2. Running AI/ML workloads in industrial facilities
Posted: Mon Feb 10, 2025 5:47 am
“Many industrial facilities have multiple control systems that may or may not integrate,” Nelson says. “IoT/edge can take data from different systems, correlate events, and predict failures.”
Latency—meaning reducing or eliminating it—is a key consideration in edge computing strategies. This is especially true for AI and machine learning applications, as well as other forms of automation that require data—and lots of it—to be effective.
IIoT has huge potential for AI/ML and automation, but also huge challenges with data and latency.
“Keeping smart machines running at the edge requires a lot of data,” says Brian Sathianathan, CTO of Iterate.ai. “Good AI requires data. Great AI requires a lot of data, and it needs it now.”
This can be problematic in the context of the first flow bulgaria mobile database above by Huff: sensor data coming from the edge to the core.
“I’ve seen situations in manufacturing plants where there’s ‘too much’ data flowing from the robot to the shop floor, through the local network, and then to the cloud and back again,” says Sathianathan. “That’s not good because, as manufacturing CIOs know, decisions need to be made instantly to be effective.”
If latency is a problem, then actual downtime is a killer, especially in industrial settings (where data outages or network problems can, for example, shut down a gas pipeline) and related segments such as manufacturing.
While some downtime is usually acceptable in standard IT environments, this is not the case in manufacturing. The cost of stopping production lines due to edge application failures can be hundreds of thousands of dollars per minute – there is simply no room for error.
Latency—meaning reducing or eliminating it—is a key consideration in edge computing strategies. This is especially true for AI and machine learning applications, as well as other forms of automation that require data—and lots of it—to be effective.
IIoT has huge potential for AI/ML and automation, but also huge challenges with data and latency.
“Keeping smart machines running at the edge requires a lot of data,” says Brian Sathianathan, CTO of Iterate.ai. “Good AI requires data. Great AI requires a lot of data, and it needs it now.”
This can be problematic in the context of the first flow bulgaria mobile database above by Huff: sensor data coming from the edge to the core.
“I’ve seen situations in manufacturing plants where there’s ‘too much’ data flowing from the robot to the shop floor, through the local network, and then to the cloud and back again,” says Sathianathan. “That’s not good because, as manufacturing CIOs know, decisions need to be made instantly to be effective.”
If latency is a problem, then actual downtime is a killer, especially in industrial settings (where data outages or network problems can, for example, shut down a gas pipeline) and related segments such as manufacturing.
While some downtime is usually acceptable in standard IT environments, this is not the case in manufacturing. The cost of stopping production lines due to edge application failures can be hundreds of thousands of dollars per minute – there is simply no room for error.