Industrial IoT Edge Architecture for Machine and Deep Learning
Note that two sets of models are created by
training – one for the edge tier inference and
one for the Platform tier inference with real-
time streaming data.
O VERVIEW
This article presents an architecture for
Machine and Deep Learning at the edge and
Platform tiers for Industrial IoT. The
architecture extends the classical lambda
I NTRODUCTION
architecture in which both streaming and
The Industrial IoT architecture promoted by
batch processing are handled simultaneously.
the Industrial Internet Consortium (IIC) has
It proposes data aggregation and protocol
three tiers: Edge, Platform and Enterprise.
normalization at the edge with open source
This article focusses on the edge and Platform
software such as Linux Foundation’s EdgeX
tiers with stress on the computation and
Foundry™. This is followed by Machine or
storage at these tiers for Machine and Deep
Deep Learning inference at the edge. The data
Learning applications. The table below shows
from the edge is cached and distributed to the
Machine and Deep Learning use cases and
Platform tier online or offline via open source
challenges for two Industrial IoT market
software. Subsequently, at the Platform tier a
verticals – Smart Manufacturing and
data lake is created of edge and historical
Transportation:
data, which is used for Deep Learning training.
Use Cases
Drivers
Challenges
Smart Manufacturing
Predictive maintenance
Increase yield and asset
Low latencies
utilization
Process optimization
Data security
New revenue streams
Supply chain optimization
Interoperability between
Operational
efficiencies
diverse sets of equipment
Remote asset management
Increased worker
Rapid interpretation of large
Product lifecycle
satisfaction/safety
volumes of data
monitoring
Eco-sustainability
Indoor/outdoor reliability in
Integrated plant
harsh environments
Risk avoidance
management
Connectivity across access
Product-as-a-service
technologies
Remote asset tracking
Predictive asset
maintenance
Warehouse capacity
optimization
Real-time fleet
management
Route optimization
Transportation
Better fleet utilization/
opex reduction
Reduce emergency
response times
Monitor driver behavior
Maintenance cost
reduction
Constant connectivity
Local regulation compliance
for international fleets
Low latency
Table 1: Smart Manufacturing and Transportation Use cases for Machine and Deep Learning
(Adapted from SDx Central IoT Infrastructure Report 2017)
- 58 -
September 2017