Synthesis of a Conceptual Architecture for Cognitive IoT Device Management

Abstract

The Internet of Things (IoT) is characterised by diverse, dynamic and distributed deployments of sensors, actuators, tools and gadgets. These devices often have limited resources like energy, memory, and processing power. Moreover, they are frequently located in unstable environments where conditions change over time. The limitations of IoT nodes often require them to delegate resource-intensive tasks to a device management platform that can act on their behalf and ensure optimal performance. However, in addition to the limited resources of the devices and the context in which they operate, these platforms also have to deal with challenges related to the scale of the deployment, the network topology, and the type of management that is performed. The size and complexity of IoT deployments often require device management platforms to operate autonomously. In this thesis, we study and explore how to autonomously manage a multitude of distinct and constrained IoT equipment that operate in ever-changing contexts. To solve this problem, we synthesised a generalised conceptual architecture for autonomous management of IoT devices deployed in non-stationary environments. The conceptual architecture is based on the theoretical foundation of two distinct research fields: IoT device management and cognitive architectures. The research was guided by the design science research methodology and followed both a bottom-up and top-down approach. Initially, we conducted six case studies, where we focused on designing specific parts of the proposed model. Following that, we carried out a structured literature review, where we analysed architectural models from 32 case studies in the field of autonomous IoT device management. Based on the insights that we gathered through the work of this thesis, we synthesised a conceptual architecture for cognitive IoT device management that fulfilled our research goal. The model describes adaptive behaviour on three levels. On the highest level, there are three system components: The device, the device manager and the system manager. On the second level, we have five distinct adaptation components that are contained within the system components. They are responsible for handling perception, action, adaptation process, declarative knowledge and procedural knowledge, respectively. The adaptation process component is further detailed in five types of adaption processes, namely, monitor, analyse, learn, predict and plan. On the lowest levels, we find adaptation mechanisms and triggers, which describe the data flow between the components. Apart from the conceptual architecture for cognitive IoT device management, this thesis has three additional contributions. First, we present a comprehensive taxonomy of adaptation mechanisms for cognitive IoT device management. Second, we describe a model of cognitive planning. Third, we provide a list of best practices to guide the design and implementation of cognitive IoT device management platforms, along with recommendations for when and how to apply them. These contributions will be useful for those who aim to develop such solutions.