A Knowledge Graph Driven Approach for Edge Analytics
the downstream control message schema to modify the operation of the sensor or actuator, relevant protocol information, and any run-time information and parameters( e. g., container references, etc.).
To enable the edge devices and analytics applications to communicate with one another, we deploy a messaging backbone within the edge appliance gateway. Since the messaging protocol employed by the sensor may not be the same as chosen for our messaging backbone, the device expert would need to create a translation container specific to a device type that would consume messages from the edge device ' s protocol and push them to the messaging backbone ' s protocol( and vice versa). While a container only handles a single edge device instance, a containerized approach allows for additional instances of the same container to be deployed whenever a new sensor is onboarded.
Application Expert
The application expert can query all necessary information for development from the knowledge graph. The analytics container developed by the application expert pulls the data and control message schema from the knowledge graph allowing the application to consume data from the edge device. This information, along with any known hardware dependencies, is documented in the knowledge graph when metadata for model sensor types is onboarded into the RDF ontology. The application container communicates with the edge device( or another application container) through the centralized messaging backbone to seamlessly produce and consume data.
Our framework ' s containerized approach allows the application expert to reuse or to develop an application in whatever environment they choose so long as containerization is supported for that operating system( currently supported by all major OS’ s). For clients that seek to reuse applications from their existing application portfolio, the only required addition is communication with the messaging backbone. Additionally, because the production and consumption of data is abstracted through the message bus, this application container may be deployed on any edge appliance that has access or subscribes to the edge device ' s message topics. Messages are labeled as belonging to a topic category for subscription models and filtering of data messages, allowing for simplified vertical mobility of analytics containers as hardware resource constraints or uplink backhaul quality requires.
Field Engineer
The field engineer needs only to concentrate on the physical deployment of the sensor itself. Once the sensor is installed, it will begin to generate identifying heartbeat messages, which in turn will be consumed by an auto-discovery agent resulting in the registration of the new unique sensor instance into the knowledge graph. The registration of a new instance of an edge device will trigger the deployment of the container instances required to communicate with the edge device( e. g., message protocol translator) by our orchestrator. Every new sensor that is
- 50- September 2017