A Knowledge Graph Driven Approach for Edge Analytics
highly intermingled device expertise and
support functional modularity. The
challenge is to reduce the high-touch
environment of sensors so that field
operations can focus on their core business
competencies, free from distractions of IoT
deployment, maintenance and systems
upgrading.
Edge Framework
Our edge framework tackles the challenges
from the generalized use case and specific
use cases as follows: (1) To decouple the
different components of our solution, we
employ a knowledge graph (graph that
bridges concepts together) that serves as a
centralized abstraction layer between each
of the components that enable the domain-
knowledge expert to focus on their relevant
components while promoting knowledge
retention
for
future
infrastructure
modifications and deployments. (2) To
enable
seamless
scalability,
simple
maintainability and application reusability in
heterogeneous infrastructure deployments,
we use the knowledge graph to manage and
employ a containerized approach that allows
for development in whichever OS or
language the developer chooses (or to reuse
applications that were previously developed
for a differing OS in another language). This
is done by creating relationship links in a
schema to act as a reference to various
containers that can be deployed to an edge
gateway instance given a set of features,
capabilities, etc. We can then perform a
search and infer which equipment can
handle which containers after querying an
actual graph instance.
In-flight Impl ementation Requirements:
1) Decoupling
of
HW/SW
stack:
modularity of the hardware capabilities
from specific applications and their
versions
2) Simplified Deployment at Scale:
capability to deploy thousands of
sensors and provide a gated, pipelined
and streamlined onboarding approach
needing only minimal interaction of
experts
3) Data
message
and
protocol
management: capability to queue
messages, filter information and focus
only on relevant information
This client has a critical need to reduce high
maintenance touch from the device expert
and modularize the deployment approach to
avoid confusion when problems arise from
installation or maintenance. When a device
goes down, equipment is sidelined and
revenue may be lost from unwarranted
prolonged maintenance periods. There is
also a deluge of events that stream from
these devices that may or may not be
important. Proper filtering and handling of
said data is paramount to avoid operator
attrition when parsing data feeds.
By analyzing our client's deployment
scenarios, we developed the following core
framework to enable scalable and
maintainable edge analytics that seamlessly
integrate with heterogeneous hardware and
software infrastructures:
Knowledge Graph
A centralized ontology containing schema
information describing each edge device
- 48 -
September 2017