Industrial IoT Edge Architecture for Machine and Deep Learning
to EdgeX, and orchestrates the data to Pulse
IoT Center to manage.
The Platform tier receives real-time edge data
as well as stored data from the edge cache.
The real-time edge data is processed and
aggregated across edges to create features
which are stored in the platform cache. The
real-time features are used by the platform
inference models. Inference results from the
edge and platform are aggregated and sent to
the Enterprise tier for BI reporting and
visualization as well as remediation if
necessary.
The Platform tier uses the cached raw and
processed data to train edge and Platform tier
models, which are served in real-time to the
edge and platform inference engines via Pulse
IoT Center OTA (over the air) software update
capability. Figure 9 summarizes the
architecture.
C ONCRETE I LLUSTRATION OF APPLYING
THIS ARCHITECTURE
Consider a set of surveillance cameras
attached to the edge device in the truck. The
edge device processes the footage from all
cameras and detects events of interest
(accident, intrusion, fist fight, pedestrian near
misses, a comet falling from the sky, etc.).
Thanks to edge computing, the voluminous
video feeds are processed locally, thus at low
latency (important for quick alerting; e.g., for
crash avoidance). Only footage of such
significant events is sent OTA to a cloud
service for archival and future search,
retrieval and learning/training. Therefore,
such important footage is protected in the
event of vandalism or crash of local
equipment. In addition, this approach
significantly lowers OTA traffic; we estimate
that
important
footage
constitutes
Figure 16: Trucking Use Case Architecture
- 68 -
September 2017