Multimodal Prompt Querying on Subsurface Models Using Vector Embeddings
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current seismic interpretation of subsurface models is manual, prone to human errors, and limited by human expertise, while existing generative AI models fail to generalize well on domain-specific subsurface data.
Innovation Solution
A method and system using generative AI-enabled multimodal prompt querying that extracts image embeddings from seismic data, stores them in a vector database, and retrieves similar images based on input prompts, leveraging a fine-tuned multimodal foundation model like CLIP for efficient subsurface data analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual seismic interpretation is performed by human interpreters, then domain expertise and contextual understanding are applied, but the process is time-consuming, prone to human errors, and limited by individual expertise
Solution Approach 1:
The patent replaces the mechanical human interpretation process with an automated AI-based system that uses machine learning models to analyze seismic data. The system substitutes human visual inspection and manual interpretation with computational algorithms that can process seismic cubes automatically, eliminating human errors and significantly reducing analysis time while maintaining or improving interpretation accuracy through consistent application of geological principles
Solution Approach 2:
The system enables self-service by allowing the seismic interpretation process to perform automatically without continuous human intervention. The AI model independently processes seismic data, generates interpretations, and can even self-correct through iterative refinement, freeing interpreters from routine manual scanning while preserving expert-level analysis capabilities
2Extent of automation
If generative AI models are directly tested with domain-specific subsurface images, then automation capability is provided, but the models fail to generalize well due to lack of domain adaptation
Solution Approach 1:
The patent applies preliminary action by fine-tuning the generative AI model on domain-specific seismic data before deployment. The system pre-trains the model using seismic cubes and associated geological annotations, preparing it with domain knowledge in advance. This preliminary domain adaptation ensures the model generalizes well to new subsurface images while maintaining full automation capability
Solution Approach 2:
The system changes the parameters and characteristics of the AI model through domain-specific fine-tuning. By adjusting model weights, learning rates, and training data distributions to match seismic data characteristics, the model transforms from a generic image processor to a specialized seismic interpreter that generalizes effectively across different subsurface scenarios
Data Source
AI summary
A method for performing generative artificial intelligence (AI)-enabled multimodal prompt querying on subsurface models includes receiving input data. The input data includes seismic data that represents a subsurface formation. The method also includes generating a plurality of images based upon the input data. The method also includes extracting first image embeddings based upon the plurality of images. The method also includes storing the first image embeddings in a vector database. The method also includes receiving an input prompt. The method also includes extracting a prompt embedding based upon the input prompt. The method also includes storing the prompt embedding in the vector database. The method also includes identifying a similar one of the images based upon the prompt embedding.