IIC Journal of Innovation 5th Edition | Page 54

A Knowledge Graph Driven Approach for Edge Analytics
egw01
Dell Edge
Gateway
5100
egw03
HPE
Edgeline
EL10
Microsoft
Windows
egw04
instance_of
egw02
Dell Edge
Gateway
5100
Ubuntu
Copyright © 20 17 Ac c enture.
All rights reserved.
Figure 6: Small section of a client knowledge graph showing HW, OS, and physical gateway instance. Depicts four edge gateways
(egw01-egw04) and their HW and OS types.
inferencing from both the device and
application centric perspectives. The device
domain expert can run queries to determine
which devices can connect to one another as
well as which theoretical network topology
could host a given application.
protocol for communication, in determining
a standard for how data is streamed to
devices. This allows the application expert to
develop models without intimate knowledge
of the devices or sensors and only need to
know what data messaging protocol and
handshaking is needed.
The model types are the complementary
component to the devices in the knowledge
graph. When an application expert onboards
a model type, the knowledge graph
maintains the data schema as well as other
dependencies which are required for that
model to run. When a model is deployed to
an instance, a "model_instance"
relationship is automatically associated with
the device instance to denote that a model
is now running on a device (see Figure 6 for
a small example segment taken from the
O&G client graph as an illustrative example).
Our ontological approach to these
components decouples the different
domains and allow the experts to work
independently removing any interaction
roadblock that may hinder a deployment
while the platform takes on the complex task
of orchestration (which requires knowledge
from all fields of expertise). Given a
deployment template, the orchestration
agents will query the knowledge graph to
identify deployment targets based on the
device capabilities, device attributes, and
application dependencies.
The last component within the knowledge
graph is the data schema property
reference. The data schema reference allows
us to follow and extend upon the OPC UA
format, a standardized machine to machine
- 52 -
September 2017