AI is moving off the server rack and onto the plant floor, where machines run and decisions have to be made in milliseconds. In a recent episode of The Automation Podcast, host Shawn Tierney spoke with Emerson’s Gene Juknevicius, Solution Architect, and Stephan Reichenauer, who leads business development and sales in EMEA for Sima.ai, about how the two companies are working together to bring physical AI to the industrial edge.

Why It Matters

Traditional machine intelligence, such as a built-in loop tuner or a pass/fail vision camera, depends on hard-coded rules. Physical AI uses vision and object detection models that adapt to what they see and feed results back to the control system in real time. For automation engineers, that makes closed-loop quality control at the machine practical. Doing it well takes compute that is fast, power-efficient, and rugged enough for harsh industrial environments.

Key Takeaways

• Physical AI perceives, decides, and interacts with the real world rather than living only on a screen.
• Low latency is essential. There is no time to send data to a data center and wait for an answer.
• Fanless, passively cooled PACSystems industrial PCs (IPCs) operate from -40 to +70°C and accept a Sima.ai PCI Express AI card in modular extension slots.
• ai’s Modalix chip is a machine learning system on chip built for performance per watt, which also eases heat management.
• AI results reach the PLC over OPC UA, MQTT, Modbus, or other fieldbuses to close the control loop.

What Makes AI “Physical”

Stephan described physical AI as intelligence built into products that perceive, decide, and act in the real world, from industrial PCs to drones, robots, and vehicles. Gene framed it from the hardware side: the goal is to bring AI close to where the industrial action happens, next to the material being observed and the machines making decisions.

Gene offered a mining example. A camera watches rocks on a conveyor belt, an object detection model estimates their size, and the result goes to the PLC over OPC UA or MQTT. If the rocks are too large or too small, the control system adjusts the mining equipment in real time.

Purpose-Built Silicon Instead of a GPU

Shawn asked whether running AI simply means adding a good graphics processing unit (GPU). Stephan explained that the Modalix chip is more than an accelerator. It combines an accelerator, an ARM processing subsystem, and a digital signal processing (DSP) vector processor, and Sima.ai’s software development kit assigns each function in an application pipeline to the core best suited to run it. The result is high performance, low latency, and low power consumption.

Gene added that GPU environmental specifications fall well short of what industrial settings demand. A GPU can work in an air-cooled control cabinet, but getting close to the machine is a different challenge.

Rugged Industrial PCs Built for AI Workloads

Emerson’s industrial PCs are designed to deliver high compute performance with no fans or active cooling while operating from -40 to +70°C. Gene noted that the latest generation was designed specifically to enable AI workloads, using a modular concept with extension slots that hold the Sima.ai PCI Express card. The PACSystems portfolio spans industrial PCs from entry-level to AI acceleration, built for consistent performance, software compatibility, and deployment flexibility.

Closing the Loop on Pipe Quality

One application involves high-pressure pipe reinforced with tape wound around it. A camera measures the gap between windings, and the IPC sends that measurement to the PLC controlling the winding operation so it can adjust in real time. Because the pipe next goes into an oven, catching a bad wind early avoids scrapping long lengths of finished product. Stephan added that timestamped video gives the manufacturer a record showing each section was produced correctly.

Listen and Learn More

Listen to the full episode of The Automation Podcast, and visit the PACSystems section on Emerson.com to learn more about uniting industrial-grade control with AI-ready edge intelligence and intuitive software tools for smarter, more efficient operations.