Latent Seismic Fault Detection Without Labeled Training Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional fault mapping in seismic data is labor-intensive, time-consuming, and sensitive to signal-to-noise ratio, while supervised machine learning methods require costly and time-consuming data labeling.
Innovation Solution
An unsupervised fault detection method using a pre-trained autoencoder to generate latent representations of seismic data, allowing for automated fault mapping without manual intervention or predefined labels, and robust to noise.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual two-dimensional fault picking methods are used, then domain expertise can be applied to identify faults, but the process becomes labor-intensive and time-consuming
Solution Approach 1:
The patent replaces manual mechanical fault picking with an automated machine learning system. The autoencoder neural network automatically processes seismic data to detect faults, substituting human manual analysis with an automated computational system that maintains high accuracy while dramatically reducing time requirements.
Solution Approach 2:
The autoencoder performs self-service by automatically learning to identify fault patterns from seismic data without requiring manual labeling or intervention. The model trains on unlabeled seismic data and autonomously develops fault detection capabilities, eliminating the need for continuous human expert involvement.
2Productivity
If structural seismic attributes such as variance and edge detection are used to extract fault-like features, then manual effort is reduced, but the quality becomes sensitive to signal-to-noise ratio and data defects
Solution Approach 1:
The patent transforms the approach by changing from using fixed structural attributes (variance, edge detection) to using learned latent representations from a trained autoencoder. The model learns optimal feature parameters directly from the data, adapting to different signal-to-noise conditions rather than being sensitive to them.
Solution Approach 2:
The autoencoder acts as an intermediary between raw seismic data and fault detection. It processes the input data through learned transformations, creating robust latent representations that are less sensitive to noise and data defects compared to direct structural attribute analysis.
3Extent of automation
If supervised machine learning techniques are used for fault detection, then automation is achieved, but costly and time-consuming data labeling is required
Solution Approach 1:
The patent inverts the conventional supervised learning approach by using unsupervised learning with autoencoders. Instead of requiring labeled data to train the model, the system uses unlabeled seismic data to train the autoencoder, which then automatically detects faults without needing human-annotated training examples.
Solution Approach 2:
The autoencoder performs self-service by automatically learning fault patterns from unlabeled seismic data. The model trains itself without requiring external labeling or human intervention, eliminating the costly and time-consuming data annotation process while maintaining high automation levels.
Data Source
AI summary
Fault detection in seismic data using a latent representation of seismic data. Sample data may be created from seismic data and used to train an autoencoder. The autoencoder is used to generate a latent representation of seismic data. A fault attribute is computed by selecting two sets of traces in proximity to each other in the seismic data, generating their corresponding latent representations using the trained autoencoder, and computing a fault attribute between the two latent representations. The fault attribute is used to identify a fault in the seismic data.