A Knowledge Graph Driven Approach for Edge Analytics
software solutions unencumbered by legacy
issues, many companies have already
invested in IoT infrastructure and software
development resources. The result of which
is a heterogeneous, piecemeal, or disparate
deployments
coupled
with
legacy
applications where replacement of existing
infrastructure and system refactoring is not
economically nor technically feasible. To
overcome the associated challenges of
legacy and heterogeneous ecosystems, we
utilize containerization techniques to
integrate existing applications and enable
the deployment of applications to any node
within
an
existing
heterogeneous
deployment. The core of our framework
tackles the challenges with a knowledge
graph in the cloud, an orchestration layer
that enables communications and container
instances, and a distributed messaging
backbone for resiliency and filtering.
I NTRODUCTION
The rapidly expanding Internet of Things
(IoT) market introduces new challenges
surrounding the processing and storage of
large quantities of data produced by edge
analytics systems and the management of
applications deployed on varied array of
devices that comprise these systems. To
address these challenges, we present a
novel, operationalized approach to the
deployment of a practical, edge analytics
framework that accomplishes two key goals:
(1) Enable the simplified integration of
heterogeneous hardware and software
resources with existing applications (or
models) and (2) Utilizes a knowledge graph
to capture domain knowledge for
reusability,
relationship
inferencing,
maintainability,
and
data
stream
communication
normalization
for
infrastructure instantiation. Inspired by the
Web of Things (WoT), the approach has a
“network layer” and “accessibility layer”;
however, expands upon it conceptually to
provide an abstraction for expertise
modularity. Through addressing these goals
the framework streamlines a division of
roles, operations, and deployments while
minimizing associated maintenance of IoT
ecosystems.
This
is
enabled
by
bootstrapping domain knowledge via a
graph and instantiating device-specific
information through an onboarding
interface.
In this discussion, we present the
components and benefits of our system
through two in-flight use cases (a large oil
and gas client, and a large construction and
mining client) along with an explanation of
the problem space, framework walkthrough,
and brief comparison to other approaches
throughout.
M OTIVATION
By 2020, the Internet of Things (IoT) market
will reach $267 billion with 50% of spending
driven by industrial and commercial
While companies without existing IoT
deployments
can
deploy
vertically-
integrated edge analytics hardware and
IIC Journal of Innovation
- 43 -