Synthesizing Sonic Logs via Machine Learning Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
It is challenging to obtain complete sonic logs for wellbores due to limited or absent sonic logging in certain areas, which hinders accurate determination of mechanical properties and equipment selection for drilling, as sonic logs from appraisal wells may not accurately represent the formation type of planned wellbores due to geological variability.
Innovation Solution
A model is trained using downhole data, including drilling parameters and other log types, to synthesize sonic logs by predicting characteristics such as compressional and shear wave slowness, enabling the generation of complete sonic logs even in areas where direct sonic logging is not performed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If sonic logging is performed to obtain complete sonic logs, then accurate determination of mechanical properties and formation type is improved, but the cost and time required for logging operations increases
Solution Approach 1:
The system performs sonic logging at selected depth intervals beforehand to build a training dataset, then uses machine learning models to predict sonic log characteristics at unlogged intervals. This preliminary action at representative locations enables subsequent synthesis without requiring complete sonic logging throughout the entire wellbore.
Solution Approach 2:
The system creates synthetic copies of sonic log data at depth intervals where actual sonic logging was not performed. By training machine learning models on actual sonic log data from selected intervals and applying them to generate synthetic sonic logs at unlogged intervals, the system reproduces the characteristics of complete sonic logging without the associated time and cost.
2Reliability
If sonic logging is performed at all depth intervals to ensure complete coverage, then comprehensive formation characterization is improved, but the complexity and cost of the logging operation increases
Solution Approach 1:
The system performs sonic logging at selected depth intervals beforehand to build a training dataset, then uses machine learning models to predict sonic log characteristics at unlogged intervals. This preliminary action at representative locations enables subsequent synthesis without requiring complete sonic logging throughout the entire wellbore.
Solution Approach 2:
The system changes the approach from direct measurement at all intervals to a hybrid approach combining limited direct measurements with model-based predictions. By transforming the problem from complete physical logging to a data synthesis problem using machine learning, the system achieves comprehensive coverage with reduced operational complexity.
3Loss of information
If sonic logging is performed in appraisal wells to obtain formation data, then data availability is improved, but the applicability to planned wellbores deteriorates due to geological variability
Solution Approach 1:
The system segments the wellbore into intervals where sonic logging was performed and intervals where it was not. By processing and synthesizing data for each segment independently using machine learning models trained on local characteristics, the system preserves the geological specificity of each section while achieving complete coverage.
Solution Approach 2:
The machine learning models are trained on actual sonic log data from specific depth intervals to capture local formation characteristics. This local training approach ensures that the synthetic logs at unlogged intervals reflect the specific geological conditions of each section rather than applying generic formations characteristics from appraisal wells.
Data Source
AI summary
Aspects of the subject technology relate to systems, methods, and computer-readable media for synthesizing sonic logs for downhole environments. Data associated with a wellbore in a formation is accessed. A model configured to identify one or more characteristics associated with sound traveling through one or more formation is accessed. The data is applied to the model to predict a characteristics of the formation that is identifiable through sonic logging.