Model-Driven Edge Analytics: Practical Use Cases in Smart Manufacturing.

AbstractIn the Internet of Things (IoT) era, devices and systems generate enormous amounts of real-time data, and demand real-time analytics in an uninterrupted manner. The typical solution, a cloud-centred architecture providing an analytics service, cannot guarantee real-time responsiveness because of unpredictable workloads and network congestion. Recently, edge computing has been proposed as a solution to reduce latency in critical systems. For computation processing and analytics on edge, the challenges include handling the heterogeneity of devices and data, and achieving processing on the edge in order to reduce the amount of data transmitted over the network.In this paper, we show how low-code, model-driven approaches benefit a Digital Platform for Edge analytics. The first solution uses EdgeX, an IIoT framework for supporting heterogeneous architectures with the eKuiper rule-based engine. The engine schedules fully automatically tasks that retrieve data from the Edge, as the infrastructure near the data is generated, allowing us to create a continuous flow of information. The second solution uses FiWARE, an IIoT framework used in industry, using IoT agents to accomplish a pipeline for edge analytics. In our architecture, based on the DIME LC/NC Integrated Modelling Environment, both integrations of EdgeX/eKuyper and FiWARE happen by adding an External Native DSL to this Digital Platform. The DSL comprises a family of reusable Service-Independent Building blocks (SIBs), which are the essential modelling entities and (service) execution capabilities in the architecture’s modelling layer. They provide users with capabilities to connect, control and organise devices and components, and develop custom workflows in a simple drag and drop manner.