Evolutionary Approach to Efficient Provisioning and Self-organization in Wireless Sensor Networks (WSN)

Graduation date: 2016Access restricted to the OSU Community, at author's request, from March 16, 2016 - March 16, 2018Advances in low-power digital integration and microelectro-mechanical systems\ud (MEMS) have paved the way for micro-sensors. These sensors are equipped\ud with data processing capabilities along with sensory circuits. Sensor data are\ud processed on these individual sensors and transmitted to the target (sink). Lowcost\ud integration and small sizes of these sensors have generated special interest\ud in the area of disposable-sensors and large scale platform management. Queries\ud to these sensors are addressed to nodes which have data satisfying the same\ud condition. However, these sensors may be constrained in energy, bandwidth,\ud storage, and processing capabilities. Large number of such sensors along with\ud these constraints creates a sensor-management problem. At the network layer it\ud amounts to setting up the efficient route that transmits the non-redundant data\ud from source to the sink in order to maximize one or more sensor objectives (e.g.\ud battery (and sensor's) life, Sensor-Data yield). This is done while adapting to\ud changing connectivity due to failure of some nodes and new nodes powering up.\ud First part of the thesis propose a reduced-complexity genetic algorithm (GA)\ud for optimization of multi-hop battery-constrained sensor networks. The goal of\ud the system is to generate optimal number of sensor-clusters with cluster-heads.\ud It results in minimization of the power consumption of the sensor system while\ud maximizing the sensor objectives (coverage and exposure). The genetic algorithm\ud is used to adaptively create various components such as cluster-members,\ud cluster-heads, and next-cluster. These components are then used to evaluate\ud the average fitness of the system based on the sequence of communication links\ud towards the sink. We then enhance the genetic algorithm (GA) approach for\ud secure deployment of resource constrained multi-hop sensor networks. The goal\ud in this case is to achieve secure coverage and improve battery life by dynamically\ud optimizing security attributes (Like authentication and encryption).\ud Further, we augment the GA approach for intrusion detection of resource\ud constrained multi-hop sensor networks. Traditional intrusion detection mechanisms\ud have limited applicability to the sensor networks due to scarce battery\ud and processing resources. Therefore, we propose an effective scheme that would\ud offer a power efficient and lightweight approach to identify malicious attacks. We\ud evaluate sensor node attributes by measuring the perceived threat and its suitability\ud to host local monitoring node (LMN) that acts as trusted proxy agent for\ud the sink and capable of securely monitoring its neighbors. Security attributes in\ud conjunction with genetic algorithm jointly optimizes the selection of monitoring\ud nodes (i.e., LMN) by dynamically evaluating node fitness by profiling workloads\ud patterns, packet statistics, utilization data, battery status, and quality-of-service\ud compliance.\ud Second part of the thesis delves into application of Information Technology\ud (and Industrial) Systems and devices where the use of sensor networks\ud can deliver non-intrusive and effective telemetry for group-based server management.\ud These systems (Like Data Centers or Shipment tracking) face major\ud challenges in seamless integration of telemetry and control data that is essential\ud to various autonomic management functions related to power, thermal, reliability,\ud predictability, survivability, locality and adaptability. Such systems that\ud are supported by a dense network of sense-points operating in noisy environment\ud (Metals, Cables) are required to deliver reliable trends, measurements and\ud analysis in a timely fashion. The traditional approaches to provide distributed\ud observability and control using wired solutions are static, expensive, and nonscalable.\ud We apply the proposed GA approach for this unique environment that\ud replaces static wired sensors with dynamically reconfigurable battery-powered\ud wireless sensors. The proposed technique employs machine learning approach to\ud optimize sensor node function assignment, clustering decisions, route establishment\ud and data collection trees for improved throughput that results in effective\ud controls