Machine Learning Acoustic Imaging for Flaw Detection Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional total focusing method (TFM) and delay-and-sum algorithms for acoustic techniques lose significant information in the summation process, limiting the accuracy of flaw detection in non-destructive testing.
Innovation Solution
Applying a trained machine learning model to acoustic imaging data, avoiding summation, to generate a probability of flaws per pixel or voxel, thereby creating a flaw characterization image.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional total focusing method (TFM) and delay-and-sum algorithms are used for acoustic imaging, then the processing is computationally straightforward, but significant information is lost in the summation process, reducing measurement precision
Solution Approach 1:
The patent extracts and preserves individual delayed acoustic signals from the summation process, preventing information loss. Instead of immediately summing all signals, the system separates and processes individual signals through machine learning models, extracting useful information from each signal before integration.
Solution Approach 2:
The patent transforms the raw acoustic signal processing approach by applying machine learning models that change the parameters of signal analysis. The trained models process signals in ways that conventional summation cannot, extracting features and patterns that preserve information while improving measurement precision.
2Measurement precision
If machine learning models are applied to acoustic imaging data, then information retention and measurement precision are improved, but device complexity increases
Solution Approach 1:
The machine learning models are trained in advance on representative acoustic data, performing preliminary learning of flaw patterns and signal characteristics. This pre-training allows the models to process new inspection data efficiently without requiring complex real-time computation, reducing operational complexity while maintaining high precision.
Solution Approach 2:
The patent uses trained machine learning models that capture and replicate the complex relationships between acoustic signals and flaw characteristics. These models act as computational copies of expert analysis, replacing complex manual or algorithmic processing with pre-trained intelligence that simplifies the inspection system's operational complexity.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
Retains and utilizes information lost in summation processes, enhancing the accuracy and efficiency of flaw detection by generating images depicting the probability of flaws per pixel or voxel.
Implementation Method 1
an ultrasonic transducer or an array of such transducers can be used to inspect a structure using acoustic energy
Implementation Method 2
Inhomogeneities on or within the structure under test can generate scattered or reflected acoustic signals in response to a transmitted acoustic pulse
Data Source
AI summary
A computerized method of image processing using processing circuitry to apply a previously trained machine learning model in a system for non-destructive testing (NDT) of a material is described. The method can include acquiring acoustic imaging data of the material, the acoustic imaging data acquired at least in part using an acoustic imaging modality, generating an acoustic imaging data set corresponding to an acoustic propagation mode, applying the previously trained machine learning model to the acoustic imaging data set, and generating an image of the material depicting a probability of a flaw per pixel or voxel based on the application of the previously trained machine learning model.