Many recent studies have focused on various aspects of wireless sensor networks (WSNs) [1]. These networks are composed of the following: A significant number of sensor nodes, which consist of a processing unit with limited computational capability and memory; sensors, a wireless communication device; and a limited power source. Although sensor nodes have strong constraints on their energy resources and computational capacity, they can be used for continuous sensing, event detection, event identification, location sensing, and local control of actuators [2]. As a result, the range of applications of WSNs is very wide, and includes environmental monitoring systems, intelligent agriculture, surveillance, health monitoring, traffic monitoring, and industrial control and monitoring.

In spite of the fact that sensor nodes have highly constrained resources (microcontroller, memory, battery, communications), numerous new functionalities have been proposed for WSNs. An example of an application adapted to WSNs is the integration of soft computing (SC) technologies, such as fuzzy logic, neuronal networks, and fuzzy rule-based systems (FRBSs) in sensor nodes [3]. FRBSs are considered to be knowledge-based systems in which the knowledge of the system is represented by a set of “IF-THEN” rules whose antecedents and consequents are composed of fuzzy logic statements (fuzzy rules). One of the main characteristics of these systems is the capacity to incorporate human knowledge by accounting for its lack of accuracy and uncertainty or imprecision.

On the other hand, WSNs represent an ideal scenario for distributed applications in which several application layer protocols have been proposed, such as sensor management protocols, task assignment and data advertisement protocols, and sensor query and data dissemination protocols.

Although considerable research has been devoted to different WSN applications, less attention has been paid to knowledge-based distributed applications for WSNs. In this sense, it is necessary to design an application layer protocol that allows the distribution of knowledge bases (KBs) among the sensor nodes, the collection of inferred data, the management of the sensor nodes, and the integration with other measurement applications.

Following our previous research on FRBS WSNs [4–6], this work proposes a new distributed system composed of multi-agent knowledge-based sensor nodes. The contributions of this work are as follows:

We propose a distributed system in which every sensor node can execute a set of applications (FRBS, Collaborative FRBS, measurements, actions, etc.).

The rest of the paper is structured as follows: the next section deals with related work and motivation. Section 3 describes the methodology that was used, including the multi-agent structure of the sensor nodes and a description of the application protocol. Section 4 shows the experimental results related to the application protocol performance, as well as two real-world applications of the proposed system. Finally, conclusions are presented in Section 5.

Communication capabilities are one of the main characteristics in sensor networks, and enable the networks to share information, knowledge, and sensing effort in sensor nodes. The introduction of intelligent capabilities into sensor networks requires the use of communication resources and their optimization. In this sense, Brignell [7] defined an intelligent sensor as one that modifies its internal behavior to optimize its ability to collect data from the physical world and to communicate the data in a responsive manner to a host system. Benoit et al. [8] presented a model of intelligent sensor systems that emphasized the ability to exchange knowledge with other actors. Karlsson [9] defined an intelligent sensor network as autonomous sensor nodes that exchange information, reason, and collaborate with each other. The specific application implemented should preserve energy resources and work as one unit when delivering fused and compiled sensor information to the end user. A new structural concept of intelligent sensors and networks with intelligent agents which provide communications elements was suggested by Mekid [10].

The past few years have witnessed a growing interest in the use of techniques based on SC to optimize the communication process between intelligent sensors. In this sense, the use of Artificial Neural Networks to discover redundant input data was proposed in [11]. Cui et al. [12] proposed a FLC algorithm to ensure that the sensor network attains a large coverage region and maintains dynamic ad hoc network connectivity between nodes. Shu [13] proposed a fuzzy optimization algorithm (FRBS) to efficiently adjust the sensor placement after an initial random deployment. A fuzzy logic control based QoS management scheme for WSANs was developed in [14]. It utilized a fuzzy logic controller inside each source sensor node to adapt the sampling period to the deadline miss ratio associated with data transmission from the sensor to the actuator. Averkin [15] showed a combination of embedded fuzzy logic and neural network models for information processing in complex environments. The most interesting aspect of this approach is the use of a WSN as a distributed computing environment for intelligent data processing methods. Srinivasan [16] presented a novel scheme for data-centric multipath routing in wireless sensor networks utilizing a fuzzy logic controller architecture at each node in the network to determine its capability to transfer named data packets based on its own battery power levels and the type of data being forwarded.

Marin-Perianu [17] proposed a distributed general-purpose reasoning (D-FLER) algorithm that uses fuzzy logic for fusing individual and neighborhood observations. Nakamura [18] described how information fusion is closely related to data communication in WSNs.

As described in [19], collaboration enriches functionality and enhances scalability and manageability of networked sensor systems, particularly in those comprised of a large number of heterogeneous sensor networks deployed over a large area. Effective collaboration may require that various sensor networks share synchronized data replicas. That paper proposed a novel data replication mechanism suitable for the limited bandwidth of sensor networks. The scheme uses neural networks for scheduling of replication. In both [20] and [21], a cooperating object was defined as a single entity or a collection of entities consisting of sensors, controllers (information processors), actuators, or cooperating objects that communicate with each other and are able to autonomously achieve a common goal.

In all of the previous works, the use of SC techniques was proposed to fuse or aggregate raw data in nodes to reduce its redundancy and transmit only the processed data. In [6] and [17], the use of a collaborative algorithm, based on embedding FRBSs into WSNs, was proposed to implement these fusion tasks. This algorithm achieves an improvement in the reliability, responsiveness, and accuracy of the sensors in WSNs.

Akyildiz [1] proposed a sensor network communication architecture that defines a protocol stack composed of five layers. Akyildiz described three possible protocols inside the application layer, including a sensor management protocol, a task assignment and data advertisement protocol, and a sensor query and data dissemination protocol.

To the best of our knowledge, although many application layer protocols for sensor networks have been defined and proposed, protocols for KBs and fussed data delivery that are adapted to sensor limitations in sensor networks remains a largely unexplored region. In this sense, the application protocol presented in this paper provides an appropriate solution to fill these requirements.

The main objective of our research is to design and implement a distributed system, based on a WSN, that allows the execution of a set of applications in the sensors (measures, actuations, FRBS, collaborative FRBS, etc.), updating of the sensor KBs, obtaining sensor data and inferences generated in the sensors, and managing the sensors.

The proposed system is composed of a computer connected to the Internet, where the host application runs, an access point or base station that allows the host application to transmit data to the sensors, a WSN, a FRBS adapted for execution in the sensors, and the communication protocol. The main functions of the components are as follows:

Computer: edit the KBs (variables, fuzzy sets, and rules), access the sensor network, communicate with sensors, and monitor sensor state.

The proposed system was designed with the aim of minimizing sensor communication to prolong battery life. Therefore, the sensors operate in a work cycle in which they first execute the application (measure, actuate, infer its output, etc.), decide if it is necessary to connect with the computer or other sensors, and then are configured in a sleep mode. On the other hand, the protocol is independent of the platform, so it can be used in different sensors.

This work is focused on an application protocol that allows communication between the computer and sensors, as well as among sensors. The following sections show the multi-agent structure of the sensor software and a detailed description of the communication protocol.

This work follows our previous research on the integration of FRBSs in WSNs. In [6] we showed the structure of two knowledge-based systems (FRBS and collaborative FRBS) that were designed to be executed into a sensor, the first version of the KBs distribution protocol, and different ways to integrate data and knowledge. In the results we showed the knowledge-based sensor performance (execution time and consumption of the sensor due to the FRBS systems) and the evaluation of the collaborative scheme showing the effects of collaborative FRBS sensors on a modeling system of pests in the culture of the olive tree. Moreover, a comparative analysis of the FRBS sensor behavior, using its control surfaces, was shown in different situations.

The application protocol and the multi-agent structure proposed in this work allow the integration of several applications inside the sensor. Therefore, this work presents a noise pollution monitoring application in order to show that the proposed system not only supports knowledge-based applications, but also measurement applications.

The following sections show the experimental platform that was used in the experiments, the results of data distribution and robustness tests including the power consumption due to communication, and a noise pollution monitoring application that was executed in the application control agent.

This work presented a distributed system adapted for WSNs where sensor nodes present a multi-agent structure that allows sensors to execute a set of applications (such as FRBS, measurements, and actions), manage sensor services, and communicate with other sensors and a computer. Communications are carried out using an application layer protocol that allows the nodes to transmit and receive information such as KBs, inferred or collected data, and alarms. Two real world applications of the system, a FRBS and a noise pollution monitoring system, were shown.

The noise pollution monitoring application presented in this paper, which was composed of the application, the multi-agent software, and the application layer protocol, represents a new monitoring opportunity that can be used to obtain maps of noise levels in time and space.

Moreover, the results show that it is possible to use this system in a wide range of applications, as well as the effectiveness of the application layer protocol. The distribution times of the KBs obtained in the real WSN were very short, and both throughput and delay showed similar values to the values obtained in previous network simulator experiments.

The multi-agent structure presented in this paper allows users to integrate different applications in a sensor node in an effective way. The definition of a service interface in the application layer allows new applications to be designed independently of hardware, memory, communications, sensors, and actuators, and access all of the components in a standardized way.