IIC Journal of Innovation 5th Edition | Page 53

A Knowledge Graph Driven Approach for Edge Analytics
onboarded will inherit a triggered
deployment of unique instances of the same
required containers.
as inherit properties from other device
types.
The strength of the knowledge graph is in
the capturing of relationships such as an
"instance_of” link between a device type
and a device instance when a field engineer
S OLUTION A RCHITECTURE
Knowledge Graph
Knowledge Graph
Device Metadata
Device
Type
Data type
Data
Type
Model Metadata
Data type
OS type
GW
Instance
Connected to
Capability
Instance
{ Resource
Requirements }
OS
Data Type:
{ Data
Schema }
Model
Type
HW_Req
Device
Type
Device
Type
{ Data Schema }
Inherits
{ Model
Parameters }
Model
Instance
Hardware
Running on
Copyright © 2017 Accenture All rights reserved.
Figure 5: Knowledge Graph
onboards a sensor. All capabilities of the
instance are documented via the device type
through the knowledge graph. However,
instance-specific information, such as what
sensors are connected, which applications
are running, location, current resource
usage, etc. are associated to that instance
alone. The instance and device type
relationships together enable deployment
and management teams to query for subsets
of devices based on capabilities and
properties when determining which
applications to deploy where. Furthermore,
these relationships allow for more complex
The knowledge graph (simplification shown
in Figure 5) portion is implemented as a
graph in the cloud. It stores all relationships
for devices, sensors, models, and data types.
When onboarding a new device, the device
engineer creates a new device type which in
turn is associated to data schemas
(documenting the semantics and format of
the data produced by a device), associated
hardware (e.g., GPU or other unique
property specifying which types of
applications can run on the device), as well
IIC Journal of Innovation
- 51 -