Neural Network Segmentation for Drilling Cuttings Lithology Identification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for automatically evaluating and characterizing subterranean drilling cuttings images are inefficient due to variability in size, shape, texture, and color, as well as challenges in accurately identifying and labeling multiple rock types within a single image, leading to limited success with machine learning algorithms.
Innovation Solution
A system utilizing a trained neural network to segment and label lithology types in drilling cuttings images, with continuous retraining based on corrected errors to improve segmentation accuracy, including image preprocessing, retraining with a subset of crops, and fine-tuning specific neural network layers for efficient identification of new lithology types.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If machine learning algorithms are used to automatically evaluate rock cuttings images, then automation is achieved, but computational time is excessive and accuracy is insufficient
Solution Approach 1:
The patent segments the rock cuttings images into individual particle images through automated detection and extraction. Each particle is separated into its own image crop, allowing the neural network to process and evaluate each particle independently. This segmentation enables more efficient computation by processing smaller, isolated images rather than analyzing entire complex images simultaneously, thereby reducing overall computational time while maintaining evaluation accuracy.
2Extent of automation
If machine learning algorithms are used to automatically evaluate rock cuttings images, then automation is achieved, but accuracy is insufficient for reliable deployment
Solution Approach 1:
The patent implements a feedback mechanism where geologists review and correct the neural network's labeling of particle images. These corrected labels are then fed back into the system to retrain and fine-tune the neural network model. This continuous feedback loop allows the system to learn from its mistakes and progressively improve its accuracy in identifying rock types, ensuring reliable deployment while maintaining automation.
3Measurement precision
If neural network is retrained to identify new lithology types, then accuracy improves, but retraining time increases
Solution Approach 1:
The patent applies partial retraining by selectively retraining only the neural network layers that are most relevant to identifying new lithology types, rather than retraining the entire network from scratch. This partial action approach allows the system to update its knowledge base efficiently, incorporating new lithology type information while preserving previously learned characteristics, thereby improving accuracy with minimal time investment.
Solution Approach 2:
The patent performs preliminary processing of particle images by extracting and preparing individual particle crops before they are used for training. This preliminary action includes automatic detection, extraction, and pre-processing of particle images, which prepares the data in advance for efficient neural network training. By performing these preparatory tasks beforehand, the actual training process is accelerated, reducing the overall retraining time while maintaining high accuracy in identifying new lithology types.
Data Source
AI summary
A method for evaluating drill cuttings includes acquiring a first digital image and processing the first digital image with a trained neural network (NN) to generate a first segmented image including a plurality of labeled segments in which at least one label includes a lithology type. The segmented image and the acquired first digital image are processed to retrain the NN. A second digital image is then be processed with the retrained NN to generate a second segmented image including a plurality of labeled segments in which at least one label includes a lithology type.