Leveraging AI to Supercharge Existing Fabriq Vision Systems

May 13, 2022 | Kevin Bird

 

Need help on eliminating false positives from your fabric defect detection process? You're not alone. 


If you are a manufacturing company that uses an Uster Fabriq Vision System to identify defects in your product, but you still have a high amount of returned product or are still required to review the images after the roll has been cleaned by Fabriq, I would like to share an alternate solution.  You can start leveraging a computer vision model to eliminate the need for human inspection while also improving the quality of the inspection process.
 

Mariner is a company that specializes in taking images from legacy vision systems and applying new computer vision techniques that have been developed in the past few years. I’m not going to oversell you on our solution but instead am going to give you five tips that we have learned while working with companies that are spending hundreds of thousands of dollars on human capital to inspect fabric and STILL shipping defective product.   

  1. Don’t label too much of any single defect map 
    If your company has a large variety of data, don’t label too much from any one dataset.  One strategy that we use to do this is to select a random 1% of the images from each defect map that Fabriq produces.  This ensures that no single roll dominates the labeled images.  

  2. Include your Subject Matter Expert from the start 
    If you have somebody that is really good at determining quality issues based on the images, include them in the meetings from the beginning.  Ask them what defects they expect to see, which defects the manual reviewers have trouble determining, and if there are any defects that are easier to see in certain cameras.  This is valuable information when validating that the model is performing at a level that at least matches what a human is able to do.   

  3. Don’t focus too much on the computer vision model 
    It is important that the computer vision model does a good job at accurately identifying defects, but the rest of the process is even more important to ensure that the defects are being properly handled once they are identified. This is also a good time to mention the importance of having a feedback loop that a human is involved in.  It is great to reduce the amount of human effort required to ship high-quality products, but this won’t eliminate the need for a human to ensure the process is performing adequately.  Make sure you allow a human to double check the computer and give those results back to the model to help improve performance over time.   

  4. Partner with a company that has these skills 
    I know I said I wasn’t going to oversell our solution, but this is one point where I want to add a little sales.  Just like you went out to the market to buy a vision system because your company isn’t experts at hardware, I would recommend finding a partner to help you with the Computer Vision Model.  It isn’t that you can’t do this on your own, and if you have the expertise internally then feel free to ignore this advice, but training a computer vision model that performs well is just one piece of the puzzle when it comes to adding AI to an existing vision system.  Besides this, you also have to figure out how to read the data that is coming from your vision system, find a repeatable process to read it in, and determine the best way to run the images through your model. Each of these steps takes time and knowledge that you may or may not have skillsets for internally.  It’s the classic “build-vs-buy" dilemma, but it really is easier to partner with a company that has done this before and leverage their expertise.  If this seems like an expensive endeavor, consider the customer satisfaction improvement, the reduction in manual labor, the reduction in returns, and the increase in product output that you can achieve – all of which go directly to your bottom line. 

  5. Have a weekly check-in meeting 
    Don’t forget to have a weekly meeting to discuss any roadblocks with your team.  This is a great cadence call to make sure the project is on track and to review how things are going overall.  We have had great success using a project tracking tool called Trello for keeping track of to-do items and who is responsible for what task, but you might have another solution that you’re already using internally. If so, that’s fantastic – the important point here isn’t the specific tool, but rather that you identify and break out anything that’s important for your team to track. 
 
Conclusion 


I hope these tips resonate with you. Whether you are using an Uster Fabriq Vision System or another vision system, if you are unhappy with the value that you are receiving after the images have been taken you might be missing out on a huge opportunity to improve your system with Artificial Intelligence. 
 

We’ll be happy to talk the whole process over with you from start to finish, and how it might benefit your production lines – just shoot a quick email to sales@mariner-usa.com and we’ll get the ball rolling. 

And if you’re interested in a real-world customer story on how we helped Sage Automotive Interiors with their Fabriq systems, be sure and visit the case study that Microsoft wrote on Sage and Spyglass Visual Inspection. It’s eye-opening (see what I did with the vision joke right there?), and you just might find that your textile production lines are suffering from the same problems that Sage was experiencing.  

Reading this before May 17, 2022? We'll be at TechTextil May 17 - 19, so if you have any questions please do drop by Booth 2923. Need free passes? We can hook you up with that, too -- shoot an email to sales@mariner-usa.com and let us know, and we'll see you there!

author

Kevin Bird

Kevin spent 4 years in the transportation industry and another 2 years in the utility industry. He has been working on problems in the manufacturing industry for the past year. In his career, Kevin has focused on automation, applying academic research to business problems, and parallelization. Kevin focuses on unlocking and maximizing subject matter expert impact by reducing the technical barriers. Kevin is a problem solver that focuses on iterative improvements in whatever projects he works on. Currently, Kevin is a Senior Machine Learning Engineer at Mariner-USA and a Co-Founder of The Problem Solvers Guild.