Machine-Learned Adherence Curves From Drilling GAV and PVT Data

The patent introduces an intermediary processing layer that bridges raw GAV data and final adherence curves. This intermediary layer includes quality control procedures, data transformation routines, and machine learning models that act as mediators between the complex raw data and the interpretative curves, thereby improving reliability while managing system complexity through structured abstraction.

The patent replaces manual, subjective analysis methods with automated machine learning algorithms. The mechanical system of manual data interpretation is substituted with computational models (Kernel Ridge regression, B-Spline, Gradient Boosting) that objectively process GAV data, enhancing reliability while the automation manages complexity through standardized algorithms rather than ad-hoc analysis.

The patent implements preliminary quality control actions before the main analysis. Data is pre-processed to identify and remove outliers, check for missing values, and validate data quality in advance. This preliminary action ensures that the subsequent machine learning analysis works with clean data, improving accuracy while the pre-processing steps are optimized to minimize time consumption.

The patent transforms data parameters during processing, converting raw GAV measurements into standardized formats suitable for machine learning. Data transformation routines adjust parameter scales, normalize distributions, and create feature representations that enhance model accuracy. These parameter changes are applied efficiently through automated routines that balance precision improvements with processing speed.

The patent converts the harmful effect of drill bit metamorphism into a beneficial signal. Instead of discarding GAV data contaminated by DBM, the machine learning models are trained to recognize and utilize these metamorphic signatures as indicators of reservoir contact. The harmful thermal and mechanical effects at the drill bit are transformed into useful data features that enhance reliability when properly interpreted through the adherence curve models.

The patent implements feedback mechanisms where adherence curve models continuously refine their predictions based on the relationship between GAV data, PVT samples, and drilling parameters. The models learn from the feedback loop between predicted and actual reservoir conditions, adjusting their interpretation of metamorphic effects. This feedback process enables the system to distinguish between metamorphic artifacts and true reservoir signals, improving reliability while maintaining productivity.