Role of Machine Learning (ML) in Drilling Fluid Design
1. Drilling fluid essentially relates either directly or indirectly to ‘the problems or solutions’ of wellbore instability; lost circulation; kicks and blowouts; ‘stuck pipes’; and so on. Now, upon introducing ML approaches (supervised learning: regression and classification; or, unsupervised learning: clustering; or, reinforcement learning); whether, these computer algorithms would be able to automatically perform ‘the required tasks’ ‘through experience’ and ‘by the use of data’ such that ML ‘learns the patterns and inference’ from the data; and in turn, would be able to make the ‘drilling fluid’ to remain as an ‘architect’, which in turn, could make ‘the drilling operation’ either ‘materialize’ or ‘unrealizable’?
2. Although, drilling fluids might represent only 5-15% of drilling costs, it may however cause 100% of drilling problems. If so, how easily could we smooth drilling, mud and log data? Won’t the output be ‘drastically’ varying depending upon the choices such as (a) using domain knowledge of drilling fluid design; or, (b) using principal component analysis (which can condense multidimensional input variables to lower dimensional input variables); or, (c) using correlations between input and output parameters?
Is there a physical reasoning based on which, we split (60/40; or, 70/30; or, 80/20) the ‘processed data’ into ‘training data’ and ‘test data’?
Leaving aside ‘no free lunch’ theorem, with imbalanced data, whether the ‘data manipulation’ technique would remain to be expert-dependent?
3. Would it remain feasible to design and adopt a ‘drilling fluid technology’ with ‘intelligent features’ such as ‘self-identification’; ‘self-tuning’; and ‘self-adaptation’ that would remain conducive to fundamentally solving various technical problems of drilling fluids; and thereby, enabling the drilling fluid to automatically identify the complex downhole environment; and to adjust the performance of the drilling fluid automatically?
With intelligent drilling fluids;
(a) Will we be able to achieve successful drilling in difficult drilling areas such as shale gas and gas hydrate bearing sediment formations?
(b) Will we be able to minimize the cost of treating agents?
(c) Will we be able to facilitate the sustainable use of drilling fluids?
(d) Will we be able to initially identify changes in the external environment of the fluid such as downhole pressure and temperature?
(e) Will we be able to initially identify changes in the internal environment of the fluid such as pH value and salinity?
(f) Will we be able to adjust the physical and chemical properties such as density, rheology and emulsification type – through such intelligent additives?
Can ML effectively be used to address all the above queries?
- Badges
- Science topic
- Similar topics
- Biological Science
- Bioecology
- Systems Ecology
- Landscape Ecology
- Connectivity