slim-hole. Results from some of these will be shown, including a kick on a semi-submersible detected at less than 3 barrels in the presence of 25 barrel/minute peak-to-peak variation due to rig heave. Kick Detection History It is widely accepted in the literature that flow measurements give the most rapid indication of a kick [1]. There are many other indicators: for example, pit volume, gas levels, increased ROP (due to porous formation), mud density decrease (due to dissolved gas), and annular pressure rise. However, flow data give the most rapid unambiguous indication at surface. One problem with using flow data is the high level of noise in the measurements. Mud flow is often measured with crude sensors (pump strokes for flow in, and the flow paddle for flow out) which are particularly noisy. Also, the presence of heave on deepwater rigs has caused problems with existing kick detection systems.