Neural Network Mineralogy Estimation from Borehole Logging
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for determining mineralogy in geological rock formations from borehole logging measurements are limited by the need for indirect measurements, which do not provide direct mineral composition data, leading to inaccuracies in estimating hydrocarbon content and drilling decisions.
Innovation Solution
Employing a trained artificial neural network to derive a mapping function that determines bulk mineral component concentrations from atomic element measurements, allowing for the reconstruction of elemental concentrations and characterization of geological formation parameters such as porosity and water saturation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If indirect measurements are used to determine mineralogy from borehole logging measurements, then downhole mineralogy determination becomes possible, but measurement precision deteriorates due to lack of direct mineral composition data
Solution Approach 1:
The patent uses machine learning models as an intermediary to bridge the gap between indirect borehole logging measurements and direct mineralogy determination. The model is trained on direct measurement data (XRD, infrared spectroscopy) and then applied to indirect borehole data, acting as a mediator that translates indirect measurements into accurate mineral composition estimates without requiring direct downhole spectroscopic capabilities
Solution Approach 2:
The patent creates a virtual copy of direct mineralogy measurement capabilities by training machine learning models on extensive datasets from direct measurement techniques. This virtual model replicates the accuracy of direct measurements (XRD, infrared spectroscopy) and enables downhole mineralogy determination with precision comparable to surface laboratory methods, effectively copying the functionality of direct measurement instruments in a downhole-compatible format
2Measurement precision
If direct measurement methods (XRD, infrared spectroscopy) are used, then mineralogy determination accuracy is improved, but downhole applicability is lost as these methods are not available downhole
Solution Approach 1:
The patent replaces the physical measurement mechanisms of XRD and infrared spectroscopy (which require surface laboratory equipment) with a computational system. Machine learning models substitute the mechanical/optical measurement processes with algorithms that process borehole logging data, enabling the same mineralogy determination functionality to operate downhole where physical XRD and infrared equipment cannot be deployed
Solution Approach 2:
The machine learning model serves as an intermediary that translates borehole logging measurements into mineralogy information that would otherwise require surface laboratory analysis. This intermediary layer enables downhole applicability while maintaining the accuracy characteristics of direct measurement methods by learning from extensive training data generated using those methods
3Measurement precision
If machine learning models are trained on extensive datasets, then estimation accuracy of hydrocarbon content is improved, but loss of time increases during model training and data processing
Solution Approach 1:
The patent performs preliminary action by training machine learning models on extensive datasets beforehand, creating pre-trained models that can then be deployed for rapid real-time predictions. The time-consuming training phase is completed in advance during model development, allowing the deployed models to provide accurate hydrocarbon content estimates without incurring time delays during actual drilling operations
Solution Approach 2:
The patent implements a dynamic approach where machine learning models are trained on extensive datasets during the development phase to achieve high accuracy, then deployed for real-time predictions during drilling operations. The system adapts between offline training mode (where time is less critical) and online prediction mode (where speed is essential), optimizing the balance between training time and operational efficiency
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
Enables accurate characterization of mineralogy and formation properties in real-time during drilling operations, improving the estimation of hydrocarbon content and informing drilling strategies with enhanced precision.
Implementation Method 1
Each artificial neuron has an associated activation function that defines an output value dependent on a input value or set of input values received on input connections to the artificial neuron
Data Source
AI summary
Methods and systems are provided to learn and apply a mapping function from data representing concentrations of atomic elements in a geological formation (or other data corresponding thereto) to mineral component concentrations in the geological formation (and/or from mineral component concentrations to reconstructed elemental concentrations in the geological formation). The mapping function can be derived from a trained neural network (such as an autoencoder). The output of the mapping function can be used to determine estimates of one or more formation properties, such as formation matrix density, formation porosity, matrix Sigma, formation saturation, other formation property, or combinations thereof.