WSNs clustering based on semantic neighborhood relationships

Abstract

We propose a semantic clustering model based on a fuzzy inference system to find out the semantic neighborhood relationships in wireless sensor networks in order to both reduce energy consumption and improve the data accuracy. As a case study we describe a structural health monitoring application which was used to illustrate and assess the proposed model. We conduct experiments in order to evaluate the proposal in two different scenarios of damage with different data aggregation methods. We also compared our proposal, using the same data set, with a deterministic clustering method and with the LEACH algorithm. The results indicate that our approach is an energy-efficient clustering method for WSNs, outperforming both the deterministic clustering and LEACH algorithms in about 70% and 47% of energy savings respectively. The energy saving comes from the fact that we have a more efficient in-network data aggregation process since by exploiting the semantic relation between sensor nodes we can potentially aggregate more similar data and consequently, decrease the data redundancy (thus minimizing transmissions). Nodes that are semantically unrelated can operate in low-duty cycle, further reducing the energy consumption. Moreover, our proposal has the potential to improve the data accuracy provided for the application where accuracy is a QoS requirement in typical WSN applications.

Introduction

Wireless sensor networks (WSNs) are examples of resource-constrained networks in which processing resources, storage and energy are limited. In spite of the fact that cost and size considerations imply that the resources available to individual nodes are severely limited, recent advances on technology and research in this field lead us to believe that limited processor and memory are temporary constraints in WSNs that tend to disappear with fast developing fabrication techniques [1]. The energy constraint, on the other hand, remains as a critical issue that needs to be tackled so that WSNs can be widely employed. There is a wide range of applications that can benefit from the use of WSNs, with different features and QoS requirements. Typical QoS requirements in WSN environments are coverage, data accuracy and packet delay. On the one hand, it is important to adopt strategies and protocols that assure a given level of QoS for client applications so that the WSN can meet its design goals. On the other hand, while meeting QoS requirements the resource consumption of the WSN nodes must be managed in an efficient way in order to extend the network operational lifetime. Since the energy is the most critical resource in such networks, the tradeoff between energy saving and QoS provision is a major concern.

Clustering has been used both in ad hoc networks as in WSNs as an effective technique for extending the network lifetime, and supporting network scalability [2], [3]. The general idea is to perform the cluster formation based on the received signal strength indicator (RSSI) and to use local cluster heads (CHs) as routers of data gathered by the sensors in their clusters towards the sink node. Since the distance among cluster members and the respective cluster-head is, in general, smaller than the distance between these sensors and the sink, sensors in a cluster save transmission energy. Clustering can also be beneficial for purposes of energy saving because it favors data fusion procedures. Cluster members can collaborate about recent data measurements and determine how much information should be transmitted to the sink node. By averaging data values collected within the cluster, the algorithm can trade data resolution for transmission power.

Some advantages of network clustering can be identified as [4] (i) size reducing of the routing tables stored at each individual sensor node, (ii) saving the communication bandwidth, since it limits the scope of the inter-cluster interactions to CHs and avoids redundant exchange of messages among sensor nodes, (iii) stabilizing the network topology at the level of sensors and thus cuts on topology maintenance overhead and (iv) the use of optimized management strategies to prolong the battery life of the individual sensors.

Several works using clustering for WSNs have been developed [4], [5], [2], [6]. However, although in some of these works the authors use information from sensor nodes to make decisions about the network organization, the semantic of collected data to group sensor nodes into clusters is not frequently exploited [7], [8]. In this work we propose a semantic clustering for WSNs in order to minimize the communication resource usage and energy cost. The key feature of our proposal is a computation of semantic neighborhood relationships based on collected data from sensor nodes. We define as semantic neighbors the set of sensors that are semantically related. In traditional clustering approaches, when the concept of neighborhood relates to the geographical distance between nodes, sometimes neighboring nodes cover areas which are not related at all, or neighboring nodes provide measurements that are not semantically related. For example, in the airport security applications [9], sensor nodes both do video and audio processing and communicate with their neighboring nodes in order to share a global view of the monitored environment. However, sensors nodes fixed on opposite sides of a same wall will cover different areas of this airport environment that are not related at all.

It is also worth emphasizing that a semantic clustering method favors applying fusion/aggregation algorithms only in semantically related data within the network. For example, in an airport security application, if a cluster consists of nodes fixed on opposite sides of the same wall, the CH will not be able to eliminate any redundancy (one of the goals of fusion/aggregation data techniques) before sending messages to the sink, since the data sent from individual sensors are not related to the same monitored environment. Thus, the semantic clustering proposal brings advantages compared to the clusters generated only by taking into account the distance between the sensor nodes. Moreover, in case of applications based on events, it is interesting that clusters encompass only nodes (semantic neighbors) that detect the event of interest, while the unrelated nodes can remain working at a lower duty cycle in order to save energy.

In our work, the set of semantic neighbors is grouped into semantic clusters. The semantic clustering is a service provided by a semantic middleware for WSNs described in our previous work [7]. A fuzzy system is responsible to establish the relationships of the semantic neighborhood. Fuzzy inference systems match two of the most challenging requirements [10] of WSNs: (i) they are simple and can be executed on limited hardware and (ii) they can deal with imprecise data. Several WSNs applications use crisp values to specify the numerical variables that characterize a monitored event. For example, in a fire detection application, the WSN might classify an event when the temperature is above 50 °C. However, sensor readings are not always accurate and the sensor nodes, even if they are neighbors, often measure different values. In that example, if a sensor report 49.9 °C, a wrong decision will be made. Therefore, the ability of handling imprecise data is desirable since individual data from sensor nodes often are inaccurate due to calibration problems, environmental noise, wireless transmission loss, faulty sensors, among other items. Thus, fuzzy logic can handle the fluctuating of sensor readings in a proper way.

Our clustering method can minimize the computational resources usage because (i) involves only the relevant nodes to the event monitored by the network; (ii) the semantic neighbors can be grouped into a semantic cluster that provides the use of techniques of local collaboration as data aggregation; and (iii) nodes semantically unrelated can operate in low duty cycle. Moreover, by using the semantic clustering, we have the potential to improve the data accuracy provided for the application, where accuracy is a typical QoS requirement in WSN applications.

In this work, our proposal is applied in the structural health monitoring (SHM) domain. Some experiments were performed in order to evaluate the impact of our proposal in the network. First, we evaluate the proposal in two different scenarios of damage in the WSN. Second, we evaluate the impact of different aggregation method in the proposal related to energy saving. Third, our proposal was compared, using the same data set, with two algorithms: a deterministic clustering and the well-known LEACH (low-energy adaptive clustering hierarchy) [11]. This comparison includes the evaluating of the data accuracy and the energy consumed by the clustering phase. It is shown that our approach is an energy-efficient clustering method for WSNs, where it outperforms both the deterministic clustering and LEACH methods in about 70% and 47% of energy save respectively.