The main objective in this work was to develop and test automated pore pressure and wellbore stability predictions ahead of bit using a digital twin. The new automated real-time workflow, with three smart agents for pore pressure, mudweight and well pressure was tested on historical data and real-time drilling campaigns for one exploration well and several production wells in the North Sea and offshore Mid-Norway. A digital workflow is set up from pre-drill 3D basin scale pore pressure modelling using a stochastic Monte-Carlo approach, including an automated update of pressure prognosis while drilling using sonic or resistivity data. Additionally, the innovative approach will also reduce the uncertainty in the predicted mud-weight window ahead of the bit. Three new smart agents have been implemented in a real-time data platform for drilling; one for pore pressure prediction, one for mudweight real-time calculations and one for well pressure real-time predictions.
The new automated workflow has been tested on several wells, either in real-time, or on historical data as playback. Both pressure real-time updates using sonic log (for exploration well) and resistivity data (for production well) have been tested with very good results, especially when filters on the raw log data are used, removing artefacts in input data, effect of well inclinations, lithologies etc. In addition, the real-time Equivalent Circulating Density (ECD) prediction has been compared with downhole ECD measurements. Digital twin using playback data from a well, offshore Mid-Norway has been carried out with automated correction and smoothing of the calculated pore pressure from resistivity log. The new automated workflow will contribute to increase the efficiency of drilling operations subject to narrow pressure margins and uncertain reservoir pressure, by providing a quantitative and adaptive automated workflow to support the drillers/rig crews.
The new automated workflow is presented for both exploration and production wells and the workflow can contribute to more cost effective and safer operations. The methodology can in the future be combined with machine learning.
