Advanced AI solutions for fast and efficient packaging inspection

Since the beginning of the Covid-19 pandemic, the packaging market has soared in value. Between lockdowns and social distancing, and the resulting need to secure food and supplies at home, the EU alone generated 25% more packaging material in 2021 compared to a decade earlier. The same trends, meanwhile, are reflected worldwide. According to work by Mordor Intelligence, global packaging is now worth $1.1tn, and could grow another $200bn by 2028.

Yet, if packaging is now a sprawling global industry, offering vast opportunities for manufacturers and suppliers alike, shippers are increasingly under added pressure. That’s true across a variety of areas, from ensuring labels are readable to checking the quality of boxes or bags before they reach customers. And though technology to support packaging has existed for decades, recent years have seen remarkable advances – as one US company is elegantly proving. 

Binary options

Before packages can be sent out to customers, they need to be checked. Perhaps the most obvious example here is the packaging itself. Deliver a ripped bag or leaky box and the financial cost can be painful, especially when research suggests that returned items are only worth two-thirds of their original price. It’s a similar story in-house too. If, for instance, a faulty food container ruptures mid-conveyer belt, it risks damaging millions of dollars of equipment. 

Just as importantly, stakeholders need to certify that packages are traceable. If a barcode is faulty, or otherwise unreadable, how can it reach its destination safely – or protect shippers from angry customers? “Shippers want the peace of mind to say: ‘When this package left my warehouse, it was fine,’” is how Gregory Lewis puts it. “‘And here is a visual image captured to confirm it was great when it left.’”

Nor is the executive channel account manager at Zebra Technologies alone in this. As Future Market Insights recently uncovered, the packaging inspection systems sector is already worth nearly $500m – and could reach almost $1bn over the next decade.

All the same, traditional packaging inspection systems have suffered from a range of limitations. Fundamentally, this is down to the technology underpinning them. Designed around a set of rigid rules, their conclusions are binary: either a package is acceptable, or it isn’t. 

The issue, Lewis explains, comes when managers try to understand exactly what the problem entails. “Is there water damage, are the boxes torn, or are the labels hanging off the bags?” he asks, speaking rhetorically. It goes without saying that, in many cases, getting to the bottom of a complication requires what Lewis calls the “human eyeball” – wasting time and money better spent on other things. 

Another difficulty with older models involves the immense variation of modern packaging. Though binary vision systems are good at examining predictable items like cardboard boxes, plastic bags crumpled a thousand different ways are far harder to parse.

It’s a similar story, moreover, with traceability inspectors. “A normal barcode on a white background?” asks Lewis. No problem. “But a laser-etched 2D barcode on a rough piece of metal? That’s pretty ugly. Traditionally, that would be quite difficult to pick out.”

The power of AI

How are shippers approaching these varied challenges? In a phrase: deep learning. Leveraging the latest advances in AI, and partnering with experts in the field, they’re deploying machine vision solutions that can dramatically improve the inspection process.

Zebra Technologies, for its part, is at the centre of these developments. From its Illinois headquarters, and enjoying a legacy stretching back 50 years, the company offers a range of cutting-edge platforms – which together can make packaging inspection faster and more efficient. 

Consider, for instance, Zebra’s Aurora machine vision software suite. A deep algorithmic system, it enables users to integrate powerful packaging inspection capabilities into their production lines. Because it’s based on AI, Aurora isn’t limited to ‘yes’ or ‘no answers – but can rather be trained on real-world data to understand packaging in all its complexity.

This has important consequences in practice. By way of example, Lewis highlights plastic bags. Under rules-based systems, the technology would struggle to distinguish a merely crumpled bag from a genuinely ripped one. Machine learning has no such issues, saving a heap of manpower in an instant.

Aurora can also differentiate between various problems – for instance, if a box is wet or ripped. Just as importantly, Lewis adds, Aurora is user-friendly, ensuring the AI can be trained to only flag exceptions that matter. “Luxury consumer brands may want a perfect box going out,” he says, which may not be the case for more humdrum products. 

The same principles can be applied to traceability. Apart from easily being trained to examine rough metal barcodes, Zebra’s Deep Learning Optical Character Recognition platform can scan actual letters equally well. Among other things, that’s invaluable in food production, for example when spotting misspelt or missing allergen warnings. 

Nor do shippers need to invest in multiple machines here. Describing Zebra’s range of Iris GTX smart cameras, Lewis says they can “do barcode reading, colour checking, and visual inspections all in one go” – a far cry from older setups, which may have obliged shippers to install two or even three separate machines. 

In a similar vein, Zebra’s machine-vision software development kit has a deep collection of tools for image capture, processing and archiving, ideal for chargebacks or proofs. 

Augmenting the technology

All this has clear and immediate consequences for shippers that choose Zebra. Some advantages are obvious: delegating time-consuming inspection tasks to AI will inevitably bolster a company’s balance sheet. Lewis, for his part, envisages situations where workers can check thousands of packages armed with only a forklift, scanning down the tunnel as they drive. 

Altogether, Lewis says, that can save a fifth of the cost on traceability tasks, particularly crucial when logistics margins can be as low as 1%. It helps that AI systems can get complex inspection protocols up and running in as little as half a day. That’s compared with days or weeks for rules-based platforms, systems that need to be totally reconfigured if anything changes. 

At the same time, there’s evidence that Zebra’s technical wizardry will only mature over time. Quite apart from the algorithms themselves – as they’re exposed to more data, the AI will inevitably sharpen – that’s true in terms of R&D. 

Zebra employs an international engineering team dedicated to AI-related R&D. That’s unusual, Lewis notes – most packaging inspection technology firms don’t have so many coders working on the same issue at once, especially when you factor in the quality of Zebra’s physical machinery. 

That’s echoed by robust customer service. “Something that is fairly unique to Zebra is that we sell support packages at the item level,” Lewis says, adding that elsewhere in the industry, warranties only extend to manufacturer error. If, in other words, a forklift accidentally crashes into a piece of hardware, the shipper will have to swallow the cost – not a concern when working with Zebra.

Combined with an extensive network of industry contacts – Zebra boasts 10,000 partners in over 100 countries – it means any technical issue can be resolved at speed. 

As Lewis says, that’s obviously useful all year round, but especially at Christmas and other busy shipping periods, when “people want you there within a day”. Considering how popular Zebra’s work has proved already, don’t be surprised if they’re invited to stick around for a drink. 

To learn more about Advanced AI solutions for fast and efficient packaging inspection please click here to download this Zebra produced White Paper."