Industrial IoT Edge Architecture for Machine and Deep Learning
Figure 10: Typical components of the edge
Data from the devices are aggregated and
normalized at the gateway. The gateway is
also used for Machine or Deep Learning
inference. Raw data from the edge are
temporarily stored at the edge cache which is
offloaded to the Platform tier cache for
training. Transfer of data from the edge to the
platform can happen in batches when
network bandwidth is available. Alternatively,
edge data can be transferred to the platform
without edge caching on available transport.
The architecture above can be implemented
with several frameworks. We discuss
implementing it with two open source
IIC Journal of Innovation
- 63 -
frameworks: (1) EdgeX Foundry, and (2)
Liota™.
EdgeX Foundry
EdgeX Foundry is an open source project
hosted by The Linux Foundation offering a
common framework for IoT edge computing.
See details of EdgeX™ Foundry at
www.edgexfoundry.org. EdgeX offers many
services including security, identification,
scheduling, logging, deployment and console
management. Due to the high volume of data
transfer between devices and edge as well as
from edge to platform, the services we most
use for Deep Learning architecture are: