Key takeaways from Zebra Technologies’ webinar, in association with MVPro Media, on the benefits of using machine vision innovations in the packaging sector.
Earlier this year, MVPro Media and Zebra Technologies collaborated to produce a webinar on the challenges of packaging inspection, and the different solutions available using Machine Vision technology.
The webinar was hosted by Zebra Technologies’ Tom Lambert, Zebra’s EMEA Machine Vision Marketing Lead. Joining him to provide expert insight on the technical aspects was Ivar Keulers, EMEA Solution Engineering Manager at Zebra.
The discussion began with a look at the main challenges faced in packaging inspection, including regulatory requirements, quality verification and traceability information.
All packaging contains critical information which must be present on every item, with consequences for the manufacturer and the consumer if any is missing. Tom explained the importance of including accurate regulatory information, such as best-before dates, allergy information, and recycling information.
He added: “Across every region, where regulatory bodies are mandating the inclusion of this critical information, we need to make sure that it’s present. A lot of this can be embedded in the code, but best-before dates and allergy information cannot be verified with a barcode, … it needs to be human-readable.”
“If we miss something critical like allergy information, that can pose a public health risk. Similarly, if we’re checking products, we need to ensure there are no contaminants, especially in food or anything being consumed.
“Then, if anything goes wrong, we need to be able to trace each batch, or each individual item depending on the industry. This is especially important for the EU, where we will soon have the Digital Product Passport Legislation which requires each individual product will be traced.”
The Rules-Based Approach
Tom then handed over to Ivar, to explain the different machine vision solutions available to address these challenges. To begin, he explained traditional rules-based inspection, providing an example of a typical inspection cell which is used for food and packaging lines to inspect the total package.
The example came from Clearview, a UK-based Zebra specialist Machine Vision Industrial Automation partner. Ivar shared an example of packaging inspection using the rule-based technique.
The example inspection is performed using Zebra’s Aurora Image Library, a software package with advanced text and barcode reading algorithms. Ivar explained how the software allows comprehensive inspection to ensure label elements such as best-before dates, codes, and pricing promotions are present and correct. The software also offers high accuracy at high speeds which makes it the preferred option for inspection of products with consistent packaging presentation and low variability.
Ivar said: “It can reach up to 30 parts per second if needed. Once you create an application with slight changes, you can add another label, and a third, perhaps even 50 labels.
“If the labels or the product changes a lot, it requires a lot of programming to make that rule-based inspection work. That’s one of the things that we want to highlight in our new technology which uses artificial intelligence and deep learning to make it possible to carry inspections that are not as strict as this example.”
Rule-Based or Deep Learning?
This takes us to deep learning tools, which are preferable when the inspection requires more variation. Tom explained that rule-based inspection is highly effective if the product is consistent, but if the products are highly varied, using rule-based inspection can be impractical, making Zebra’s deep learning tools an attractive prospect.
Tom explained: “At Zebra, we have deep learning-based tools which do all the training for you. For every product, you need to have a machine-readable element for traceability, but you might also need a human-readable element. If the font or information changes a lot, then a rule-based system is not practical, because of the level of complexity and programming involved.”
“We’ve developed a drag-and-drop tool, so you can program in a barcode read and use that as a reference point to read a piece of human-readable text, such as a best-before date. This can be done simply by using the drag-and-drop interface.”
Applications can be built in a few clicks with this software, with thousands of fonts already pre-trained. This means users will no longer need to go through the training required with a standard machine vision AI tool.
Tom added: “It will deliver industrial-grade quality without having to do any programming and it can work in high-speed industrial applications. It’s highly flexible and can work across multiple platforms.”
Avoiding Damages with AI-Powered Quality Inspection
Tom then introduced full AI-powered machine vision solutions for packaging, drawing on examples from the logistics industry.
One of the biggest challenges for the logistics industry is the sheer amount of money wasted due to damaged packaging. He stated that missed damage to boxes in logistics costs the industry more than $1 billion every year. He pointed out that, in the past, the process of checking boxes for damages has been completely manual due to the variability involved. However, advancements in AI have allowed for major improvements in this area. AI-powered tools can be used to make subjective judgements on damage, much like how a human would.
Tom explained: “We’re able to perform quality inspection even in this very extreme, highly variable environment, and detect defects, which has made huge improvements to the processes of some of our logistics customers.”
Together, Tom and Ivar discussed one of Zebra’s partners, Denali, which has developed an AI-based system for this exact problem.
Tom said: “They’ve used machine vision technology to train a system to identify normal, torn, and crushed boxes and categorise them into these three categories with 98.6% accuracy. They’ve done that using a sample of 800 various boxes. That solution has reduced operational costs for customers by 13% and saved them millions of dollars in concessions.”
Denali’s system can also support additional functions, such as liquid damage inspection, which is highly valuable in packaging. If liquid damage were to go unnoticed for too long, it could cause damage to the products as well as the machinery, which would be very costly to fix.
Ivar then provided his technical insight into how this system works. He explained that Denali created a six-sided scan tunnel, where several cameras scan the box at different angles to check for damages. The conveyor is opened from below to allow the base of the box to be scanned using a line scan camera. The images are then checked using Zebra’s AI deep learning application to determine which of the three trained categories best reflects the state of the package.
Ivarsaid: “The software from Zebra is called Aurora Design Assistant. It is a drag-and-drop software with different tools, one of which is AI deep-learning classification. They have trained all kinds of different products, which we classify into the three categories. The AI will capture the images from six sides, and then decides what fits best, and decides if it’s a normal, crushed or torn box.”
Before moving on to questions, Tom concluded by clarifying which option is best suitable for clients. He noted that the answer depends on what you are trying to inspect, stating that while consistent product inspection can be carried out with the rules-based method, a highly variable product will likely require deep learning or even artificial intelligence to deliver consistent accuracy.
He added: “The most important thing is having the right expertise to help you make that judgement and support you in getting the best system and technology in place to support what you’re trying to achieve.
“At Zebra, we have a full portfolio of hardware and software and most importantly, an ecosystem of partners and system integrators that are here to support you. If you are looking at packaging inspection applications or you need support in delivering a new system or updating an old one, we are here to help you.”
In the final section of the webinar, Tom opened the floor for questions from the audience. One viewer asked for clarification on when you should choose an AI-based system over a rule-based system.
In response to this, Ivar reiterated that it depends on the required flexibility of the inspection, before providing an excellent example of a case when AI-based systems are the best option, He referred to a Zebra partner based in Portugal which is responsible for sorting tuna.
Ivar stated: “I can’t tell the difference, but the tuna is a different size or colour. The system recognises which type of tuna it is, which you can’t do using rule-based algorithms. So, if you have a lot of variation like this, deep learning AI is the way to go.”
Conclusion
All-in-all, Zebra Technologies’ packaging-focused webinar provided excellent insight into the benefits and pitfalls of different approaches to machine vision-powered packaging inspection and provided viewers with helpful advice on which method is best suitable for their specific requirements.
In fielding the viewer questions, Tom and Ivar were able to alleviate any additional concerns, while providing real-world examples of companies who are benefitting from the advancements of AI in packaging inspection.
We hope you enjoy watching the webinar as much as we did.
Check out the full webinar for yourself.
If you missed the webinar or would like to watch it again, the full recording is available here
