Image processing Contract manufacturer produces more flexibly with AI-based image processing

Contract manufacturers rely on flexible automation to realize quick product changes while remaining economical. Therefore, the American contract manufacturer Federal Package uses AI-based vision systems from Cognex in its quality control.

Supermarket shelves around the world are full of products whose manufacturers are well-known, but in many cases, they are not the actual producers of the goods. The reason behind this seemingly paradoxical situation is that many original manufacturers, from sectors such as food, beverages, personal care, or cleaning products, have their products manufactured, filled, or even packaged by contract manufacturers for a fee. This approach offers several advantages compared to complete production on their own facilities and in their own factory halls. Among other things, this strategy allows for greater flexibility regarding production volumes and better scalability with changing demand. In many cases, it is more economical to outsource production capacities to external partners and thus keep the organization lean. For these reasons, numerous larger and smaller manufacturers are increasingly relying on such contract manufacturers.

Essential for brand image is the quality of the goods: defective products or packaging can quickly damage the reputation of the original manufacturer and result in drops in sales and profits in no time. Flawless products are also crucial for contract manufacturers, as otherwise there is a risk of losing orders.

Gallery

Increasingly automated processes

An example of this business model is the American contract manufacturer Federal Package, based in Chanhassen, Minnesota. The company specializes in the development and packaging of a wide range of health, beauty, and personal care products, including deodorants, sunscreens, lip balms, serums, and a variety of lotions and creams.

When filling such materials into their respective containers, overflow can occur at any time because the fill levels often cannot be controlled with 100 percent certainty, or droplets of the content may land on the containers.

Jerry Bilse, Senior Vice President Operations & Engineering, Federal Package

Adding to the challenge is that the company works for many manufacturers, which means that the containers, fill levels, and ingredients frequently change.

We therefore increasingly rely on automated manufacturing processes. This also applies to quality control, which we are implementing more and more frequently using industrial image processing systems.

Jerry Bilse, Senior Vice President Operations & Engineering, Federal Package

The fact that increasingly powerful AI-based vision systems are being used can be easily explained in the case of Federal Package: Overflowing or dripping material during filling results in highly variable defect patterns. In these cases, no contaminated shampoo bottle is like the other.

With the help of artificial intelligence and edge learning tools, we can easily teach such defects to the image processing system, thus achieving reliable detection of incorrectly filled containers and their fully automatic removal from the process.

Jerry Bilse, Senior Vice President Operations & Engineering, Federal Package

Edge learning allows the seamless training of good and bad examples by uploading the corresponding images into the system. "By using such systems, we are now able to switch our production lines to new products much faster than before."

In a recently conducted project, Federal Package received an order to produce deodorants for a personal care company and invested again in industrial image processing. Two tasks within the manufacturing process were particularly suitable for automating quality control: One was checking the plastic packaging on the front and back of each deodorant container for drips, which can result from overfilling and lead to poor product presentation. In the second quality control, the labels were inspected. This involves checking whether the correct label is used and whether it is applied in the correct position and alignment on the packaging. "At the production line selected for this task, the goods were previously inspected manually," recalls Bilse. "To increase productivity and enhance quality, automating the inspection of each individual package on this line was the ideal solution."

Federal Package had previously invested in industrial image processing for various business areas and had corresponding experience with Cognex and other leading providers of industrial image processing technology. "For this deodorant project, we chose Cognex again for various reasons and opted for In-Sight 2800 systems," said Bilse. According to him, essential arguments for Cognex included the hardware's performance, the software's ease of use, and the positive experiences with the manufacturer's support. Particularly, the ability to use the integrated edge learning tools of the In-Sight 2800 systems for defect training was decisive for this decision.

Edge Learning is a subset of artificial intelligence (AI) where processing takes place directly on the device, using a set of pre-trained algorithms. The technology is easy to set up and requires less time and fewer images for training compared to other AI-based solutions, such as deep learning. Choosing the In-Sight 2800 significantly accelerated the commissioning process, Bilse emphasizes.

The image processing component of this system was set up and ready to use within just 20 minutes – previously, such changes took us up to a whole day.

Jerry Bilse, Senior Vice President Operations & Engineering, Federal Package

Such short changeover times are crucial for Federal Package, as the company produces between 30 and 40 different deodorant designs with varying scents, colors, and labels for different manufacturers. The system's edge learning capability is smart enough to filter out features like different colors and labels and focus on the actual defects of the product.

Federal Package uses the monochrome version of the In-Sight 2800 for drip inspection and the color version for label control, as it can better differentiate between the edge of the label and the deodorant container. The line for drip inspection runs at nearly eighty units per minute, while the label inspection operates at about sixty units per minute. In both cases, automation allows for a higher throughput and ensures that all units are checked 100 percent, which would be impossible manually. The accuracy rate for detecting defects is over 99 percent, meaning that the operating personnel only need to intervene in the few cases where defective containers are identified and rejected.

Bilse is thrilled with the results of the upgraded system: "This image processing system has helped us further optimize our processes and elevate them to an even better level. The production line, where deodorants are now filled, typically runs 24 hours a day, five days a week, without requiring a single manual intervention. This would be impossible without suitable image processing systems."

In addition to increased reliability, Federal Package benefits in another way from using Cognex image processing: "Over time, we have gained more confidence in our systems and know that, due to the high level of automation and powerful image processing, we produce virtually no scrap. On this basis, we now operate the systems at slightly higher speeds, further increasing our throughput and productivity. We owe this success also to the experience of Cognex experts who support us in becoming even better. In this area, I see tremendous added value for the future."