Fault Diagnosis in Wireless Sensor Networks using Artificial Immune System

Abstract

Wireless sensor network (WSN) consists of a huge number of sensor nodes, which are tiny in size, low battery powered and low cost. These sensor nodes are deployed in the environment for applications such as healthcare monitoring, environmental monitoring, and military application. Due to their physical limitations and environmental conditions, these nodes are subjected to different types of faults such as hard and soft fault. On the occurrence of hard fault a sensor node can not communicate with other sensor nodes. In case of soft faults, they are categorized into soft permanent, soft intermittent, and soft transient fault. On the occurrence of soft fault the sensor nodes can communicate with other sensor nodes but erroneously. A soft permanent faulty sensor node persists the faulty behavior by giving erroneous results. A soft intermittent faulty sensor node may behave like faulty for some period of time and fault free at other times. A soft transient faulty sensor node may occur for some duration of time and fault free for the rest of the time. In order to obtain accurate data from the deployed WSN, the sensor nodes need to be fault free. Motivated by the need of a fault free WSN in the deployed field, it is crucial to identify the faulty sensor nodes in WSN and categorize them into their respective fault types. In this thesis, four fault diagnosis algorithms have been proposed based on the concept of artificial immune system (AIS) such as clonal selection principle (CSP), negative selection algorithm (NSA), dendritic cell algorithm (DCA) and artificial immune network (AIN). To evaluate the performance of the proposed methods simulation and experimental validation are conducted using standard generic performance parameters such as fault detection accuracy (FDA), false alarm rate (FAR), false positive rate (FPR), fault classification accuracy (FCA), false classification rate (FCR), fault detection latency (FDL) and energy consumption (EC). Throughout the thesis, it is assumed that the sensor nodes can be either hard or soft faulty whereas the communication links are assumed to be fault free. To resemble the scenario in which the sensor nodes are randomly deployed in the field, the WSN follows an arbitrary network topology.
An efficient fault detection algorithm based on clonal selection principle(FDCSP) of AIS has been proposed to detect faulty sensor nodes such as hard permanent, soft permanent, soft intermittent and soft transient. This is followed by a fault classification algorithm based on the above respective fault types using the probabilistic neural network. After the actual fault status is detected, the faulty sensor nodes are isolated in the isolation phase. In fact, the proposed algorithm follows three phases such as fault detection, fault classification and fault isolation. The performance of the algorithm is evaluated by using the performance metrics such as FDA, FAR, FPR. It is shown that, the fault detection accuracy of the proposed FDCSP algorithm is improved by 1.8%, 5.06% and 6.45% over Mohapatra et al., Panda et al. and Elhadef et al., respectively. The false alarm rate of the proposed algorithm is improved by 0.43%, 0.78% and 1.88% over Mohapatra et al., Panda et al. and Elhadef et al., respectively. The false positive rate of the proposed algorithm is improved by 1.81%, 5.07% and 6.46% over Mohapatra et al., Panda et al. and Elhadef et al., respectively. The fault classification performance is measured by fault classification accuracy and false classification rate. The simulation result also shows that the FDCSP algorithm provides less fault detection latency i.e., 4.11%, 6.19% and 10.35% over Mohapatra et al., Panda et al. and Elhadef et al., respectively and consumes less energy i.e., 11.40%, 20.95% and 49.03% over Mohapatra et al., Panda et al. and Elhadef et al., respectively
An improved negative selection algorithm (INSA) has been proposed to diagnose the faulty sensor nodes and classified into soft permanent, soft intermittent and soft transient using the support vector machine. The performance of the algorithm is evaluated by using the performance metrics where it is shown that, the fault detection accuracy of the proposed INSA algorithm is improved by 1.55%, 4.97% and 6.49% over Mohapatra et al., Panda et al. and Elhadef et al., respectively. The false alarm rate of the proposed algorithm is improved by 0.33%, 0.85% and 1.86% over Mohapatra et al., Panda et al. and Elhadef et al., respectively. The false positive rate of the proposed algorithm is improved by 1.55%, 4.97% and 6.49% over Mohapatra et al., Panda et al. and Elhadef et al., respectively. The fault classification phase gives the average classification accuracy approximately 97% and the average misclassification rate 0.03. The simulation result also shows that the proposed algorithm provides less fault detection latency i.e., 4.26%, 8.75% and 13.09% over Mohapatra et al., Panda et al. and Elhadef et al., respectively and consumes less energy i.e., 11.64%, 21.81% and 50% over Mohapatra et al., Panda et al. and Elhadef et al., respectively.
Fault diagnosis using dendritic cell algorithm has been proposed to detect the faulty nodes. The important feature of this algorithm is that no training data is required. The performance of the algorithm is evaluated by using the performance metrics where it is shown that, the FDDCA algorithm gives better result as compared to the existing algorithms in terms of FDA, FAR, FPR, FDL and EC. The fault detection accuracy of the proposed FDDCA algorithm is improved by 1.51%, 4.7% and 6.46% over Mohapatra et al., Panda et al. and Elhadef et al., respectively. The false alarm rate of the proposed algorithm is improved by 0.25%, 0.9% and 1.85% over Mohapatra et al., Panda et al. and Elhadef et al., respectively. The false positive rate of the proposed algorithm is improved by 1.51%, 4.7% and 6.46% over Mohapatra et al., Panda et al. and Elhadef et al., respectively. The proposed algorithm provides less fault detection latency i.e., 7.48%, 12.54% and 16.56% over Mohapatra et al., Panda et al. and Elhadef et al., respectively and consumes less energy i.e., 18.49%, 26.99% and 53.51% over Mohapatra et al., Panda et al. and Elhadef et al., respectively
An artificial immune network based fault diagnosis algorithm has been proposed to diagnose the faulty sensor nodes. In this algorithm to train and optimize the fault samples, learning, memory and suppression mechanism of immune network is used. Fault type information has been added to memory antibodies so that it can learn and memorize the same types of faults. Hence, classification accuracy can be improved. Experimental result shows that the fault detection accuracy of the proposed AINFDA algorithm is improved by 1.51%, 5.01% and 6.46% over Mohapatra et al., Panda et al. and Elhadef et al., respectively. The false alarm rate of the proposed algorithm is improved by 0.39%, 0.75% and 1.74% over Mohapatra et al., Panda et al. and Elhadef et al., respectively. The false positive rate of the proposed algorithm is improved by 1.51%, 5.01% and 6.46% over Mohapatra et al., Panda et al. and Elhadef et al., respectively. The proposed algorithm provides less fault detection latency i.e., 4.44%, 6.52% and 10.94% over Mohapatra et al., Panda et al. and Elhadef et al., respectively and consumes less energy i.e., 12.16%, 22.15% and 49.80% over Mohapatra et al., Panda et al. and Elhadef et al., respectively.