Boost quality with a no-code, AI-powered vision system

A recent customer survey by Rockwell Automation shows that improving product quality is considered to be the top accelerant for manufacturing’s digital transition. In addition, survey respondents said the number one area of impact for artificial intelligence (AI) is in closed-loop quality control.

At Automation Fair 2024, Amanda Thompson, Rockwell Automation product manager for FactoryTalk Analytics VisionAI, held a session highlighting the newest edition of VisionAI that was released in September 2024. This new release of VisionAI addresses the quality and AI capabilities ranked so highly by Rockwell Automation’s customers.

Thompson explained that, while VisionAI does leverage machine vision technology, it’s more than a machine vision system. “This is a quality inspection platform designed to help you understand the quality of the product you're producing,” she said. “It not only tells you if a part or product is good or bad, it also tells you why.”

She added that the system can support vision inspection at line speeds up to 500-600 parts per minute, depending on the application, to read barcodes, classify parts for identification and sorting, detect defects on surfaces, perform presence/absence detection and read/verify text.

Beyond inspection, VisionAI also provides traceability so that users can see the images highlighted by the system as defective to drive comparisons with acceptable images and provide proof of specific defects. “VisionAI features root cause analysis capabilities that will tell you why an item failed, and it provides notifications so that detected issues can be addressed immediately,” Thompson said.

In keeping with Rockwell Automation’s direction of developing AI-powered technologies that depend on end user domain expertise to refine them, Thompson pointed out that VisionAI is a no-code platform designed for operations and quality control personnel and “relies on your expertise to provide the information needed for the system to make good quality decisions.”

VisionAI architecture and operation

The architecture of VisionAI stretches from the cloud to the edge, incorporating embedded analytics, data storage, and application programming interfaces for integrations with manufacturing execution and enterprise resource planning systems. The cloud hosts the AI engine and remote access features for monitoring, management, and deployment of VisionAI's edge functionalities. At the edge are local hardware such as edge computers, human machine interfaces, PLC integration and the vision system cameras, lenses and lighting.

Thompson noted that VisionAI currently works with Basler cameras but support for more third-party vision systems will be announced in 2025.

Explaining how VisionAI works, Thompson said the first step is for the user to “define the type of inspection they want VisionAI to run. Then they can start capturing images and labeling them as good or bad. The number of images needed to train the system will depend on the type of inspection.”

She added that VisionAI provides a chart for image training that shows the user when enough images have been supplied to successfully train the model. A training report generated by VisionAI shows the model accuracy as it’s developed to indicate if you’ve labeled images incorrectly and allows users to correct them to improve accuracy before deployment.

Once the cloud-based model has been trained, it’s then deployed to the edge where up to eight cameras can be supported on one system, with live camera feed capabilities available on site and via remote access.

“The defect carousel feature shows failed images, which lets users double-click on the images to learn more about the failure reasons,” she said. “VisionAI also aggregates data across systems and time periods so that the system’s dashboard supports comparison of batches, stations, production days and other factors to support specific troubleshooting needs.”

Quality reports are generated directly by VisionAI so that every stakeholder receives the same report.

Thompson added that VisionAI’s remote access also lets users label images so that they can train and deploy models even if they are not on site. “Plus, version control allows users to see when models were changed so that if a new model doesn’t perform as expected, you can revert back to a previously deployed model,” she said.

Industry applications

Thompson noted several different industry applications suitable for VisionAI, including:

Packaging. VisionAI can do numerous packaging inspections, such as checking a bottle for defects, checking the expiration date and time stamps on labels, and assessing a cap to see if it’s twisted correctly or leaking. Thompson pointed out that all of these inspections can be done at the same time with VisionAI to provide one quality result for the product. She added that VisionAI works on a recipe concept. “If you are running different shampoo brands on a line and you want to run a slightly different set of inspections, you can create an inspection recipe per product or SKU,” she said.

Food and Beverage. VisionAI can identify issues even on complex backgrounds. For example, it can be used to inspect cereal as it passes on a conveyor to verify correct shape and color and to flag imperfections or foreign objects. It can also assess the placement of toppings on a product to identify issues like gaps in topping placement on bagels or doughnuts.

Automotive. Applications here can include presence/absence of required labels, verification of correct label placement, identification of wrinkles/bunching on car seat covers and note the specific area on the seat where these wrinkles are located for ease of physical assessment.