Complex tasks such as machine/deep learning, predictive maintenance or visual inspection are increasingly moving from the cloud to decentralized processing on site at the edge. "As a result, the demands on our range of edge computing devices are increasing enormously," says Norbert Hauser. "Our AI strategy therefore includes all services from consulting, hardware and software development to the training of neural networks by our own data scientists."

Norbert Hauser : "Many AI tasks are moving from the cloud to the edge."

© Kontron

At SPS, Kontron announced a strategic technology partnership with the Israeli AI company Hailo. The collaboration focuses on high-performance AI acceleration solutions. "With 26 Tera Operations Per Second (TOPS) at less than 2.5 watts, the Hailo-8 AI processor is currently the most powerful AI accelerator in its class," enthuses Hauser and continues: "With this performance booster for X86 and Arm systems, completely new edge AI applications can be realized that were previously only possible in cloud environments." One of the strengths is the real-time inference processing of high-resolution video streams, which enables object recognition or semantic segmentation over long distances. As part of the partnership, Kontron is now presenting two 'AI-ready' products with an integrated Hailo-8 chip. Hailo is also a sub-exhibitor at the Kontron booth and will show a live AI demo.

Real-time control with AI

Together with technology partner Intel, a Kontron C-Series KBox with an 11th generation Intel Core processor and the Intel Edge Controls for Industrial Software Framework was used to create what Hauser calls an "Industrial Next-Generation Smart Factory (NGSF) Edge Compute System" for real-time control applications with artificial intelligence. As part of workload consolidation, control, visualization, TSN communication and AI functionality can now be implemented as an integral solution with a multi-core processor. With up to eight computing cores, the systems are suitable for high-end applications with high bandwidth. They are supported by Intel Iris XeGraphics and Intel Deep Learning Boost for increased AI performance.

Kontron will continue to use a similar Intel Gen 11-based AI vision system based on Intel Edge Insights in its own paperless and networked system production for the visual inspection of systems. Six IP cameras are used to inspect the systems on a glass table from all sides and, using the barcode on the system and a correspondingly trained neural network, the interfaces, connectors, labels and screws required for each variant are recognized and checked at the respective positions.

These AI solutions round off Kontron's spectrum in the upper performance range. In the lower and middle, cost-sensitive range of AI solutions, two platforms - one with an Arm processor and the other with an Intel Atom processor - are already in the program. Each with an integrated Google Coral Edge TPU (Tensor Processing Unit) and Tensorflow software platform.