Using a self-contained vision system to inspect unlabeled cans to improve food safety
The potential liability involved in mislabeling a product which might be inadvertently ingested by someone who is allergic to the ingredients is a major concern for food and beverage producers. When you are labeling hundreds of thousands of cans at a rate of 1,000 per minute, it’s a major challenge to spot a few incorrect cans or labels that were mixed in with the batch. With machines operating at this speed, 100% manual inspection is not a viable option. The only way to be sure about the ingredients is to read the character-based product code printed on the can and match it up to the Universal Product Code (UPC) barcode on the label. This operation is called brightfield inspection because the metal can creates a bright background that makes it hard to read characters.
Brightfield inspection is so difficult that it has only been attempted a few times by complex custom vision systems that are assembled from a camera, frame grabber board and computer. More recently, the power of much more robust and easier to apply and operate self-contained vision systems has increased to the point that they are now capable of inspecting brightfield product codes at the required line speeds. Matrix Technologies has developed what is believed to be the first vision system-based brightfield inspection solution, which also reads the UPC barcode and makes sure the label is fully attached to the can. This approach provides improved regulatory compliance and traceability while being much easier to configure and use and is more reliable than systems based on vision building blocks.
The brightfield inspection challenge
Producers of canned foods typically make a large volume of a particular product, such as tomato soup, then store the cans in a warehouse without labels while waiting for orders from customers. The cans are labeled just before shipment, often with the customer’s private brand label. This process is called bright stock labeling because it begins with bright unlabeled cans. The challenge for the food producer is to make sure that each can is labeled correctly. This is important not only because customers will be unhappy if they buy a can of bean soup and discover that it is actually cream of mushroom, but an even greater concern is that they might be harmed by unknowingly ingesting an ingredient that they are allergic to. Inspection is difficult due to the large volumes and high speeds involved as well as by the challenge of inspecting the hard-to-read product code. Additionally, private brand labels usually have the same motif regardless of product which makes it problematic to tell one product label from another.
The cans go by at a speed of one every 60 milliseconds so conventional manual inspection is not possible. Traditionally, canners have addressed this challenge with a series of checks and balances. A supervisor might spot-check the finished product but it is unlikely that there is time to inspect each can. Another approach is to inspect the registration marks on the edges of the labels prior to attaching them to the cans. The marks should all line up if the labels are the same. But because the labels are very thin it’s hard to detect one or two incorrect labels. Also, this type of inspection won’t detect cans with the wrong product codes.
The only known effort at applying machine vision to this problem used a camera connected to a frame grabber board on a computer. Optical character recognition (OCR) software on the computer reads each product code and also reads the barcodes on the label, achieving the goal of providing 100% inspection of both labels and product codes. Its weakness is that the specialized hardware is not designed for use in a factory environment. The cameras and frame grabber boards are susceptible to heat and dust. A considerable level of expertise is also required to set up and maintain this type of system, expertise that is typically not found in a canning plant. Canning companies frequently use new labels and a program needs to be developed for each one.
A new approach Matrix Technologies utilized recent advances in vision system technology to develop a better approach to brightfield automated inspection.
“The key to the new approach is the use of the Cognex In-Sight 5600 vision system to inspect the product codes against the bright can background at a speed of 1000 products per hour,” says Les Haman, Department Manager for Matrix Technologies. “In-Sight 5600 vision systems offer the same rugged design and outstanding performance as the In-Sight 5400 series, but with twice the processing speed and memory to perform inspections at line rates no other vision systems can match. In-Sight vision systems are an excellent fit for the factory environment because they are completely self-contained in an IP67 (NEMA 4) rating to withstand dust and wash down without an accessory enclosure. Cognex In-Sight vision sensors also provide a software interface that simplifies setup and operation to the point that many users allow machine operators to configure the system to inspect new parts.”
Matrix Technologies’ bright stock labeling solution inspects products immediately after a label is applied to a bright product. The Cognex PatMax pattern matching tool inspects the product code. This application takes advantage of the ability of the PatMax tool to recognize a pattern regardless of its location. Rather than reading individual characters the application is configured to simply look for an image that matches the three-digit product code. A new product code can be configured simply by putting a can with the new code in position to be viewed by the vision system and positioning a rectangular box around the product code. From that point, the vision system will detect that product code even if it is in a different position or at a different angle as long as it is in the field of view. This approach is much simpler, more robust and more economical than the machine vision technology used on this application in the past.
Matrix Technologies’ bright stock inspection solution also includes a laser scanner that reads the barcode on the label of each product. A fiber optic sensor identifies labels that have not been properly glued to the can by detecting a protruding flap. A proximity sensor triggers both the vision system and the barcode reader. The vision system, barcode scanner and fiber optic sensor independently inspect each product and send pass and fail signals to the programmable logic controller (PLC) that overseas the inspection station. The pass or fail signals are buffered until the product travels to the reject mechanism. The buffer is then processed at the reject mechanisms to either allow the product to proceed or eject the product from the conveyor.
Network connectivity simplifies setup
A human-machine interface (HMI) running on a PC displays real time image updates, inspection statistics, diagnostics, and setup functions. Images are displayed on the screen with an overlay to indicate a Pass or Fail result. Running counts of passed and failed inspections as well as maximum and minimum values of the previous 10 inspections are also displayed, providing users with precise information about each failed inspection. The vision system records images of all products that fail inspection. The most common cause of failure is that the product code is not in the proper orientation so these records provide a useful tool for trouble-shooting the canning machinery. System configuration and manual control is password protected.
Corporate ERP connectivity relates product codes to UPC codes. The system also utilizes Microsoft SQL Server to centralize configuration parameters and to retain failed inspection results. Setup mode leverages the ERP and SQL Server connections to ensure that the latest updates are deployed. The vision system communicates with the PLC using static outputs and communications with the PC running the over an Ethernet connection. The operator presses a button for ‘setup mode’. The barcode from the next product that runs through the system is retrieved and is automatically used to download the correct product code. No data entry is required. All products that run through the system in ‘setup mode’ are rejected.
Matrix Technologies is in the process of deploying 10 of these systems to its initial customer. “Two of these machines are already up and running and they are working very well,” Haman concluded. “The bright stock labeling solutions have already proven their ability to provide accurate inspection results with virtually no downtime. Matching the images of the product codes has proven to be a much more reliable and robust solution than attempting to convert the images to characters. The existing plant personnel are able to maintain the equipment and program them to read new product codes without any difficulty. Based on performance criteria, benchmark testing, and user acceptance, this solution delivers an attractive value for bright stock labeling in the food and beverage industry.”
For more information, visit Cognex