Improving the dependability of safety critical wireless sensor network scheduling using artificial intelligence

To ensure optimal functionality and adherence to specified requirements, a Wireless Sensor Network (WSN) must prioritise the validation of its three fundamental attributes: Connectivity, Coverage, and Network Lifetime. Existing literature highlights numerous research endeavours to resolve reliability issues in WSNs, often concentrating on singular properties such as coverage and/or connectivity. These properties are frequently treated interchangeably and seldom concurrently addressed due to the intricate challenges arising from the distributed nature and resource constraints typical of WSNs. Moreover, safety critical WSNs applications require to follow rigid and strict policies such as strict deadlines while deploying in emergency situations which adds an additional layer of complexity. This layer of complexity is caused by new set of additional requirements not only connectivity, coverage, and lifetime. That is technically challenging in terms of optimizing the basic parameters required for the implementation of WSNs. Notably, there is a scarcity of published work comprehensively analysing and testing all three primary requirements; connectivity, coverage, and network lifetime of WSNs simultaneously, owing to their inherent complexity. This thesis tackled the three mentioned properties of safety critical WSNs as a Multi-Objective Optimisation (MOO) problem. The research methodology encompasses six key principles. Firstly, the Randomised Coverage-based Scheduling (RCS) algorithm is replicated, validated, and verified using a MATLAB simulation environment, revealing insights into node utilisation imbalances. Secondly, the performance of the RCS algorithm is scrutinised using the Perceptron Multilayer Artificial Neural Network (ANN) scheduling algorithm, exposing limitations. Thirdly, the Hidden Markov decision-process Model (HMM) is employed to enhance the service availability and reliability of the RCS algorithm, demonstrating superior optimisation metrics over RCS by increasing network lifetime while improving coverage and connectivity. Fourthly, a novel Bio-Inspired Bat algorithm was developed to address the identified limitations of previous scheduling algorithms, utilising objective optimisation functions and Pareto optimisation. The Bat algorithm outperformed the HMM algorithm across all metrics. Fifthly, the Self-Organising Feature Map (SOFM) algorithm surpasses the Bat algorithm with its straightforward approach to dimension reduction and classification. Sixthly, critical analyses of implemented algorithms (HMM, Bat, and SOFM) reveal similar patterns in coverage data. Consequently, a Long Short-Term Memory (LSTM)-based node scheduling algorithm was introduced to analyse and provide an energy-efficient scheduling solution, addressing the Multi-Objective Optimization (MOO) highlighted in this research.