Nondestructive Inspection Using Color Image Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current nondestructive inspection systems using radio waves struggle to accurately identify the type of internal state of a product, such as the presence of bubbles or foreign objects, due to simple threshold determination and phase jump issues, leading to inaccuracies in foreign object detection.
Innovation Solution
A learning apparatus and method that preprocesses relative phase and intensity differences between radio waves into color images in the HSV color space, enabling the creation of an identification model using training data to accurately identify the internal state of a product, employing a convolutional neural network for feature extraction and metric learning to improve identification accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If simple threshold determination is used to detect foreign objects, then the system is simple to operate, but the accuracy of identifying the type of internal state is insufficient
Solution Approach 1:
The patent introduces color images as an intermediary representation that transforms complex radio wave phase and intensity data into visual formats. This mediator enables machine learning models to effectively process and classify internal states while maintaining system operability. The color image generation process converts multi-dimensional radio wave data into a two-dimensional visual representation that preserves essential features for classification.
Solution Approach 2:
The patent replaces simple threshold-based mechanical determination with a machine learning-based identification model. This substitution enables the system to automatically learn complex patterns from training data and accurately identify different internal states (such as bubbles, foreign objects, or defects) without requiring manual threshold adjustment or complex rule-based logic.
2Measurement precision
If machine learning algorithms are applied to identify internal states, then the identification accuracy can be improved, but phase jump issues reduce the reliability of the system
Solution Approach 1:
The patent transforms the phase difference parameter into a color representation in the color image, where phase information is mapped to hue values. This parameter transformation handles phase wraparound issues naturally, as the hue parameter in color space is inherently periodic and continuous, eliminating discontinuities caused by phase jumps and providing stable input for machine learning models.
Solution Approach 2:
The color image serves as a robust intermediary that stabilizes phase information against jumps. By converting phase differences into continuous hue values in the HSV color space, the system creates a representation that is immune to the 2π phase wraparound problem, ensuring reliable and consistent input for the identification model regardless of phase discontinuities.
3Device complexity
If only presence or absence of foreign objects is determined, then the inspection process is simple, but the quality improvement is limited
Solution Approach 1:
The patent segments the internal state identification into multiple distinct categories (such as bubbles, foreign objects, defects, and normal states) rather than treating it as a single binary classification. This segmentation enables the system to provide detailed quality information that can drive targeted quality improvements while maintaining a unified inspection process through the color image-based machine learning approach.
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
The system achieves accurate identification of internal states with improved accuracy, as demonstrated by experimental results showing identification rates of 90% or higher, enhancing the quality of product inspection by distinguishing between different internal states and unknown defects.
Implementation Method 1
relative phase differences and relative intensity differences between a plurality of transmission/reception waves based on radiation of radio waves to an object to be measured
Data Source
AI summary
A learning device that comprises a preprocessing circuit and a learning circuit. The preprocessing circuit performs processing that converts the relative phase differences and relative intensity differences between a plurality of transmission/reception waves that are based on the radiation of electromagnetic waves at a measured object into a color image. The learning circuit uses first color images that have been processed by the preprocessing circuit and training data that associates second color images and types of internal states for measured objects to learn an identification model for identifying the types of internal states.