If expected cookies don’t arrive, customer satisfaction crumbles. But ensuring a happy outcome is often a tall order. Evans Distribution Systems Inc. of Melvindale, Mich., used to manually perform quality control to help ensure the arrival of correct cookie orders. Then the company switched to machine vision and got better results. Evans Distribution, which provides third-party logistics and supply solutions by packing and shipping in response to customer orders, sees the bulk of its business from the automotive industry, according to Stephen Ruch, vice president of warehousing. Using a camera, a vision system detects and identifies boxes. Courtesy of ADLINK. Over the past few years, though, the company has expanded the industries it serves. So, Evans Distribution now fills consumer orders for seasonal cookie offerings, with this line of business a growing part of company revenues. But it must get the fundamentals exactly right, Ruch said. “What we’re trying to do is to ensure the cookies that are ordered are the cookies in the box. Nothing more and nothing less.” Verifying box contents can be difficult — the number and types of product must match customer orders. During the busiest time of year, a shipping box full of varied boxes of cookies must be inspected every three to six seconds, and the results compared to the customer order. This process runs 10 hours a day, with up to 20,000 orders shipped daily. In the past, Evans Distribution handled this task using highly experienced senior personnel, with one employee scanning a barcode on the side of the shipping box and two others checking box contents. The barcode provided order information, which was checked against what was inside the box. Due to fatigue, these employees had to be rotated in and out of the inspection point, which meant that the inspection process tied up a considerable number of skilled workers. Based on previous job experience, Ruch believed that machine vision could offer a better way to handle this necessary quality control function. Evans Distribution therefore turned to ADLINK Technology Inc., a supplier of machine vision AI systems with a U.S. headquarters in San Jose, Calif. Machine learning AI vision systems use two cameras, one on the side and another above, to ensure that shipping box contents are correct. Courtesy of Evans Distribution Systems. Daniel Collins, the company’s senior director of IoT solutions, said that the solution eventually implemented was based on the ADLINK Edge Smart Pallet technology, with some of the application straightforward. Existing tools, for instance, made reading the barcode on the side of the shipping box relatively simple. Determining what was in the box, however, was not so easy. The varied products came in different-size boxes and were arranged in various ways inside the shipping box. ADLINK used a neural net and machine learning to solve this problem, training the system with images from actual production. Changes were then made to constrain Evans Distribution’s process somewhat, for the benefits it would bring. “The more static the environment, the easier this whole thing becomes — the less training that you have to do and the less horsepower you need to run the [classification] model effectively and efficiently,” Collins said. So, ADLINK and Evans Distribution instituted several changes. For instance, the warehouse had variable lighting due to windows and the movement of the sun throughout the day. Putting up a curtain minimized this issue. Another change involved training and operations: Product pickers were instructed to orient the cookie boxes in a consistent way within the shipping boxes. The cookie boxes could still fall over during movement, but this restriction on placement simplified the possible configurations and the classification challenge. A consistent box-loading pattern reduced the machine learning training time required to get the system up and running, allowing this to happen in a two-month window. In operation, the system uses a side-mounted vision system to read a packed box’s barcode and determine which customer order should be in the box. It uses a camera located above the line to compare what was requested to what’s in the box, and it indicates whether the result is a match. If it’s not, the system alerts staff and lists the missing cookies. Although there are plans for further refinements to the system, the machine vision solution is already a success, according to Ruch. Based on comparison runs during the most recent shipping season, the quality of the automated inspection was higher than that of the manual approach. Also, the labor savings alone will pay back the system cost in less than a year. Another benefit could be as appetizing as the cookies themselves. While speaking of the labor savings and the machine vision system payoff, Ruch added, “That doesn’t even count the improved customer service.”