Automatic Borehole Sonic Classification via MLADI
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional borehole sonic dispersion data analysis is time-consuming, requires exacting calibration, and is prone to noise, limiting its effectiveness in distinguishing between different types of sonic data for characterizing mechanical formation properties.
Innovation Solution
A method using machine-learning automatic dipole interpretation (MLADI) to process sonic data from boreholes, extracting scattered dispersion points as smooth curves, and inputting them into a classifier with equivalent isotropic and homogeneous curves for automatic classification, enabling quick and efficient identification of dispersion types.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual dispersion analysis is used, then data quality can be evaluated, but the process is time-consuming
Solution Approach 1:
The patent replaces manual mechanical analysis with an automated computer-based system that uses machine learning algorithms to classify sonic dispersion data. The system automatically processes raw sonic data, applies dispersion analysis, and generates classification results without manual intervention, thereby maintaining data quality evaluation while dramatically reducing analysis time.
Solution Approach 2:
The automated classification system performs self-service by automatically evaluating and classifying sonic data without requiring manual calibration or expert intervention. The system includes built-in quality control mechanisms that automatically assess data quality and flag problematic measurements, enabling the system to serve itself in the data evaluation process.
2Measurement precision
If conventional dispersion analysis is used, then formation properties can be characterized, but exacting calibration on mud velocity and borehole calipers is required
Solution Approach 1:
The patent transforms the calibration problem by changing from absolute parameter calibration (mud velocity, borehole calipers) to relative parameter analysis. The machine learning model is trained to recognize patterns in dispersion data that are invariant to specific calibration values, allowing formation property characterization without exacting calibration of individual parameters.
Solution Approach 2:
The patent introduces dispersion analysis as an intermediary step between raw sonic data and formation property characterization. By first analyzing dispersion characteristics and then using machine learning to classify these dispersion patterns, the system mediates the complex calibration requirements, converting them into more robust pattern recognition tasks.
3Productivity
If noise from running machinery is not quieted, then data collection can proceed, but data quality deteriorates
Solution Approach 1:
The patent converts the harmful effect of machinery noise into a beneficial classification feature. The machine learning model is trained to recognize noise patterns and distinguish them from genuine formation signals. By incorporating noise characteristics into the classification process, the system can maintain data collection efficiency while improving data quality through intelligent discrimination of noise versus signal.
4Productivity
If automatic classification is implemented, then analysis speed increases, but system complexity increases
Solution Approach 1:
The patent segments the automatic classification system into distinct functional modules: data acquisition, dispersion analysis, feature extraction, machine learning classification, and result interpretation. Each module performs a specific function and can be independently optimized or replaced. This segmentation manages system complexity by breaking down the automated process into manageable components while maintaining high analysis speed.
Data Source
AI summary
A general-purpose workflow for automatic borehole sonic data classification to identify data into different physical categories and logging conditions, which are traditionally manually evaluated. The workflow uses machine learning techniques and physical knowledge for data classification, including pre-processing the high-dimensional high-quality dispersion modes extracted using a recently developed physical-driven ML enabled approach.