Ultrasonic Defect Depth Detection With Self-Supervised Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Ultrasonic inspection for defect detection in industrial applications, such as nuclear power plants, faces challenges in accuracy due to inspector-dependent results and the difficulty in obtaining label data for AI training, especially when surface states vary.
Innovation Solution
A non-destructive inspection method using self-supervised learning that synthesizes defect signals with original signals through a denoising autoencoder, applying random scaling and location augmentation, and uses a residual layer to predict defect location and depth based on statistical thresholds and time of flight.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional ultrasonic inspection is performed, then non-destructive defect detection is achieved, but inspection accuracy varies depending on the inspector and surface state
Solution Approach 1:
The system uses self-supervised learning where the model trains on its own output without requiring external label data. The denoising autoencoder automatically learns defect characteristics by processing ultrasonic signals through encoder-decoder architecture, enabling the system to independently perform accurate defect detection without relying on inspector expertise or pre-labeled data from multiple inspectors
Solution Approach 2:
The patent creates synthetic defect signals by superimposing defect patterns onto normal ultrasonic signals. These synthesized defect signals are then used to train the denoising autoencoder, allowing the model to learn defect detection capabilities without copying actual defect data from multiple inspectors or requiring manual labeling of diverse inspection scenarios
2Reliability
If AI technology is introduced into ultrasonic inspection, then inspection consistency is improved, but separate label data are required which are difficult to obtain
Solution Approach 1:
The self-supervised learning framework allows the AI model to generate its own training data by processing ultrasonic signals through the denoising autoencoder. The model learns to distinguish defect signals from normal signals by training on synthesized data, eliminating the need for manual acquisition and labeling of defect data from actual inspection scenarios
Solution Approach 2:
The patent converts the difficulty of obtaining labeled defect data into a benefit by using synthetic defect signal generation. Instead of requiring actual defect samples for training, the system creates artificial defect signals through signal processing, turning the limitation of unavailable labeled data into an opportunity to generate training data through computational methods
3Adaptability or versatility
If ultrasonic inspection is performed on samples with varying surface states, then real-world applicability is improved, but detection accuracy deteriorates
Solution Approach 1:
The denoising autoencoder learns to invariantly detect defects by training on synthesized defect signals with varied characteristics. The model adjusts its internal parameters to recognize defect patterns regardless of surface state variations, enabling consistent defect detection accuracy across different inspection conditions without requiring manual adaptation for each surface state
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
Enables accurate detection of defects without destructive methods, surpassing existing techniques by predicting defect presence and depth reliably even without separate label data, enhancing the precision of ultrasonic inspection.
Implementation Method 1
generating a reflected signal of a corresponding floor by radiating ultrasonic waves to the sample
Implementation Method 2
a denoising autoencoder that is trained to receive the plurality of data sets and to output an original signal from which the defect signal has been removed
Implementation Method 3
a residual layer unit configured to output a defect signal through a residual operation of the original signal output by the defect analysis model
Implementation Method 4
calculating the depth of the defect through time of flight (TOF) with respect to the defect signal determined as the defect
Data Source
AI summary
Disclosed is a non-destructive inspection method and system based on self-supervised learning, which detect the inside of an inspection object in a non-destructive way by using ultrasonic waves and also predict the depth of a defect through self-supervised learning. According to the non-destructive inspection method, it is possible to predict whether a defect is present within an inspection object and the depth of the inspection object in a non-destructive learning way, by augmenting a floor reflected signal into which physical characteristics of a defect reflected signal are incorporated through random scaling, applying an arbitrary defect signal to a random location, and determining whether a defect is present based on an average of the absolute values of a defect prediction signal and a statistical threshold by training a model in a way to remove the arbitrary defect signal through the structure of the denoising autoencoder.