Metallographic Phase Classification for Fast Property Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for quantitatively evaluating the metallographic structure of metallic materials, such as steel plates, are prone to large errors due to operator variability and require significant time, making it difficult to accurately link material properties to the structure.
Innovation Solution
A method and device that utilize image preprocessing, feature value calculation, and machine learning to classify and predict the phase of metallographic structures, optimizing luminance levels for accurate phase classification and enabling efficient prediction of material properties.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual color-coding method is used to evaluate phase fraction, then operator can identify different phases, but large error occurs among operators and significantly long time is needed
Solution Approach 1:
The patent replaces the manual mechanical color-coding process with an automated image processing system using luminance value binarization. The computer automatically converts grayscale images to binary images by comparing pixel luminance values against a threshold, eliminating manual intervention and enabling rapid, consistent phase fraction measurement without operator variability.
Solution Approach 2:
The patent changes the measurement parameter from subjective color perception to objective luminance value thresholding. By converting the continuous grayscale luminance values into binary categories (phase present/absent) based on a determined threshold value, the system achieves both speed and consistency in phase fraction evaluation.
2Productivity
If luminance value binarization method is used to evaluate metallographic structure, then analysis time is reduced significantly, but large error occurs when difference in luminance value for each phase is not clear
Solution Approach 1:
The patent employs feedback through iterative threshold optimization. The system determines the luminance value threshold by analyzing the histogram distribution of luminance values in the metallographic image, using the feedback from the image data itself to establish an optimal threshold that maximizes phase classification accuracy while maintaining fast processing speed.
3Reliability
If repeated observation by optical microscope is conducted to control metallographic structure, then material properties can be evaluated, but significantly long time is needed for quantitative evaluation
Solution Approach 1:
The patent replaces repeated manual microscopic observations with a single automated image processing operation. By substituting the mechanical process of multiple manual measurements with an automated computer-based luminance thresholding system, the patent achieves reliable material property evaluation in a fraction of the time required for traditional repeated observations.
Data Source
Figure 1~2
Figure 3
Figure 4~5
AI summary
A metallographic structure phase classification method includes a feature value calculation step of calculating one or more feature values for each pixel of an image captured of a metallographic structure of a metallic material, and a phase classification step of classifying the phase of the metallographic structure in the image by inputting each feature value calculated in the feature value calculation step to a learning-completed model subjected to learning using feature values to which one of labels of a plurality of phases of the metallographic structure are allocated as inputs and the labels of the phases as outputs and acquiring the label of the phase of a pixel corresponding to the input feature value.