Real-time vision in the machine process

Unlike conventional solutions that rely on external PCs or controllers, B&R 's Smart Camera processes image data locally. AI inference takes place in real time - powered by an edge AI processor that is up to 15 times more efficient than comparable chips, according to Hailo benchmarks. A variety of AI-supported image processing functions are supported with models from MVTec and others - including optical character recognition (OCR), anomaly detection, object recognition, classification and semantic segmentation. These functions can be combined with conventional, rule-based algorithms in an end-to-end, hybrid workflow. According to the manufacturer, the solution combines the flexibility of AI with the speed and precision of conventional image processing. This hybrid approach is particularly suitable for complex inspection tasks, such as identifying product types, detecting defects or checking printed characters. The Smart Camera is designed for rapid commissioning and is configured using a tool integrated in 'Automation Studio'. A low-code environment allows automation engineers to configure AI vision tasks, simulate results and manage models without specialist knowledge. Existing smart camera installations can be easily upgraded via a drop-in upgrade.

"Our new smart camera can be integrated directly into the control loop so that machines can be adjusted during operation," says Florian Schneeberger, Chief Technology Officer, ABB Machine Automation (B&R). "With our solution, machines not only detect product defects, but also react to them immediately."
Another advantage of the smart camera is its flexible use. "Customers can change models and combine AI functions in user-defined sequences - without interrupting production," adds Schneeberger. This feature makes it possible to quickly adapt the process to new product variants or to detect new types of defects. The solution also supports the gradual integration of AI into existing rule-based systems.