A Predictive Fault-tolerance Framework for IoT Systems

1 Context
When the term Internet of Things (IoT) is used, its definition spans a wide range of applications which extend the Internet to interconnected objects forming, where a diverse array of sensor devices are connected with the intention of mining the data which they generate. The purpose of IoT is to connect all of these devices to the Internet to allow systems to identify, locate, track, monitor in real time. IoT environments often suffer from failures such as device energy depletion and poor connectivity, which may result in data loss. If an IoT system is to remain resilient to ongoing challenges in its environment, such as changes in sensor location, failures in service provision, and device faults, we need a way to anticipate and to mitigate the problems before they occur. This PhD proposal seeks to investigate how real-time complex event processing and the long-term predictability of machine learning can be combined with fault tolerance mechanisms to provide and maintain an acceptable level of service in the face of faults.

2 Aims and Objectives
The aim of the proposed research is to develop a framework where features can be interchanged that exploit real-time and historic data to improve the resilience of an IoT environment. As such, the objectives for this research are as follows:
Objective1. To identify and classify faults events and fault patterns in IoT systems
In order to become resilience to faults, one must understand the faults that can occur within the system. As such, the first objective is to have a thorough review of the different faults which can occur within IoT systems and classify such faults so that patterns can be identified on what faults occur, their severity, the effect the fault has on the system, and also what is likely to cause such a fault.
Objective2. To develop a service-oriented fault anticipation framework
A service-oriented framework will be developed so that these known faults can be identified and anticipated as they occur within an IoT environment. It will utilise complex event processing so that faults can be identified in real-time, and it will also use machine learning to predict when faults are likely to occur, where the algorithm(s) will be trained using the known faults. The framework will generate predictive models, and live data will be used to infer whether the threat of failure is probable, so that the system can react to the potential fault in good time.
Objective3. To incorporate fault mitigation strategies and mechanisms
An extensive review of strategies and mechanisms to resolve system failures will be undertaken. The outcome of the review will inform the development and integration of strategies and mechanisms for mitigating the faults into the framework.
Objective4. To scale to more complex IoT scenarios
The prior objectives would be initially tested within a controlled, small-scale environment. As large-scale IoT systems (or perhaps ones which grow to a large scale) are to be expected in real-world environments, it is important to assess whether the framework is able to cope when scaled up. As such, the framework will be extended to adapt to a cloud-based architecture, with a close evaluation of existing cloud-based IoT solutions.

3 Applications
Some of IoT's recent applications include in agriculture where sensor devices are used to monitor farmland and supply water when a farm is dry and remove any water when in excess, as well as using triangulation to help the visually impaired navigate around the home.