ML Decoder Reconstructs Acoustic Data from Compressed TFM Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional acoustic imaging techniques, such as the total focusing method (TFM), suffer from data loss due to dimensionality reduction, making it impossible to reconstruct original acoustic data like the full matrix capture (FMC) matrix, and cannot generate new images using different generation parameters, such as altered part thickness or acoustic modes.
Innovation Solution
Applying a previously trained decoder machine learning model to encoded acoustic images, like TFM images, to generate reconstructed acoustic data, which can then be used to recreate new encoded images using different generation parameters, effectively overcoming the limitations of existing methods by enabling the modification of parameters like part thickness and acoustic mode.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If dimensionality reduction is applied in acoustic imaging (TFM method), then image processing efficiency is improved, but data loss occurs making original acoustic data irretrievable
Solution Approach 1:
The patent creates a digital twin or copy of the original acoustic data through machine learning reconstruction. The decoder model generates a reconstructed FMC matrix that serves as a faithful copy of the original full acoustic data, allowing the system to work with compressed TFM images while recovering the complete information content when needed.
Solution Approach 2:
The patent transforms the data representation parameters by converting between different acoustic data formats (FMC matrix to TFM image and back) using machine learning. The decoder model changes the parameter space from compressed image representations back to full acoustic data representations, enabling flexible data manipulation without permanent information loss.
2Adaptability or versatility
If original acoustic data is preserved for future reprocessing, then data flexibility is improved, but data storage requirements increase
Solution Approach 1:
The patent transitions between different data dimensions and representations. Instead of storing only one format, the system maintains the ability to convert between compressed TFM image representations and full FMC matrix data through machine learning, effectively adding a computational dimension that allows flexible data retrieval without proportional storage increases.
Solution Approach 2:
The patent performs preliminary encoding of acoustic data into compressed TFM images while preserving the capability for future reconstruction. The machine learning decoder is pre-trained and ready to reconstruct full acoustic data when needed, allowing the system to store compressed representations while maintaining adaptability for various inspection scenarios.
3Device complexity
If acoustic images are generated with fixed generation parameters, then processing simplicity is improved, but adaptability to different inspection conditions is reduced
Solution Approach 1:
The patent introduces dynamic parameter adjustment capability through machine learning reconstruction. The system can regenerate acoustic images with modified generation parameters (such as different part thicknesses or acoustic velocities) by reconstructing the full acoustic data first, then re-processing with new parameters. This makes the processing pipeline dynamic and adaptable rather than static and fixed.
Data Source
AI summary
Acoustic data, such as a full matrix capture (FMC) matrix, can be reconstructed by applying a previously trained decoder machine learning model to one or more encoded acoustic images, such as the TFM image(s), to generate reconstructed acoustic data. A processor can use the reconstructed acoustic data, such as an FMC matrix, to recreate new encoded acoustic images, such as TFM image(s), using different generation parameters (e.g., acoustic velocity, part thickness, acoustic mode, etc.).