Web Inspection Image Processing for Real-Time Defect Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current web inspection systems are inadequate for real-time detection and analysis of defects and formation irregularities in web manufacturing, particularly failing to classify weak defects and provide full web coverage at high dirt densities, and are not capable of online floc analysis.
Innovation Solution
A method that acquires digital images of the web and processes them to identify regions of interest by selecting local pixel units, deciding their interest based on thresholds and impact areas, and merging or initializing new regions, enabling real-time detection and classification of defects and formation irregularities across the entire web.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If snapshot images or limited cross direction band scanning imaging methods are used, then device complexity is reduced, but measurement precision and productivity are insufficient for real-time full web coverage at high dirt densities
Solution Approach 1:
The patent divides the web inspection task into multiple scanning bands across the cross direction, with each band processed independently and sequentially. This segmentation allows the system to achieve full web coverage by combining results from multiple passes, effectively increasing productivity without requiring a single overly complex imaging system to capture the entire web at once.
2Measurement precision
If current web inspection systems are used, then operation is simpler, but they cannot classify weak defects or provide full web coverage at high dirt densities
Solution Approach 1:
The patent performs preliminary classification of dirt particles by size and intensity before final counting and reporting. By pre-processing the data to categorize particles into different size ranges and intensity levels, the system can accurately classify both strong and weak defects while maintaining real-time processing capability through efficient data organization and reduced computational burden during final analysis.
3Productivity
If online real-time analysis is implemented, then productivity is improved, but measurement precision for weak defects and formation irregularities deteriorates
Solution Approach 1:
The patent employs dynamic threshold adjustment based on the local intensity distribution and dirt density in different regions of the web. By adapting the detection thresholds in real-time according to the specific characteristics of each scanned band, the system maintains high measurement precision for weak defects while preserving online processing speed, as the thresholds are updated continuously without requiring re-scanning.
4Measurement precision
If full web coverage is achieved, then measurement precision is improved, but processing time increases
Solution Approach 1:
The patent implements continuous scanning across the full web width with overlapping bands, ensuring that no region is missed and maintaining measurement precision. The continuous action is achieved by systematically moving the scan head or web in a coordinated manner, processing each band immediately as it is scanned, thereby minimizing idle time and maintaining high processing efficiency while achieving complete coverage.
Data Source
AI summary
A method for detection of distinctive features in a web being transported in a moving direction during a web manufacturing process including the steps of: a) acquiring an image of the web, including a plurality of pixels P i with iϵ{1; . . . ; p}, b) identifying a plurality of regions of interest each corresponding to a distinctive feature by processing the plurality of pixels P i by: c) selecting a local pixel unit including a subset P j with jϵS⊂{1; . . . ; p} of the plurality of pixels, the subset i) being representative of a subregion of the digital image, and ii) different from previously selected local pixel units, d) deciding whether the local pixel unit is of interest or not, i) if the local pixel unit is of interest, 1) identifying whether the local pixel unit is located within an impact area A k of a previously identified region of interest R k with kϵA⊆{1; . . . ; n}, 2) if the local pixel unit is not located within any impact area A k of any previously identified region of interest R k with kϵA⊆{1; . . . ; n}, or no regions of interest have previously been identified, (a) identifying the local pixel unit as a new region R n+1 of interest; (b) initializing an impact area A n+1 for the new region R n+1 of interest; (c) incrementing a counter n representative of the number of previously identified regions of interest; if the local pixel unit is located within an impact area A k 0 of a previously identified region of interest R k 0, (a) merging, depending on a merging condition, the local pixel unit with the previously identified region of interest R k 0, (b) if the merging condition is fulfilled, updating the impact area A k 0 of the region of interest R k 0; ii) preferably, if the local pixel unit is not of interest, identifying whether the local pixel unit is located within an impact area A k of a previously identified region of interest R k with kϵ{1; . . . ; n},if the local pixel unit is located within an impact area A k 0 of a previously identified region of interest R k 0, updating the impact area A k 0; e) repeating steps b) through d) until at least essentially all pixels of the image have been processed.