Ontology-based Consistent Specification and Scalable Execution of Sensor Data Acquisition Plans in Cross-Domain loT Platforms

Nowadays there is an increased number of vertical Internet of Things (IoT) applications that have been developed within IoT Platforms that often do not interact with each other because of the adoption of different standards and formats. Several efforts are devoted to the construction of software infrastructures that facilitate the interoperability among heterogeneous cross-domain IoT platforms for the realization of horizontal applications. Even if their realization poses different challenges across all layers of the network stack, in this thesis we focus on the interoperability issues that arise at the data management layer. Starting from a flexible multi-granular Spatio-Temporal-Thematic data model according to which events generated by different kinds of sensors can be represented, we propose a Semantic Virtualization approach according to which the sensors belonging to different IoT platforms and the schema of the produced event streams are described in a Domain Ontology, obtained through the extension of the well-known ontologies (SSN and IoT-Lite ontologies) to the needs of a specific domain. Then, these sensors can be exploited for the creation of Data Acquisition Plans (DAPs) by means of which the streams of events can be filtered, merged, and aggregated in a meaningful way. Notions of soundness and consistency are introduced to bind the output streams of the services contained in the DAP with the Domain Ontology for providing a semantic description of its final output. The facilities of the \streamLoader prototype are finally presented for supporting the domain experts in the Semantic Virtualization of the sensors and for the construction of meaningful DAPs. Different graphical facilities have been developed for supporting domain experts in the development of complex DAPs. The system provides also facilities for their syntax-based translations in the Apache Spark Streaming language and execution in real time in a distributed cluster of machines.