A Computer Vision-Based Quality Assessment Technique for R2R Printed Silver Conductors on Flexible Plastic Substrates
Round 1
Reviewer 1 Report
Summary
The manuscript tries to apply the image processing technique to detect the printing defects of R2R printed silver conductors. The authors claim that a computer vision-based inspection system was applied to visualize the defects.
Major issue:
1) Abstract, the authors described the demand, tools, producers et al. However, what is the novelty of this paper? To the reviewer’s limited knowledge, the detection algorithm employed in section 2 is one of the basic image processing techniques. The authors just applied them to the R2R printed silver conductors. The novelty of this paper is vague.
2) The defects in this paper are focused on disconnection. How about other defects? Such as the uneven thickness and width?
3) Four layouts were tested in this research. How about out-of-sample layout? For example the gird partten?
Minor issue:
1) There is something wrong in line 54 (Error! Reference source not found.).
2) Figue 5, the authors claim that the output of the defect detection algorithm highlighting the disconnected resistance lines. Figue 5, the authors claim that the output of the defect detection algorithm highlighting the disconnected resistance lines. How is it highlighted? The reviewer could not find the highlighted part.
Author Response
Dear Sir/Madam,
Thank you for your time and consideration for the submitted manuscript. Please note that the type of this paper is Communication. Contrary to a full research article, the aim is to present preliminary results of new technology or materials and due to the nature of Communication paper, more in-depth findings and results will be published once the developed system is more matured and more data is available for publication.
Major Issues:
Respond to #1:
Thank you for your view, the importance of this Communication paper can be argued by talking about the speed and accuracy of the developed algorithm. The combination of the different stages of the algorithm provided a robust tool in terms of speed and accuracy in detecting defects on the flexible plastic substrates R2R printed silver conductors. Please note that this project was funded by the European Union’s HORIZON 2020 research and innovation program under Grant Agreement no. 820789 and according to the proposal for this project, the end users that provided the samples are the industry leaders in producing these specimen. Based on the statement from the end users, the developed solution has not been implemented on these particular R2R samples with this reliability at this scale. This is due to the difficulties in the R2R printing process of these samples which are rather challenging (some of the issues are mentioned in the manuscript).
Respond to #2:
The subtraction stage of the algorithm would score an image with the reference sample image as an output parameter “quality” which shows how much different is an image to its reference counterpart including thickness or width differences. Nonetheless, the system cannot localise the location of the deviation from the reference image at the current stage of the algorithm.
Respond to #3:
As the findings were based on a fixed pattern as part of a more complex, robust and general-purpose ADR system, the algorithm at this stage was designed to detect different types of either vertical or horizontal lines.
Minor issues:
Respond to #1:
This has been fixed.
Respond to #2:
Thank you for your suggestion, we have added an indicator to the image to show the disconnected area.
Reviewer 2 Report
The author presents computer vision-based quality assessment technology for R2R printed silver conductor. The algorithm has been clearly presented, and the result demonstrates qualified accuracy for this work. I would like to recommend to publish this paper after revision.
(1) Line # 54, Please add reference.
(2) On figure 1, I recommend to add plotting scale.
(3) Line # 93, the author said "the least defects." Would you please provide more details about how many defeats will be used as good sample? Would you prefer to use the ideal synthetic model (0 defeat) as good sample?
(4) Would you please add a short description about the image alignment if image rotation in need?
(5) Would you please consider to add more features to improve algorithm robustness with some dead pixels on camera?
(6) Is there any unexpected short been detected during your imaging?
(7) Line # 161, please provide the reason the 1000 times difference between 360 square um and 0.36 square mm.
Author Response
Dear Sir/Madam,
Thank you for your time and consideration for the submitted manuscript. Please note that the type of this paper is Communication. Contrary to a full research article, the aim is to present preliminary results of new technology or materials and due to the nature of Communication paper, more in-depth findings and results will be published once the developed system is more matured and more data is available for publication.
Respond to #1:
This has been fixed.
Respond to #2:
Thank you for your recommendation, a plotting scale has been added based on your suggestion.
Respond to #3:
With the supervision of technicians at the end user’s facility who provided the sample images and based on the results from the electrical testing devices, the “least defects” sample was chosen. That sample was then retouched to remove any remaining defects and converted into a defect-free version and into a perfect condition. A synthetic model with zero defect could not be used as the CAD design of the patterns did not represent the material and pixel density of the real samples’ background and as a result, could not be used to the alignment and image subtraction stages.
Respond to #4:
Thank you for your point, we have added a line to explain why the rotation of images were not a major importance for the developed algorithm due to the circumstances of image capturing conditions in mentioned in Line #115.
Respond to #5:
Thank you for your suggestion. We welcome your suggestion and will add more feature to the algorithm in the final stage of the system. Please bear in mind that this manuscript is a communication paper rather than a full Article and we aim to include more detailed findings in the upcoming article paper.
Respond to #6:
No, our observation showed no unexpected short during the image analysis of the sample images.
Respond to #7:
We just represented the numbers in two different resolutions, one in mm and the other in μm, but both represent the same value.
Round 2
Reviewer 1 Report
This paper can be published now.
