AI at the Edge is a breakthrough innovation that transforms how companies access machine data in the field. It leverages AI models — Bayesian, neural net, TinyML, GenAI, LLM, IVA, etc. to improve operations without steep network and cloud storage costs. This blog post covers its benefits, key use cases, and critical IoT elements for success in Edge AI. 

Word Count: 1,437  Reading Time: 6 minutes

Summary: 

• AI at the Edge is the capability of running AI models — Bayesian, neural net, TinyML, GenAI, LLM, IVA, etc. on assets in the field to gain real-time insights into safety, performance, and efficiency.
• Machine data is the most prevalent, critical, and challenging to access, making Edge AI impossible without a flexible, scalable IoT platform. 
• Many companies try to custom-build models, spending millions on an IoT solution that ultimately fails to scale and requires constant maintenance.
• Before deploying Edge AI, companies should seek an IoT platform that provides a no-code application for monitoring assets’ real-time status.
• Edge AI can capitalize on best-in-breed AI models for real-time inferencing, which generates real-time field insights that increase safety and performance without the expense of hosting large data sets in the cloud.
• Intelligent Video Analytics (IVA) can be leveraged in an Edge AI application, yet personal security is an important consideration.

What is AI at the Edge?

AI at the Edge, now commonly referred to as Edge AI, is the capability of running AI models — Bayesian, neural net, TinyML, GenAI, LLM, etc. on assets in the field to gain real-time insights into safety, performance, and efficiency. Assets can range widely, but equipment and machinery can gather a remarkable amount of data when equipped with a sensor or similar device.

These assets may include buildings, vessels, trucks, solar panels, motors, drills, automobiles, batteries, bridges, scooters, electrical towers, lighting, energy meters, video cameras, water filters, radios, heavy equipment, etc. 

Think of real, physical things we see every day in our world. AI models can be applied to the data of these assets to improve safety, sustainability, and efficiency. By doing so, companies can better understand, command, and control their operations. However, accessing data at this level is incredibly difficult to leverage. 

Utilizing AI at the Edge, or Edge AI, transforms how companies access machine data in the field to improve operations without steep network and cloud storage costs. First, let’s delve into the data dilemma.

AI at the Edge: A Data Dilemma

Machine data, or data generated by assets in the field, is the most prevalent form of data worldwide. Let’s look at a few of the assets and their global prevalence:

• Solar Farms: 20,601 solar farm phases in 159 countries
• Railroad Cars: 65,000+ railroad cars worldwide
• Rail Crossing: 200,000+ rail crossings in the U.S.
• High-Speed Rail Cars: 49,000+ worldwide 
• Surveillance Cameras: 1 Billion worldwide 
• Buildings: 100 billion worldwide

The list continues, yet connecting assets to software and generating meaningful insights that improve safety, sustainability, and efficiency remains elusive. Consider this quote from ClearBlade CEO Eric Simone: 

“Machine data is the most prevalent data worldwide, the most important AND the most difficult to access.” 

The reason? Before deploying a strategy to access data and apply AI at the Edge, you need a scalable, flexible IoT infrastructure that connects with any device and protocol. 

Why Reinvent the Wheel? Or, How Not to Spend Millions on IoT Software

Companies have long awaited the ability to access their machine data to monitor and control their assets in the field, let alone take a massive leap forward with Edge AI. Often, these approaches have included attempting to build IoT solutions from the ground up or utilizing consultants to build custom solutions. The inherent problem with this approach is that building from the ground up is expensive and difficult to maintain at scale. They are often stressful, requiring the architect to watch servers around the clock. Finally, homegrown IoT solutions are never done. They will eventually have new requirements that necessitate re-development.

Contrast the building approach versus a stress-tested IoT platform built on an industrial scale. One that can connect with any device (even hundreds of thousands to millions of devices), any protocol, seamless integration to any system, and a limitless ability to add extensions. Even better, add a connected digital twin for real-time monitoring and control.

Let’s dig into the connection between IoT and AI at the Edge…

Ramping to Best-In-Class Edge AI 

Edge AI gives enterprises complete flexibility to run their applications and AI models at the Edge. Edge computing and Edge AI capabilities provide fully connected or disconnected execution with automation, intelligent optimization, or self-learning systems. More advanced AI can be applied at the Edge, taking the machine data, building advanced levels of AI models, and running them in the cloud or at the Edge.

This creates a more advanced world. For all the hype and reality around AI, it takes time to get all the equipment and all the data from that equipment connected. The most prevalent data in the world is machine data, and connecting up your physical equipment is critical to executing a vision for AI in the future.

AI can become a reality faster by focusing on the Edge. First, the Edge can use legacy protocols off sensors and equipment: Modbus, BACnet, OPC UA, and any legacy enterprise or industrial protocols. This could include wireless protocols such as LoRa, BLE, and Zigbee.

Each requires an Edge to gather data and gateway processing to bring it in. Edge computing can do logical processing and data persistence. The Edge can even do smart work, like running rules and managing how much backhaul network is used.

The Edge is critical if you’re truly going to engage with the assets in the field as they are today. Once assets are connected and gathering data, data can be accessed and inferred.

Flexibility with the data inferencing layer in the field is critical to an IoT solution. An Edge solution should be able to hydrate many Edges and feed them to your IoT solution. Flexibility should also provide data as a service to third-party data sources or targets, allowing you to gather information into any cloud.

With this base IoT infrastructure, hydrating many Edges, an application layer provides no-code command and control functionality. More than a digital twin, Intelligent Assets is a no-code flexible application that’s large-screen and mobile-responsive while being easily customizable. The best part is it’s the ideal bridge between the technical prowess of the IoT solution and the everyday operators who can easily understand their assets and customize those controls. 

Edge AI: Hot Data to Inform Ops in Real-Time 

Machine data is some of the essential data to leverage in real time. By doing so, you can increase safety in the field and ensure equipment is running efficiently. This requires data inference at the Edge to tell field operators to slow down that drill, replace a fuse or wheel, inspect a bridge, turn off lights, etc.

This brings an incredible gravity to the Edge, which is why Edge AI is such a hot topic. Data inferencing—or drawing conclusions based on data—must be done as fast as possible to inform field operators of critical issues.

There are a variety of AI models, new and old, that might be appropriate depending on your use case. This can include neural nets, Bayesian algorithms, TinyML, GenAI, and LLMsor open-sourced AI models. By embedding AI at the Edge, a world of real-time insights opens up based on machine data, including: 

• Predicting pipe leaks before they happen
• Flagging a worker to get out of the way of a dangerous situation
• Monitoring bridge vibrations to detect anomalies that indicate potential failure
• Knowing when solar panels have reduced power production
• Optimize a drill to reduce wear and tear on equipment 
• Flagging a person or vehicle that is trespassing 
• Predicting when solar panel strings need repair or cleaning 

Now, consider field operations without Edge AI. Data generated by assets and machines was not accessible for real-time insights such as these. The data loads were too great to be transmitted via satellite or cellular and too expensive to send to the cloud to be analyzed. In some cases, machine data lived on physical tapes. At best, companies could glean backward-looking insights on what should have happened. By sharp contrast, Edge AI can infer terabytes of data to change behavior right at the asset level. 

Conclusion

When paired with a flexible, scalable IoT solution and a non-code application, Edge AI can close that gap, enabling field operators to take action before repairs are needed or, in the worst case, before someone gets hurt.

To learn more about AI at the Edge, contact our expert team.