O&G operators seek to reduce CAPEX by reducing unit development costs. In drilling operations this is achieved by reducing flat time and bit-on-bottom time. For the last five years, we have leveraged data generated by drilling operations and machine learning advancements in drilling operations. This work is focused on field test results using a real-time global Rate of Penetration (ROP) optimization solution, reducing lost time from sub-optimal ROPs. These tests were conducted on offshore drilling operations in West Africa and Malaysia, where live recommendations provided by the optimization software were implemented by the rig crews in order to test real-world efficacy for improving ROP. The test wells included near-vertical and highly deviated sections, as well as various formations, including claystones, sandstones, limestones and siltstones. The optimization system consisted of a model for estimating ROP, and an optimizer algorithm for generating drilling parameter values that maximize expected ROP, subject to constraints. The ROP estimation model was a deep neural network, using only surface parameters as inputs, and designed to maximize generalizability to new wells. The model was used out-of-the-box, with no specific retraining for the field testing.
During field-tests, increased average ROP was observed after following recommendations provided by the optimizer. Compared to offset wells, higher average ROP values were recorded. Furthermore, drilling was completed ahead of plan in both cases. In the Malaysian test well, following the software's advice yielded an increase in ROP from 10.4 to 31 m/h over a 136 m drilling interval. In the West Africa well, total depth was reached ∼24 days ahead of plan, and ∼2.4 days ahead of the expected technical limit. Importantly, the optimization system provided value in operations where auto-driller technologies were used. This work showcases field-test results and lessons learnt from using machine learning to optimize ROP in drilling operations. The final plug-and-play model improves cycle efficiency by eliminating model training before each well and allows instantaneous, real-time intervention. This deployable model is suitable to be utilized anytime, anywhere, with retraining being optional. As a result, minimizing the invisible lost time from sub-optimal ROP and reducing costs associated with on-bottom drilling for any well complexity and in any location is now part of the standard real-time operation solutions. This deployment of technology shows how further optimization of drilling time and reduction in well cost is achievable through utilization of real time data and machine learning.