On Event/Time Triggered and Distributed Analysis of a WSN System for Event Detection, Using Fuzzy Logic

Abstract

Event detection in realistic WSN environments is a critical research domain, while the environmental monitoring comprises one of its most pronounced applications. Although efforts related to the environmental applications have been presented in the current literature, there is a significant lack of investigation on the performance of such systems, when applied in wireless environments. Aiming at addressing this shortage, in this paper an advanced multimodal approach is followed based on fuzzy logic. The proposed fuzzy inference system (FIS) is implemented on TelosB motes and evaluates the probability of fire detection while aiming towards power conservation. Additionally to a straightforward centralized approach, a distributed implementation of the above FIS is also proposed, aiming towards network congestion reduction while optimally distributing the energy consumption among network nodes so as to maximize network lifetime. Moreover this work proposes an event based execution of the aforementioned FIS aiming to further reduce the computational as well as the communication cost, compared to a periodical time triggered FIS execution. As a final contribution, performance metrics acquired from all the proposed FIS implementation techniques are thoroughly compared and analyzed with respect to critical network conditions aiming to offer realistic evaluation and thus objective conclusions’ extraction.1. IntroductionIn recent years, wireless sensor networks (WSNs) have emerged as a promising research field and have been applied to a wide variety of application domains including industrial control, environmental monitoring, and healthcare applications. The primary objective in such WSN applications is the accurate and reliable monitoring of an environment, based on the processing of multiple and diverse sensors values and the identification of irregular situations or dynamic real life events. The collaborative tasks lead to specific action scenarios, so as to control the monitored environment. The process of observing a real phenomenon and evaluating its behaviour in WSNs is known as event detection [1].With respect to the dependency upon the input signals, real events are distinguished in two categories: single modality and multimodality events [2]. The former concerns the examination of the monitored values of each parameter independently, based on the assumption that if any of these exceed a specific “normal” range, an event occurs [2]. The latter category includes the multimodal events which are based on the correlation of several attributes, the processing of which evaluates the occurrence of an event [3]. Critical challenges of event detection algorithms in WSN include energy saving, data integrity, and in-depth understanding of the monitored environment. In order to meet such objectives, the development of a classification model is essential for the accurate identification of an event, along with the reduction of the communication as well as processing overhead. The classification of an event can be defined as the process of evaluating an event of interest using multiple sensor nodes (multimodal event). Such processing may vary from a trivial rule engine machine to a complex machine learning algorithm, while the final outcome of this process triggers specific action scenarios. Taking into consideration typical WSN characteristics, the classification of an event is strongly affected by the quality and characteristics of the communication channel between the monitoring and actuation units in order to optimize the monitoring and control of the environment.Considering an application which is based on single modality events, when a sensor value exceeds an upper/lower threshold (i.e., temperature in an environmental monitoring application), an event generation is indicated (i.e., fire alarm). However, in many cases, such decisions may lead to false alarms, since most of the real life events depend on multiple monitored parameters in a correlated manner. For example, in the fire alarm scenario, an accurate decision should take into consideration the existence of smoke and luminosity level along with the temperature value. Alarm situations trigger specific reactions and, thus, node-to-node communication (actor-to-actor coordination schemes) [4]. Therefore, false alarms will lead to an application’s performance degradation, as well as to a network traffic and energy consumption increase. Hence, the need for more sophisticated multimodality classification processes that will maximize the application’s accuracy while mitigating resource wastage among the actuation units arises. Additionally, taking into consideration that the communication modules are the most energy consuming components of a sensor node, the lifetime of the node and the network’s robustness are anticipated to be accordingly benefited. In that respect, the utilization of classification algorithms [5] (i.e., fuzzy inference systems, FIS) in WSNs regarding the identification of complex events is crucial in achieving the aforementioned objectives.Towards this goal, existing data mining techniques can offer several algorithmic solutions [6] to this field. However, conventional implementation requires high processing capabilities and abundant memory availability, in order to meet specific execution time restrictions. Such assumptions contradict typical WSN characteristics, where the sensor nodes suffer from limited processing power and available memory. Such characteristics, in combination with the error prone nature of wireless communications, highlight the challenge for designing distributed, highly efficient, yet of low complexity and low resource-demanding data mining algorithms in WSNs. Furthermore, in realistic WSN applications, the distributed implementation of such computational intensive algorithms can be highly beneficial towards balancing CPU load among several nodes. Specifically, the distributed implementation of an algorithm increases on-site processing and can potentially reduce the number of data packet transmissions, leading to bandwidth conservation, network data transfer relaxation, and energy consumption degradation.Another critical aspect drastically affecting FIS algorithm design when aiming towards WSN application concerns the effect of network characteristics upon the algorithm’s performance. An indicative example could concern the case where the outcome of the classification process is extracted using invalid inputs. This could be caused by several factors leading to input values potentially not being up to date because of unpredictable transient problematic network conditions. Indicatively, such conditions include network congestion, resulting in increased packet loss and delay, or node mobility leading to network disconnection. Traditional data mining approaches do not consider these problems. Specifically, the input data are assumed to be always valid (i.e., in time), while the execution delay is assumed negligible due to abundant processing resources.Driven by such observations, distinguishing WSN from other traditional network areas is essential. A valuable contribution of this work, compared to relative ones, concerns a comprehensive study on the effect of such conditions in the context of realistic WSN environments. Towards this objective this paper also presents a respective framework enabling application of existing data mining algorithms in such cases, accounting for the distributed processing power, the communication cost, and the algorithm’s sensitivity to invalid input data.In this work, we study the proposed fuzzy logic system in an environmental scenario by simulating a realistic WSN infrastructure characterized by significant communication challenges. In our previous work [7] a centralized implementation of a healthcare FIS was presented, where a TelosB mote was considered as a cluster head, being responsible for the reception of all the generated packets and the FIS execution. In [8] the respective distributed implementation was presented. The evaluation results of both efforts proved the sensitivity of the system’s performance to the networking conditions and the cluster head’s overloading. Driven by these observations the main goal and contribution of this paper concern the proposal of novel and efficient approaches on implementing data mining algorithms specifically targeting WSNs applications as well as undertaking a respective comprehensive performance evaluation. Towards this objective, this work investigates the FIS’s performance under three different scenarios: centralized and distributed time triggered as well as centralized event triggered execution. Respective evaluation highlights in a quantifiable way that the centralized approach burdens the CPU utilization of the FCHN, due to the fact that all data flows are directed to it. Driven by these observations as well as taking into consideration the scarce CPU and wireless bandwidth availability in WSNs, we propose a distributed approach to partition the execution of the FIS among the nodes. In this way we aim to optimally balance the energy consumption between the nodes. Moreover, respective measurements indicate a significant diffusion of network traffic avoiding the all-to-one communication scenarios. Moreover, an exploration is conducted on the way wireless channel conditions and the packet arrival rate affect the overall performance of the FIS and the final outcome of the event detection system. Although the distributed time triggered approach is anticipated to balance the energy consumption, the evaluation revealed additional aspect and interdependencies, needing in-depth investigation. Typically, the FIS algorithm is executed periodically based on a time triggered approach. In case that the occurrence of abnormal events is rare, the periodic nature handles CPU resources and network bandwidth in a nonoptimal way. For this reason, our work proposes an eve