Inverting Anisotropic Constants via Borehole Sonic Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for estimating anisotropic parameters, particularly the three Thomsen parameters, in sedimentary rock formations are unreliable due to local minimum issues in cost function models and the lack of mud property measurements, which are essential for geomechanical analysis.
Innovation Solution
A data-driven, physically-constrained inversion method using a comprehensive random or grid searching algorithm to match theoretical borehole dispersion curves with measured modes, followed by regression to refine mud slowness and anisotropic parameter estimates, constraining search ranges with empirical correlations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If model-based inversion methods are used to estimate anisotropic parameters, then the estimation process can be performed, but the results are unreliable due to local minimum issues in the cost function
Solution Approach 1:
The patent replaces the traditional model-based inversion method (mechanical optimization approach) with a data-driven machine learning approach. Neural networks are trained on synthetic data to directly predict anisotropic parameters from borehole sonic data, avoiding the cost function minimization process that suffers from local minimum problems. This substitution of the estimation mechanism fundamentally resolves the reliability-precision contradiction.
Solution Approach 2:
The patent performs preliminary training of neural networks using extensive synthetic data generated from known formation models before actual field application. This pre-training phase establishes robust parameter relationships that enable reliable and precise estimation during actual logging operations without requiring real-time iterative inversion, thus resolving the contradiction between reliability and precision.
2Ease of operation
If model-based inversion is performed without accurate mud property measurements, then the inversion can proceed, but the results are affected by uncertainties in mud properties
Solution Approach 1:
The neural network model is designed to automatically handle uncertainties in mud properties by learning their effects during training on synthetic data that includes varied mud property scenarios. The model self-corrects for mud property variations without requiring explicit mud property measurements or manual adjustments, thus maintaining ease of operation while improving reliability.
Solution Approach 2:
The patent transforms the estimation problem from one requiring precise mud property inputs to one where the neural network learns to estimate anisotropic parameters directly from borehole sonic data while implicitly accounting for mud property variations. This parameter transformation approach allows the system to operate easily without mud property measurements while maintaining reliable results.
3Measurement precision
If comprehensive random or grid searching algorithms are used to match theoretical and measured dispersion curves, then accurate parameter estimation is achieved, but the computational complexity increases
Solution Approach 1:
The patent replaces complex iterative searching algorithms with pre-trained neural networks that provide direct parameter predictions. The computational complexity of comprehensive grid searching is shifted to the offline training phase, while field operations benefit from simple, fast neural network inference that maintains precision without algorithmic complexity.
Solution Approach 2:
The patent performs the computationally intensive parameter matching and optimization work during the offline training phase using synthetic data. Once trained, the neural networks provide rapid, accurate parameter estimation during field operations without requiring complex real-time algorithms, thus resolving the contradiction between precision and algorithmic complexity.
Data Source
Figure 1
Figure 2A~2B
Figure 3A~3B
AI summary
A method is presented wherein inversion for formation anisotropic constants is achieved using borehole sonic data.