Defect Inspection Method for High-Accuracy Crack Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional defect inspection methods struggle to accurately detect narrow cracks due to difficulties in distinguishing crack-related pixels from unrelated pixels, especially when non-defective point-like or short linear patterns exist in the background, leading to false detections and overlooked cracks.
Innovation Solution
The method involves scanning images in predetermined directions using dedicated filters to evaluate pixel luminance, selecting pixels based on specific luminance differences, connecting relevant pixels, and synthesizing results to remove noise, thereby enhancing crack detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a simple luminance threshold method is used for crack detection, then the processing is fast and easy to implement, but the detection accuracy is low and false detections occur frequently
Solution Approach 1:
The patent changes from using a single luminance threshold parameter to using multiple parameters including gradient information in multiple directions (Gx, Gy, Gxx, Gyy). This multi-parameter approach allows the system to distinguish narrow cracks from noise patterns by analyzing the spatial distribution and directional characteristics of pixel intensity changes, thereby improving detection accuracy while maintaining computational efficiency
Solution Approach 2:
The patent extends the detection from simple 1D luminance thresholding to 2D spatial gradient analysis by computing gradients in horizontal and vertical directions and their second derivatives. This dimensional extension enables the detection algorithm to capture the directional characteristics of cracks, allowing accurate identification of narrow linear defects that would be indistinguishable from point-like noise using simple thresholding
2Reliability
If a strict threshold value is used to reduce false detections, then fewer false positives occur, but actual cracks are overlooked and detection sensitivity decreases
Solution Approach 1:
The patent applies partial action by using multiple gradient components (Gx, Gy, Gxx, Gyy) rather than requiring all conditions to be met simultaneously. The evaluation value combines these partial indicators in a weighted manner, allowing the system to detect cracks even when some individual gradient measurements are ambiguous, thus maintaining high sensitivity while reducing false detections through the combined evidence from multiple partial measurements
3Measurement precision
If multiple scanning directions are used to improve crack detection accuracy, then detection reliability increases, but processing time and computational complexity increase
Solution Approach 1:
The patent segments the gradient computation into four independent directional components (Gx, Gy, Gxx, Gyy), each capturing gradient information in a specific direction. By computing these segmented directional gradients separately and then combining them through the evaluation function, the system achieves comprehensive multi-directional crack detection while maintaining computational efficiency through the modular and independent nature of each gradient calculation
Data Source
Figure 1A
Figure 1B
Figure 2(a)~2(b)
AI summary
There is provided a defect inspection method capable of detecting a crack with high accuracy. The defect inspection method includes the steps of: obtaining a shot image comprising pixels; and scanning the shot image in predetermined directions, and assigning a high evaluation value to a pixel M for each scanning direction when the luminance of the pixel M is lower than the luminances of first adjacent pixels K, O, located on both sides of the pixel M in the scanning direction and, in addition, the luminance of each of second adjacent pixels C, W, located on both sides of the pixel M in a direction perpendicular to the scanning direction, is lower than the luminances of third adjacent pixels A, E or U, Y located on both sides of the second adjacent pixel in the scanning direction. The method also includes the steps of selecting selection pixels based on the evaluation values of the pixels for each scanning direction; connecting the selection pixels for each scanning direction; and synthesizing the selection pixels of the predetermined scanning directions, and removing those pixels which do not meet the requirement for a predetermined shape from the selection pixels.