Ultrasonic Powerplant Component Inspection With Self-Supervised ML
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing non-destructive inspection methods for internal defects in components, such as rotor disks, struggle to distinguish between component variability and actual defects, often relying on individual signal peak analysis which is limited and prone to false positives.
Innovation Solution
A self-supervised machine learning technique using pretext tasks like masked reconstruction and augmentation invariance pretext is employed to process ultrasonic signals, enabling the differentiation between normal component variability and actual defects by learning invariant representations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If individual signal peak analysis is used for defect detection, then the inspection method is simple to implement, but the measurement precision deteriorates due to inability to distinguish component variability from actual defects
Solution Approach 1:
The patent replaces traditional mechanical signal peak analysis with a machine learning-based approach. The system uses a processor to receive ultrasonic signals and applies machine learning algorithms to automatically distinguish between component variability and actual defects, eliminating the need for manual signal peak interpretation while significantly improving detection accuracy.
Solution Approach 2:
The patent introduces machine learning models as an intermediary between the ultrasonic signals and the defect detection decision. The machine learning algorithm processes the signals, learns to differentiate between normal variability and defects, and provides accurate defect identification without direct human intervention in the analysis process.
2Device complexity
If traditional inspection methods are used, then the device complexity is low, but the reliability deteriorates due to high false positive rates from inability to model component variability
Solution Approach 1:
The patent implements self-service through autonomous machine learning models that automatically learn from ultrasonic signals and make defect detection decisions without human intervention. The system performs self-training and self-evaluation, continuously improving its ability to distinguish component variability from defects while maintaining high reliability.
Solution Approach 2:
The patent incorporates feedback mechanisms where the machine learning model continuously refines its defect detection capabilities based on signal analysis. The system evaluates its performance and adjusts its decision-making process, creating a feedback loop that improves reliability over time while managing system complexity through automated learning.
3Measurement precision
If machine learning techniques are applied to process ultrasonic signals, then the measurement precision improves for defect detection, but the device complexity increases due to computational processing requirements
Solution Approach 1:
The patent segments the signal processing task into distinct functional components: signal acquisition, machine learning analysis, and defect determination. This segmentation allows the complex machine learning algorithms to be applied systematically to ultrasonic signals, improving precision while managing complexity through modular processing stages.
Solution Approach 2:
The patent transforms the ultrasonic signals into different parameter representations that are more suitable for machine learning analysis. By changing the signal parameters and features extracted from the raw ultrasonic data, the system enables more effective defect detection while optimizing the computational complexity through intelligent feature selection and transformation.
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
This approach allows for more accurate detection of defects by modeling component-to-component variability, reducing false positives and improving defect identification in rotor disks without requiring labeled datasets.
Implementation Method 1
using a transducer to transmit a first signal into a component
Implementation Method 2
using the transducer to sense the component for a second signal produced as a result of the first signal being transmitted into the component
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method of inspecting a component (22) is provided that includes: using a transducer (32) to transmit a first signal into a component (22) comprising a solid metallic material; using the transducer (32) to sense the component (22) for a second signal produced as a result of the first signal being transmitted into the component (22), and produce a response signal representative of the second signal; and processing the response signal to determine a presence or an absence of a defect in the solid metallic material of the component (22), the processing using a controller (48) configured with a self-supervised machine learning technique that is trained to be invariant to a component variability portion of the response signal.