Deep Learning Neural Network for Borehole Image Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Borehole imaging tools face challenges in achieving accurate and reliable high-resolution formation property images due to the standoff effect and simplified inversion schemes, which hinder the identification of thin beds and other fine formation features.
Innovation Solution
The implementation of a Deep-learning Neural Network (DNN) system to process formation property measurements, enhancing image accuracy and reliability by learning from patterns of inaccuracy in existing inversion methods and correcting formation property images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If traditional inversion schemes based on simplified models are used, then the processing is computationally efficient and easy to implement, but the resolution and accuracy of formation property images deteriorate due to standoff effect
Solution Approach 1:
A neural network is introduced as an intermediary between the simplified inversion scheme and the final formation property image. The neural network learns the complex relationship between raw measurements and accurate formation properties, effectively mediating the trade-off between computational simplicity and imaging accuracy by capturing non-linear effects that simplified models miss
2Measurement precision
If complex 3-D inversions are used to improve image accuracy, then the resolution and reliability of formation property images improve, but the computational time and complexity increase significantly
Solution Approach 1:
The neural network is trained in advance on a large dataset of synthetic borehole images with known formation properties. This preliminary training phase captures the complex 3-D inversion relationships, allowing the trained network to rapidly predict formation properties from new measurements without performing computationally intensive 3-D inversions in real-time
Solution Approach 2:
Instead of performing computationally expensive 3-D inversions for each new measurement, the system creates a neural network model that copies the essential information and relationships learned from extensive training data. This model can then quickly predict formation properties without repeating the full 3-D inversion process
3Difficulty of detecting and measuring
If high-resolution imaging is pursued to identify thin beds and fine features, then the detection capability improves, but the impact of standoff effect and simplified models worsens image reliability
Solution Approach 1:
The system changes the approach from using simplified physical models to using a data-driven neural network model. By training the neural network on diverse synthetic data representing various formation conditions and standoff effects, the system adapts to maintain reliable high-resolution imaging across different geological scenarios
Data Source
AI summary
A method for enhancing a formation property image may include taking at least one set of formation property measurements with a borehole imaging device, arranging the at least one set of formation property measurements into a two-dimensional image with a buffer, feeding the two-dimensional image into a deep-learning neural network (DNN), and forming a corrected formation property image from the two-dimensional image. The method may further include inverting the at least one set of formation property measurements to form at least one set of inverted formation property measurements and arranging the at least one set of inverted formation property measurements into a two-dimensional image with a buffer.