Polynomial Curve Fitting for Ultrasonic Back Wall Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for automatically detecting portions of interest in image or matrix data, such as in ultrasonic non-destructive testing, face challenges due to noise and varying positions of features like the back wall, leading to incomplete defect detection and reliance on inaccurate assumptions.
Innovation Solution
A computer-implemented method that uses approximate coordinates from training data to specify a bounding region and identify candidate parts, then fits a polynomial curve to estimate the location of the portion of interest, reducing human intervention and improving detection accuracy by narrowing the search space.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If automated pattern detection techniques are used to detect manufacturing defects, then detection speed and consistency are improved, but the presence of noise (echoes) under the back wall causes false detections and reduces accuracy
Solution Approach 1:
The patent segments the image data into distinct regions by detecting the back wall structure first, then separately analyzing the region above it for defects. This segmentation isolates the noise (echoes) under the back wall from the defect detection process, allowing automated pattern detection to work accurately on the relevant region only.
Solution Approach 2:
The patent extracts and removes the back wall structure and the noisy region beneath it from the analysis. By taking out these interfering elements, the automated defect detection can focus solely on the clean region above the back wall, eliminating false detections caused by echoes.
2Reliability
If manual visual inspection is performed by trained operators, then detection accuracy can be maintained through expert judgment, but the inspection process is time-consuming and may not provide full coverage
Solution Approach 1:
The patent performs preliminary automated detection of the back wall structure and defines the region of interest before conducting defect detection. This preliminary action prepares the data in advance, enabling subsequent automated defect detection to proceed quickly and accurately without requiring time-consuming manual inspection of the entire image.
Solution Approach 2:
The patent introduces an intermediary automated processing step that bridges manual inspection requirements and automated detection. The system automatically identifies the back wall and defines analysis regions, acting as an intermediary that prepares data for both automated defect detection and potential manual verification, reducing overall inspection time while maintaining accuracy.
3Device complexity
If previous methods assume constant number of layered structures and fixed positions, then processing is simplified and faster, but detection fails when these assumptions do not hold
Solution Approach 1:
The patent employs dynamic detection of the back wall structure rather than assuming fixed positions or constant layer configurations. The system adaptively identifies the back wall and its echoes in each image, allowing the analysis region to dynamically adjust to the actual structure present. This makes the detection robust to variations in layering while maintaining reasonable processing complexity.
Data Source
Figure 1
Figure 2
Figure 3a~4(d)
AI summary
In a method of locating a portion of interest in image data derived from an item under consideration, approximate coordinates of a feature in training data likely to correspond to the portion of interest are obtained from an operator, the training data relating to an item similar to the item under consideration. A bounding region which extends around an area defined by the approximate coordinates of the feature is specified. At least one candidate part which may be a component of the portion of interest is identified in the image data, and any candidate parts having coordinates which fall outside the bounding region are removed from the image data to obtain masked data. A polynomial curve of best fit is fitted to the masked data, and the candidate parts which fall within a region enclosing the curve are identified as structures forming the portion of interest.