Towards multi-layer interoperable IoT architecture with seamless integration of structured and unstructured data

Abstract

With the advancement of IoT devices and thanks to the unprecedented visibility and transparency they provide, diverse IoT-based applications are being developed. With the proliferation of IoT, both the amount and type of data items captured have increased dramatically. The data generated by IoT devices reside in different platforms and systems, and a major barrier to utilizing the data is the lack of interoperability among the different standards and technology used to capture and share the data. This dissertation proposes multi-layer interoperability architecture which deals with both structured and unstructured data by dividing it into three parts: middleware (Interoperability among entity based data models), mediation (Interoperability between Event and Entity based data models) and unstructured data(an improved semi-supervised prediction).

From the middleware perspective this dissertation address connectivity and fragmentation among the different IoT data models which uses entity data models. By proposing distributed Resource-oriented IoT Middleware for EPCGlobal architecture,in addition to scalability, this thesis address the requirement of interoperability in terms of communication interface and data exchange. It introduces Thing Model Using Global Product Code to abstract the resources and services provided by things and applications. The proposed system uses discovery service to manage distributed middleware which are enabled to register device information with its corresponding broker address for service resolution.

From the mediation perspective this dissertation introduces interoperability between entity based IoT models and Event based IoT models. The two data models differ not only in the data encoding but also in the underlying philosophy of representing IoT data. This dissertation proposed a methodology and implementation to closes the gap between the two data models. To demonstrates the applicability and feasibility of the proposed system, it is applied in to a real-life case study of integrating transparency systems used in a meat supply chain.

Data coming from IoT device is not only structured but also unstructured which makes it difficult to process it together. From this perspective, in this dissertation we propose an architecture similar to lambda architecture used in big data processing to deal with the unstructured data. To demonstrate the applicability we proposed an Improved SSGAN, a multi-Generator/Discriminator semi-supervised GAN architecture to address the well-known problem of mode collapse in addition to an improved classification for ordinal information.