www.eejournal.com, Sept. 17, 2024 –
On occasion I amaze even myself. Sometimes this is in a good way. Other times... not so much. My wife (Gina the Gorgeous) often says I'm clueless as to what's going on around me (she says this with love). I can't argue with her. Well, I can try, but I never win, so what's the point?
One of the things I'm particularly good at is not having much of a clue about what behemoth corporations are doing at the enterprise level. I have, of course, heard the name Nippon Telegraph and Telephone Corporation (NTT), and I was aware that this was a Japanese telecommunications company, but I was unaware that it boasts ~310,000 employees and is a ~$110 billion entity.
Similarly, I was unaware that NTT Data, which is a partially-owned subsidiary of NTT, is a $30 billion company with ~170,000 employees operating in 50+ countries. The reason I now know so much more about this stuff than I did before is that I was just chatting with Parm Sandhu, who is Vice President, Enterprise 5G Products and Services at NTT DATA.
As an aside, if you use the internet (and who doesn't?), then your communications are probably flowing through NTT DATA servers. Also, it's often said that NTT created the first emojis, which are pictograms, logograms, or ideograms that can be embedded in things like text messages and on web pages. It's true that Shigetaka Kurita from NTT DoCoMo (the mobile operator of the NTT Group and partner company of NTT DATA) created a set of emojis that were sent as part of text messages via mobile communications in 1999. However, although Kurita's emoji set included sports, actions, and weather, notably absent from the set were any emojis that actually reflected emotion. Furthermore, there were some earlier contenders, but we digress...
The reason for our chat was for Parm to bring me up to date with NTT DATA's recently announced Ultralight EDGA AI Platform. This little scamp–which is also known as SPEKTRA (Sentient Platform for Network Transformation)–boasts an Intel Elkhart Lake x6425E Atom processor (4C 2.0GHz), 4GB DDR4 SODIMM-290, 32GB SSD M 2 SATA TPM 2.0, 2 x GbE, WiFi / Bluetooth, optional Power-over-Ethernet (PoE), and optional cellular connectivity (4G or 5G).
Although the artificial intelligence (AI) spotlight has been on GenAI and Large Language Models (LLMs) recently, these technologies are largely impractical for industries requiring real-time and local decision-making. NTT DATA's Edge AI solution addresses this challenge by processing massive data sets on compact computing platforms, using smaller, more efficient machine learning (ML) models to deliver real-time AI insights.
Now, your knee-jerk reaction might be that there are lots of edge AI-capable platforms around these days, and that's certainly true. What makes this platform different is that it's part of an all-inclusive managed service platform that includes all the systems, tools, and capabilities required for AI at the edge. It addresses data discovery, collection, integration, computation power, seamless connectivity, and AI model management.
There's a difference between training and using AI models in data centers and training and using them in the field. Take a pump, for example. You could have two fundamentally identical pumps deployed in different parts of the same factory–one mounted on a concrete slab, the other on a floating floor; one connected to plastic pipes, the other to metal pipes; one pumping water, the other working with some more viscous fluid–both exhibiting widely different characteristics.
As another example, consider two identical boilers located at opposite ends of the same factory and mounted in generally identical environments... except for the fact that one is located close to some railway tracks with trains randomly passing by causing vibration patterns that have nothing to do with the functioning of the boiler.
To address this, the folks at NTT DATA have created a suite of pre-defined models whose final training is performed in the field where they can learn the unique characteristics associated with the equipment under their charge.
Each Ultralight Edge AI platform can handle multiple sensors using multiple protocols (Ethernet, USB, Wi-Fi, cellular), and each deployment may involve multiple Ultralight Edge AI platforms. These platforms gather the data, contextualize it, time-stamp it, transform it as required (e.g., converting audio signals from the time domain into the frequency domain), process the heck out of it, analyze it, and action it on the operational technology (OT) side of the firewall. The resulting metadata can also be handed over to the informational technology (IT) side of the firewall for more detailed and higher-level analysis and actioning as required.
One of the things that's built into this solution is the ability for Ultralight Edge AI platforms to identify each type of sensor or PLC controller to which they are connected (this solution supports 250+ different protocols and tens of thousands of asset types), and to check that everything is up to date with the latest and greatest software versions and security patches.