Ultrasonic Defect Analysis Using Priority-Learned Enhanced Imaging
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing defect detection systems require multiple measurements to prepare original images, which can be time-consuming and inefficient.
Innovation Solution
A defect analysis device that creates enhanced images emphasizing defects using a learned model associating teacher information with standard and target ultrasonic images, prioritizing defect detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple original images are acquired by measurement a plurality of times for learning, then the defect detection accuracy is improved, but the time consumption and operational effort increase significantly
Solution Approach 1:
The patent uses synthetic ultrasonic images generated through simulation as copies of real ultrasonic images for training the deep learning model. These synthetic images are created by simulating ultrasonic wave propagation and defect interactions, providing sufficient training data without requiring multiple physical measurements. This copying approach maintains defect detection accuracy while eliminating the time-consuming repeated measurements.
Solution Approach 2:
The patent performs preliminary actions by pre-generating synthetic ultrasonic images with various defect types, positions, and characteristics before actual defect detection. The deep learning model is trained in advance using these pre-prepared synthetic images, so when real defect detection is needed, the model is already ready and no additional measurements are required, thus reducing time consumption.
2Measurement precision
If multiple original images are acquired by measurement a plurality of times for learning, then the defect detection accuracy is improved, but the operational effort increases significantly
Solution Approach 1:
The patent replaces the need for multiple physical measurements with synthetic image copies generated through computer simulation. The synthetic ultrasonic images are created by simulating ultrasonic wave propagation through materials with embedded defects, providing diverse training data without requiring operators to perform repeated measurements, thus significantly reducing operational effort.
Solution Approach 2:
The system performs self-service by automatically generating synthetic ultrasonic images and training data through simulation algorithms without requiring manual measurement operations. The deep learning model automatically learns from these synthetic images, eliminating the need for operators to manually acquire multiple original images, thereby reducing operational effort.
3Measurement precision
If a learned model prioritizes certain defects among multiple defects is used, then the detection accuracy for critical defects is improved, but the complexity of the learning system increases
Solution Approach 1:
The patent applies local quality by making the deep learning model's defect detection capability selective rather than uniform. The model is trained to prioritize detection of specific defect types or characteristics that are most critical, allocating more detection resources and attention to these important defects while maintaining adequate detection for others. This is achieved through targeted loss functions and training strategies that emphasize critical defect detection, improving accuracy for important defects without requiring a completely complex system redesign.
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 efficient defect analysis by preferentially highlighting defects, improving detection accuracy and reducing processing time.
Implementation Method 1
uses, for example, an ultrasonic flaw detector to emit ultrasonic waves to an object while changing the emission position, and measures reflected waves at each emission position
Data Source
AI summary
A defect analysis device includes a model storage unit configured to store a learned model in which teacher information representing positions, shapes, and sizes of a plurality of defects embedded in a standard test piece is associated with detection information of the plurality of defects detected based on a plurality of standard ultrasonic images acquired from the standard test piece, a target image acquisition unit configured to acquire a plurality of target ultrasonic images from an inspection target, and an enhanced image creation unit configured to create a plurality of enhanced images emphasizing the plurality of defects based on the learned model and the plurality of target ultrasonic images, in which the learned model is learned by giving a priority to defects to be detected among the plurality of defects in each standard ultrasonic image.