The artificial neural networks in machine learning are designed to mimic the way the brain learns.
Machine learning for improved drilling
17.04.2018Since 2016 IRIS Drilling and Well modelling group have been researching on how machine learning methods may be applied for management of drilling operations.
Chief Scientist in the Drilling and Well Modelling group at IRIS, Fionn Iversen explains:
“Our philosophy is that machine learning methods should be built on physical understanding to achieve as powerful models as possible. We therefore seek to build machine learning models through use of both state of the art process models and expert input in the learning process. The ongoing work is to determine applicable machine learning techniques for such development, which shall result in improved applications for use in drilling process management.”
Results presented in Fort Worth
Machine learning techniques, based on the so called hidden Markov models, have been used for the reconstruction of pipe movement utilizing triaxial accelerometer measurements. The results have been presented at the SPE Drilling Conference in Fort Worth, 6-8 March, 2018 in the paper SPE-189618-MS Reconstruction of Pipe Displacement Based on High-Frequency Triaxial Accelerometer Measurements
. This work has been undertaken as part of the project P1.3 Drilling Process Optimisation of the DrillWell centre.
Chief Scientist, Fionn Iversen.
Automatic Drilling Problem Classification
Work is ongoing on revision of KPN application on Automatic Drilling Problem Classification for September PETROMAKS 2 call. The project will combine machine learning methods with predictive physics-based drilling process models with parameter normalization for early diagnostics of drilling problems. One of the main challenges in the project is limited data sets, both with regards to process description and numbers of relevant and available data-sets. This problem is to be addressed through generating synthetic datasets and development of machine learning models and methods applicable for limited and sparse datasets.
Celle Drilling Conference
Two abstracts were submitted to Celle Drilling Conference in March, both related to learning methods, with the titles Automatic detection of anomalous drilling operations using machine learning methods
and also a paper on statistical learning methods for ROP optimization, which have been developed in a Sekal project with Statoil as funding partner. The paper on automatic detection of anomalous drilling operations will be studying detection of downlinking signals for directional control as a case.
Geosteering for IOR
There is also an ongoing effort related to the project Geosteering for IOR, studying use of machine learning methods for prediction of geology around the bit based on resistivity measurement. This work is to result in future publication.