Podcast: How modern networks are helping industrial AI deliver on the promise

Key takeaways

AI adoption in manufacturing requires full-stack integration ensuring seamless functionality across legacy and modern systems.
The IT/OT divide and outdated infrastructure create major obstacles, often overlooked until deployment gaps become urgent.
Scalable success begins with targeted, low-effort use cases and a dedicated, secure AI network designed for future expansion.

This episode of Great Question: A Manufacturing Podcast features a conversation with Rajeev Shah about integrating AI-driven operations in manufacturing industries. Shah is the co-founder and CEO of Celona, which offers private 5G platforms to host AI applications and promote enterprise connectivity.

Scott Achelpohl sat with Rajeev Shah to discuss opportunities for asset connectivity and AI in manufacturing, and how companies can integrate these technologies seamlessly within their operations and culture.

Below is an excerpt from the podcast:

SA: Rajeev, your company is focused on building a new kind of wireless edge with private 5G to power industrial AI. Can you lean into some of your ideas about AI and its impact on industrial network infrastructure?

RS: You can imagine being a person who is working on the manufacturing floor or in a refinery, and you’ve probably had some limited ability for a decade plus, to show up at your facility in the middle of all that steel and machinery, with a tablet and doing some sort of maintenance on a machine. You have the ability to pull up a user manual that describes how it works. That's probably not new. But just imagine trying to do that when you are on an elevated crane with all your hands taken up by tools. It’s impossible to even pull out a tablet and then read through a PDF a few hundred pages long, find the relevant information, and then use it right then and there. Why digitization is great, in the real world in these types of environments, that’s not feasible.

Now, imagine that world in today’s world of GenAI. When the same operation can be conducted through a hands-free handset, which is connected through an Android or IOS device that has a GenAI assistant running on it, you can ask a very specific question like, “tell me what’s wrong with this machine?” The GenAI assistant can give you specific information, being trained on the same manual, and you can right away take action. That is a quantum leap in what that exact same information in a PDF file could do before and can do now.

Think about the same thing with machines which have been the first generation of automated robots that are fixed in one location and can do one thing at a time. Versus, going to a true humanoid machine that is going to walk around and be able to change what it does depending on what’s happening; that is going to completely change how these factory lines look.

This is not the IoT that we’ve been talking about for the last 10 to 15 years. When something big changes in a particular industry, my observation has always been that it is not limited to one part of the technology stack. Every part, from the user experience, the devices and the machines, the chips that are used, to the way programming is done. Most relevant to me and Celona, the networking or the communications layer gets impacted.

SA: Tell me a little bit more about the legacy infrastructure hurdles that stand in the way of AI adoption in manufacturing operations.

RS: There is an absolute lack of basic understanding of what communication is required. I'll give you one very good example. We recently had fantastic technology of a humanoid robot deployed on a manufacturing floor. As they rolled it out, literally days before it was going to actually hit the factory floor, was the first conversation between the robot maker and the networking team of that big company. It was clear that there was such a gap in knowledge. The robot maker essentially expected a high bandwidth, seamless connectivity everywhere in the plant for their robot to work, and in reality, there was no such infrastructure in place and, quite honestly, some of their requirements were unrealistic for where wireless technology sits today.

There is a big gap where people who are making these robots and new physical AI models don't even know enough about the infrastructure needed at a communications layer. There are a couple of what I would call “infrastructure hurdles” that we see very often that, really interestingly, create these challenges.

Last but not least, there's still not enough attention being paid to the security aspects of these things. As we get into this world, how we secure communications and all operations of these devices, which are effectively going to run the entire operation, is still very misunderstood because people who are trying to do this are coming from a different world.

SA: I’m going to ask you to put your thought leader cap on for a minute. If you had a top five list of tips that would make AI adoption easier for a manufacturer, lots of them are just now getting into it, what would those tips be?

RS: There are a few key things I would start again by repeating: please think holistically. Think of your full stack, don't get bogged down into thinking this is just an AI model problem, or this is just a robot. It's not.

Think about the full stack; that's number one. Having said that, once you have thought about the full stack, start small. Pick an app that makes sense. We are seeing some really high value, relatively low effort, safety, and maintenance apps show up. That can easily build a business case for it without having a lot of investment. I would start a pilot with that and start thinking of how to scale it to full automation.

Third: think about security. Think about a secure network that is built for this AI and has security built into it. Don't let these new apps compete for bandwidth with your legacy network. Give it its own infrastructure and think about a secure, dedicated AI network from day one. Then, think about your next generation of operations where humans, machines, and agents are all going to work closely together. So, take it step by step and be ready to learn in every step and then pivot as needed. But I think this world is coming faster for all of us, so I would just start picking those infrastructure elements and planning for growth.

About the Podcast
Great Question: A Manufacturing Podcast offers news and information for the people who make, store and move things and those who manage and maintain the facilities where that work gets done. Manufacturers from chemical producers to automakers to machine shops can listen for critical insights into the technologies, economic conditions and best practices that can influence how to best run facilities to reach operational excellence.Listen to another episode and subscribe on your favorite podcast app