Neural Network Well Log Depth Matching for Thin Laminations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The synchronization of gamma-ray logs from multiple logging passes in well logging is challenging due to environmental perturbations, tool sticking, and irregular borehole shapes, leading to inaccurate depth measurements and sub-optimal interpretation, particularly in thinly-laminated formations.
Innovation Solution
A machine learning-based method using a neural network for automated depth matching of gamma-ray logs, employing data augmentation and Generative Adversarial Networks (GANs) to create a large training dataset, and incorporating data filtering and stacking techniques for precise synchronization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional synchronization techniques are used to match gamma-ray logs from multiple logging passes, then the process can be completed with existing methods, but the depth matching accuracy deteriorates due to environmental perturbations, tool sticking, and irregular borehole shapes
Solution Approach 1:
The patent replaces traditional mechanical/mathematical synchronization methods with a machine learning-based approach. A neural network model is trained to predict depth shifts between logging passes by learning patterns from labeled data pairs, automatically correcting for environmental perturbations, tool sticking, and borehole irregularities without requiring complex mechanical adjustments or manual intervention
Solution Approach 2:
The patent creates synthetic training data by copying and transforming real well log data. Synthetic gamma-ray logs are generated by applying simulated depth shifts and transformations to real log data, creating artificial training pairs that teach the neural network to handle various depth mismatch scenarios. This synthetic copying enables the model to learn robust depth matching patterns without requiring extensive manual annotation of real data
2Measurement precision
If manual labeling of training data is used, then the data quality can be controlled, but the time consumption increases significantly
Solution Approach 1:
The patent generates synthetic training data by copying real well log data and applying simulated transformations. Synthetic gamma-ray logs are created by taking real logs, applying random depth shifts and environmental perturbations, and creating corresponding label pairs. This copying process automatically generates large volumes of high-quality training data without requiring manual annotation, dramatically reducing data preparation time while maintaining data quality
Solution Approach 2:
The patent performs preliminary data transformation and augmentation during the training data generation phase. By pre-applying simulated depth shifts, noise, and perturbations to create synthetic training pairs, the system prepares robust training data in advance. This preliminary action eliminates the need for time-consuming manual labeling while ensuring the training data reflects real-world complexity
3Quantity of substance
If data augmentation is applied to expand the training dataset, then the dataset size increases, but the computational complexity increases
Solution Approach 1:
The patent expands the training dataset by copying real well log data and generating synthetic variations. Real logs are copied and transformed through simulated depth shifts, noise addition, and perturbation application to create multiple synthetic training pairs from each original data point. This copying and transformation approach efficiently multiplies the dataset size while maintaining manageable computational complexity through automated vectorized operations
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method and computing system device for receiving a plurality of well logs. A depth shift between at least one well log of the plurality of well logs and at least one other well log may be determined based upon, at least in part, processing the plurality of well logs with a neural network. The plurality of well logs may be matched with one another based upon, at least in part, the depth shift between the at least one well log and the at least one other well log.