Multimodal Lithofacies Prediction for Accurate Drilling Decisions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for lithofacies analysis face complexity and human error due to the manual evaluation of diverse data sources, leading to inconsistent and inaccurate characterization of subsurface rock formations.
Innovation Solution
A drilling decision system utilizing a multimodal approach with convolutional neural networks (CNN) to extract features from digital photos, remote sensing maps, and SEM images, combined with statistical feature extraction from seismic and well log data, and a rule-based engine for feature selection, followed by a one-dimensional CNN for classification and decision-making.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple diverse data sources are used for lithofacies analysis, then the accuracy and comprehensiveness of lithofacies identification is improved, but the complexity of the analysis system increases
Solution Approach 1:
The patent combines multiple diverse data sources (seismic data, well log data, digital photos, remote sensing maps, and SEM images) into a unified analysis framework. Different CNN models process each data type separately and then merge their features in a multimodal fusion architecture, achieving comprehensive lithofacies identification while managing complexity through modular design.
Solution Approach 2:
The patent introduces CNN-based feature extraction models as intermediary components between raw data and final lithofacies classification. These intermediaries automatically extract relevant features from each data source, reducing the complexity of direct analysis while preserving the information needed for accurate identification.
2Ease of operation
If manual analysis methods are used for diverse data sources, then flexibility in evaluation is maintained, but human error increases and analysis time extends
Solution Approach 1:
The patent replaces manual mechanical analysis with automated CNN-based deep learning models. These models consistently process diverse data sources without human error, maintaining reliability while reducing analysis time. The system preserves flexibility through configurable analysis parameters and interpretable output that supports expert decision-making.
Solution Approach 2:
The system performs automated feature extraction and lithofacies classification without requiring manual intervention. The CNN models self-adjust through training to optimize performance on different data types, reducing reliance on expert manual evaluation while maintaining consistent and reliable results.
3Productivity
If automated CNN-based analysis is implemented, then analysis time is reduced and human error is minimized, but the computational complexity increases
Solution Approach 1:
The patent segments the computational task into multiple specialized CNN models, each handling a specific data type (seismic, well log, digital photos, remote sensing, SEM images). This segmentation allows parallel processing of different data sources, improving analysis speed while managing computational complexity through modular architecture.
Solution Approach 2:
The patent transitions from traditional 2D image processing to multimodal analysis by adding temporal and spectral dimensions. The CNN models process data across multiple dimensions simultaneously, extracting features from different data types and integrating them to achieve comprehensive lithofacies identification efficiently.
Data Source
AI summary
A system and method for drilling decision system based on multimodal lithofacies identification includes an image data collection interface for collecting data from digital photos, remote sensing maps and scanning electron microscope (SEM) images. A first convolutional neural network (CNN) model to extract visual features from digital photos, a second CNN model to extract spatial-structural features from the remote sensing maps and a third CNN model to extract microstructural features from the SEM images. An output feature layer combines the extracted features from the first, second and third CNN models to output combined imaging features, a 1-D CNN model extracts the features from the combined imaging features and performs only 1D convolutions. A Neural Network (NN) classifier takes features from the output of 1-D CNN model to predict lithofacies classes with respective composition percentages. An output device configured to output a decision based on the predicted lithofacies classes.