Research On Reliability Modelling And Analysis Of Wireless Sensor Network

Due to its low cost, flexible deployment and easy expansion, the Wireless Sensor Network (WSN) has widely been applied in Intelligent Agriculture, Intelligent Transportation, Disaster Warning, Environmental Surveillance, Military and Health Care, etc. The prospective of WSN’s application will be even more extensive as the software and hardware technology is developing. Nevertheless, the reliability requirements are strictly demanded for the WSNs in these applications such as Medicine, Military and Disaster Warning where the reliability has been one of the key performance measurements. The reliability design and analysis of WSNs has therefore become one of the main focuses in the current research, and how to establish the corresponding methodology on the reliability assessment is the primary problem in WSN reliability analysis.On the WSNs reliability assessment and analysis, many research outcomes have been achieved. However, there are still some problems that are necessary to study further:1) Although a lot of research outcomes on the WSNs reliability with specified topology structure can be found, there is lack of efficient approach to study the WSNs with unknown topology structure or the topology structure cannot be identified due to its dynamic change.2) There are many research outcomes on the WSNs reliability models under specific constraint conditions, little research can be found on general but flexible and efficient models.3) Comparatively, the current study methods on WSNs reliability are mostly static and mainly based on simulation and probabilistic analysis. Little study can be found on the dynamic assessment and prediction of the WSN reliability based on the actual failure data observed in the WSN operation process.4) Currently, most of system reliability models assume that the failures of components or subsystems are independent. For WSNs reliability models, the assumption of independent failures among the nodes or subnets does not fully comply with the engineering background. For example, the failures of nodes are usually dependent since the failures of some nodes will increase the loading of other nodes so that the failure process of these nodes will be accelerated.Considering the above-mentioned problems in WSN reliability assessment and analysis and based on practical engineering background, this thesis has systematically studied the theoretical modelling, assessment and prediction methods on WSN reliability under different maintenance policies, data types, network structure and failure characteristics. By applying probability and statistics theory and methods such as Ordered Binary Decision Diagram (OBDD), Order Statistics, Non-Homogeneous Poisson Process (NHPP) and Copula Function, general models based on various maintenance policies and their corresponding statistical inference methods are developed for reliability assessment and prediction of WSN. The system reliability functions are also given for the proposed models.The main innovations of this thesis can be summarized as follows:1) For the given and known topology structure in WSNs, the reliability of two-terminal network with both imperfect nodes and imperfect edges has been studied. Based on the OBDD analysis approach, a method named as Enhanced Node Expansion (ENE) is developed, which provides an efficient solution for the reliability assessment of a large scale WSN with various complex network structures. The ENE method has integrated three effective measures to implement:reduce the redundant states by applying high efficient OBDD structure to save the node and edge states; avoid the repeated computations of isomorphic subnets by using the hash table to identify the processed subnets; use three-level variable-length vector label to mark subnet structure so that the storage space in hash tables are greatly reduced, and the efficiency of isomorphic subnet identification is therefore significantly increased. These three measures have made the ENE highly valuable in practice since it can greatly reduce the computation amount and increase operational efficiency. Additionally, based on the study of WSN reliability assessment, the Birnbaum measure is introduced to quantify the importance of network links, and then a new OBDD-based method is proposed to evaluate the importance of network links. The proposed method can be used to analyze and evaluate the stability of the operating WSN. By applying the method, the weak links that possibly cause the failures of system can be identified and detected; and the WSN reliability is consequently enhanced.2) For the unknown or uncertain topology structure due to the dynamic changes of the inner network in WSNs, the system reliability models with and without maintenance are studied, respectively.For the WSN with possible maintenance, this thesis has applied the Non-Homogeneous Poisson Process (NHPP) in WSN reliability and developed WSN NHPP reliability models. NHPPs are the stochastic point processes that are widely applied in various fields for characterizing the system failure process with deterioration or enhanced patterns. Based on the operation environment and failure mechanism of WSN, this thesis has integrated the NHPP theory into WSN failure process and, for the first time, applied the NHPP reliability models to characterize the failure process of repairable WSN. The NHPP models can also be used for making the dynamic assessment and prediction of WSN reliability. By simulation study for the failure processes of WSN with different topology structures, the proposed NHPP models have been shown to be suitable for modelling the failure processes of those WSN.For the WSNs without maintenance, a general system reliability model is developed suitable for various lifetime distributions of the nodes and the analytic formulas of the WSN reliability have also been provided. Based on the proposed model, an optimization model on the number of the nodes in a WSN system is further proposed. This model is useful in consideration of resource saving and cost control in practice. Since exponential distributions are widely suitable for characterizing the failure pattern of electronic products, the proposed general model is therefore specified by considering the exponential distributions for subnet lifetimes. The analysis method, assessment and prediction models are specifically provided.3) An additive NHPP model has been developed for large-scale WSN with failure data on system, subnets and nodes. A large-scale WSN usually consists of thousand subnets (nodes or subsystems) and the failure data can be obtained at different levels. Under the assumption of independent failure processes of subnets, the additive NHPP model for WSN system reliability is established by using all failure data. The Maximum Likelihood method is therefore used for model parameter estimation so that the WSN reliability can be evaluated and predicted. The proposed models are demonstrated by simulation tests to be good in modelling.Furthermore, based on the proposed additive WSN reliability model, this thesis has studied the reliability modelling of WSN with the masked data. Masked data is a common engineering phenomenon where the causes of system failures cannot be exactly identified, possibly just known to be one sub-sets of subnets. In WSN, the masked failure phenomena occur, for example, due to the data damage in data integration or compression that are typically implemented in the redundant operation of sensor nodes. In the model, it is assumed that a WSN is made of several subnets or clusters, and the failure processes of these subnets are independent. The additive NHPP masked-data model is then applied to describe the failure process of the WSN when the masked data are appeared. The maximum likelihood estimation (MLE) procedure is developed to estimate the parameters in the proposed model.4) By introducing the Copula function to characterize the relationship between the failures of nodes, a reliability model with dependent failures is established. In the model, the joint distribution can be obtained by using Copula function given the lifetime distributions of subnets, the system reliability can therefore be calculated for the WSN system with dependent failures. The proposed model is more compliance with the engineering background and can achieve a better WSN reliability assessment, and therefore provide the theoretical base for WSN reliability design.