Sonic Data Classification Using Visual Features and CNN
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current sonic data classification methods are time-consuming, require high-performance computation, and struggle with noise and data quality issues, leading to inaccurate predictions and lack of quality control.
Innovation Solution
A method that involves obtaining raw sonic data, preparing a graph of data slowness versus frequency, performing statistical quality control, creating a digital dispersion picture, extracting features, and using a convolutional neural network to classify the data and provide a confidence score.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning programs are used for sonic data classification, then prediction accuracy is improved, but computation performance requirements increase
Solution Approach 1:
The patent extracts and removes noise from the raw sonic data before classification, using signal processing techniques to separate useful data from distortion. This preprocessing step improves the quality of input data for the machine learning model, allowing for accurate predictions with reduced computational complexity.
Solution Approach 2:
The patent performs preliminary quality checks and noise removal on the raw sonic data before it enters the machine learning classification process. By preparing the data in advance and removing harmful elements, the system achieves high prediction accuracy without requiring excessive computation power during the actual classification phase.
2Productivity
If machine learning programs are used for sonic data classification, then classification speed is improved, but data quality control capability deteriorates
Solution Approach 1:
The patent incorporates feedback mechanisms through quality checks that evaluate the raw sonic data before classification. The system assesses data quality metrics and provides feedback on whether the data meets acceptable standards, enabling automatic rejection or flagging of poor-quality data while maintaining fast classification speeds for acceptable data.
3Reliability
If manual classification process is used, then data quality control is improved, but time consumption increases
Solution Approach 1:
The patent implements an automated system that performs quality checks and classification without requiring manual domain expert intervention. The machine learning program autonomously evaluates data quality, performs noise removal, and conducts classification, thereby maintaining reliable quality control while dramatically reducing the time consumption associated with manual processes.
4Reliability
If noise is present in sonic data, then measurement reliability deteriorates, but machine learning programs cannot separate noise from useful data
Solution Approach 1:
The patent explicitly extracts and removes noise from the raw sonic data using signal processing techniques before feeding the data to the machine learning classifier. This noise separation capability restores measurement reliability by isolating and eliminating distortion while preserving the useful sonic information for accurate classification.
Data Source
Figure 1
Figure 2(A)~2(C)
Figure 3
AI summary
Embodiments presented provide for a classification of sonic data. In one aspect, visual features of sonic data are used to classify the sonic data and provide a quality control mechanism to ensure that a researcher understands the quality of the data calculations. In one or more embodiments, the method can obtain the raw sonic data from field measurements. The field measurements can pertain to downhole geological features.