Abstract
The trend to implement the monitoring system with a wireless sensor network has been becoming urgent due to guaranteed flight safety and the passengers comfortability in travel. In this paper, a new dynamic routing algorithm is proposed to prolong the lifetime of the monitoring system with a distributed network based on the K-coverage method, and filter algorithm to be used for data fusion. Finally, the simulation results validate the effectiveness of the proposed approach.
1. Introduction
In recent years, the civil aviation industry has developed rapidly in China. Passenger and cargo transitions have maintained fast growth from 2014 to 2018. In this case, air travel is considered to be the priority option [1]. The flight safety, environmental quality, and comfort in cabin are focused by passengers. [2]. To meet the extremely high requirements of flight safety and avoid the miss alarms, the sensors used in the monitoring system are always with strict sensitivities. However, this kind of system will lead to false alarms. For examples, on November 13, 2017, the flight CZ6406 of China Southern Airlines was diverted to Changsha Airport due to a fire false alarm in cargo during flight. On March 19, 2019, a Boeing 777 aircraft of British Airways was made a forced landing in St. John's, Canada, after takeoff because of the same issue [3]. The frequent occurrence of false alarm events will cause a lot of economic losses and give passengers uncomfortable travel experience. Therefore, it is very important to reduce the fire false alarm rate in airplanes. At present, wired single-point sensors in airplanes can no longer meet the practical demands. A wireless sensor network including a large number of nodes with a specific algorithm can be used to play an important role than that of single-point wired sensors [4]. Furthermore, this kind of network can be applied in different areas such as UAV systems and environment monitoring. [5–8]. Wang et al. designed a WSN for pollutant monitoring in the cabin [7] and analyzed network failures cases [8]. They also proposed to use WSN for monitoring fire in the cargo compartment of commercial aircraft and deployed sensors based on the K-coverage index redundantly. However, they did not discuss the energy consumption and lifetime of WSN for aircraft applications [9]. In this article, a dynamic deployment method of WSN is discussed. This method is mainly focused on the dynamic balance of coverage and energy consumption in WSN aircraft environment application.
The coverage of the monitoring area and the available service time of the network are two important indicators needed to be considered for a WSN operation. So an energy-saving approach applied in cabin WSN based on the low-energy adaptive clustering hierarchy (LEACH) protocol is taken into consideration [10]. However, due to the strong randomness of the LEACH protocol in the cluster establishment phase, it will lead to creating unreasonable topology [11]. Many articles combined LEACH with other algorithm to improve overall performance. Kaddi et al. proposed a kangaroo method-based LEACH protocol, which has a good energy consumption performance and can prolong WSN lifetime [12]. Mohapatra et al. proposed a partitioned-based energy-efficient-LEACH (PE-LEACH) protocol which tends to the energy-based fault-tolerant technique, and it performs better than the LEACH protocol [13].
Therefore, in this article, a novel coverage index, K-coverage, with a certain probability is proposed to be a criterion of network lifetime. Then an improved binary artificial bee colony-LEACH (LEACH, LEACH-IBABC) algorithm is proposed with the index mentioned above. This approach makes active nodes and cluster head nodes to the global optimization in order to reduce the energy consumption and extend the lifetime of WSN. The structure of this article is arranged as follows: First, a probabilistic sensing model with a dynamic K-coverage deployment index with a certain probability is proposed. Second, a new LEACH-IBABC algorithm is proposed for the dynamic K-coverage deployment of WSN applied in aircraft cabins. Third, simulation verification and analysis are discussed. Finally, conclusions are drawn. The main contribution of this article is to combine the classical LEACH protocol with a new IBABC algorithm to reduce WSN energy consumptions and prolong network lifetime based on the constraint of the new coverage index, dynamic K-coverage.
2. Sensing Model and Coverage Rate Index of Wireless Sensor Network
2.1. Probabilistic Sensing Model of Wireless Sensor Nodes
In practical applications, the sensing probability of the monitoring grid points does not simply follow a Boolean model. It is determined by the distance, dml, between the sensor node (SN) and the monitored target, the physical parameters of the node, and the interference of the surrounding environment. The sensing probability can more accurately reflect the coverage capability of wireless sensor nodes for the real environment [14]. The probability mentioned above varying with distance dml is noted as the probabilistic sensing model (PSM).
Definition 1. Effective sensing radius: in terms of the probabilistic sensing model, when dml is smaller than the effective sensing radius Rs,max of the wireless sensor node, the target point to be monitored (TPM) can be effectively monitored and covered, so the sensing probability pml is 1. Otherwise, the probability pml is less than 1. Rs,max is a threshold value of dml. Furthermore, that probability exponentially decreases as the Euclidean distance between SN and TPM increases.
Suppose that the deployed sensor nodes are homogeneous in this article, the sensing model is optimized and obtained in (1) by deriving from the model proposed in [15]:where m is the index of SN. l is the index of TPM. And the parameter α (α > 0) describes the decreasing rate of the sensing (monitoring) probability pml when distance dml increases. As dml is continuously increasing till pml is less than a predefined threshold pthr, the possibility of TMP successfully monitored is too small to make error monitoring results. Therefore, pml is zero in this condition. pthr (0 < pthr < 1) is a relatively small value close to 0.
To calculate the threshold value of dml, let , then the threshold value of dml can be obtained as shown in the following equation:Substituting (2) into (1), we can obtain the following equation:Equation (3) shows a complete piecewise function of pml according to variable dml.
Taking an effective sensing radius Rs,max = 1.5 m as an example, Figure 1 shows the change of the monitoring probability pml as the distance dml varies.
As shown in Figure 1 (α = 0.3), if the distance between SN and TPM is 0, the monitored probability of TPM is 1. If the distance between SN and TPM is larger than Rs,max, the probability value pml decreases smoothly. And the probability decreases sharply at the beginning, and the decreasing speed is faster in the early stage than that in the latter stage. When the distance is larger than pml is 0.
2.2. Probabilistic Sensing Model-Based Dynamic K-Coverage Deployment with a Certain Probability of WSN in Aircraft Cabin
In general, the coverage rate of wireless sensor networks is defined as shown below:where the length of the area to be monitored is L. The width is W. The height is H. And the set of all sensor nodes S = {s1, s2, …, sNs}. Area (si) represents the coverage area of SN, which is the center of a sphere with the sensing radius. Due to the complexity of the calculation, a regional three-dimensional meshing method is used to find a deployment solution to WSN in aircraft cabin. In Figure 2, the solid dots are wireless sensor nodes. The hollow dots are grid points. In this article, consider each grid point as a TPM in the aircraft environment.
As mentioned before, the monitoring probability matrix P = [pml]NS × NGP where m = 1, 2, …, Ns, Ns is the number of sensor nodes, l = 1, 2, …, NGP, and NGP is the total number of grid points (equivalent to TPM). The set of all sensor nodes is expressed as S = {sq |q ∈ [1, Ns]}.
For any grid point l, the covering probability is the joint probability shown below:
In this article, a constant value, pA, is defined as monitoring accuracy. And the joint probability of all monitored TPMs should not be smaller than pA. The parameter pA should be larger for the higher requirements of environments.
When K = 1 and the wireless sensor network has achieved the coverage for grid point l with joint probability pA.
When K > 1, assume that the subset of the sensor nodes SENq ⊆ S contains ns sensor nodes (1 ≤ ns ≤ Ns), and each sensor node in the subset is different from others. Each subset SENq is independent of others:
If there are at least K SENq subsets and all the sensor nodes of each subset cover the grid point l with the joint probability pA, all the nodes of set S achieve K-coverage at grid point l with the probability pA. Moreover, if each grid point in the area to be monitored is K-coverage with probability pA, the wireless sensor network can achieve K-coverage with probability pA in that area. By changing the values of the monitoring accuracy pA and the coverage degree K, the coverage index of the network can be adjusted.
In this article, when plenty of sensor nodes have been deployed in the aircraft cabin, the dynamic K-coverage and deployment algorithm is used to decide which sensor nodes are activated in each cycle, so that the active nodes of the wireless sensor network should ensure that each grid point in the cabin is covered by at least K groups of sensor nodes with probability pA.
Parameters K and pA are two important coverage indicators of WSN. The larger the values of K and pA are, the more the nodes are required, the more accurate the fire monitor is, however, the higher the energy consumption is. On the contrary, it is impossible to meet the coverage and accuracy requirements if these two parameters are too small. In the next section, a new routing protocol is proposed to save energy.
3. Intelligent-Based Dynamic Routing Approach to Energy saving
A routing protocol is one of the key technologies in wireless sensor network-related technologies. In recent years, four types of protocols have been developed for wireless sensor networks. They are geographic-based routing protocols, data center-based routing protocols, cluster-based routing protocols, and hybrid routing protocols. And each type of protocol has produced many branches.
The cluster-based routing protocol separates the sensor nodes into different groups. And each group of sensor nodes is organized by a cluster head (CH). All cluster heads are controlled by a base station (BS) or sink node. The LEACH protocol is the most basic and popular cluster-based routing protocol. The clustering method of the LEACH protocol is shown in Figure 3.
3.1. Energy Consumption Model of Sensor Node
To prolong the lifetime of WSN applied in aircraft, it is necessary to develop an energy-saving algorithm.
The energy consumption of nodes in wireless sensor networks is mainly divided into the following categories:(1)The inherent operating energy consumption It is created from SN hardware themselves without sensing and communicating with other nodes.(2)The energy consumption on collecting and sensing It occurs in collecting information and detecting the environment. The time interval of collecting environmental information, Tsense, will affect the energy consumption speed.(3)The energy consumption of data transmission It occurs in transmitting the sensing data amplified by the power amplifiers.
Figure 4 is a block diagram of a conventional wireless sensor network node module.
A wireless sensor network node is composed of four hardware modules: sensing module, processing module, wireless communication module, and energy supply module.
According to the energy consumption model in [16], the energy consumption of data transmitting and receiving from sensor node m1 to m2 can be visually shown in Figure 5.
As shown in Figure 5, the energy consumed by the transmitter can be calculated by using the following equation:where L is the number of bits of data transmitted between two nodes. Parameter d is the distance between any two communication nodes. Eelec represents the energy consumption of the circuit. Parameters εfs and εmp represent the energy consumption of the transmitter amplifier per square meter per bit. d0 is the threshold distance and is defined in equation (8). If d is less than d0, the free space channel model is adopted, and the power amplifier coefficient is εfs. Otherwise, the multipath attenuation model is used, and the power amplifier coefficient is εmp:
The energy consumption of the receiver is shown in the following equation:
The energy consumption on data fusion of the cluster head can be calculated by the following equation:where EDA represents the required energy for data fusion per bit assuming that there is no energy consumption in the collecting process.
3.2. LEACH-IBABC Algorithm
This section proposes a new protocol algorithm-LEACH-IBABC based on the basic LEACH protocol and energy consumption model mentioned above.
Normally, “round” is a basic unit in the LEACH protocol. Each node of the wireless sensor network has a specific probability to be a cluster head in one round. Based on this idea, the load of the network can be distributed to each node uniformly to prolong the lifetime of the network [17].
A wireless sensor network organized by the LEACH protocol includes at least one base station or sink node. The base station or sink node collects data from all cluster heads without considering the energy consumption. In this article, all nodes are assumed to own the same initial energy, and the transmission energy consumption is symmetrical between the communication pair. A node can adjust the transmission power automatically to minimize energy consumption by calculating the distance between the transmitter and receiver ends based on the strength of the received signal.
All sensor nodes participate in coverage and sensing for the basic LEACH protocol. Therefore, the network coverage is high in the early operation stage. However, when some nodes’ energy is exhausted, the network coverage ratio might not be guaranteed and may decrease rapidly. Then the wireless sensor network with a low coverage ratio is not suitable for aircraft cabin scenarios. However, as mentioned above, WSN should ensure that the coverage ratio meets the K-coverage index with probability pA to decrease false alarms and missed alarms.
Inspired by the LEACH-C algorithm, the base station can be used as a center to manage the status of each node (working or sleeping) and cluster working mode (cluster head or cluster member) [18]. This algorithm operates in the cluster establishment phase of each round. In the operation process, an optimal node subset SD selected from the K-covered node set S is set to work mode and participates in the dynamic coverage of WSN in this round. And other nodes are in sleep mode to save energy. It is an NP hard problem to select the optimal node subset. To solve this problem, the artificial bee colony algorithm-based LEACH protocol is proposed. In this article, since each node has only two states: working and sleeping, a binary artificial bee colony (BABC) algorithm is used. Compared to the classic ABC algorithm, the improved BABC algorithm uses different search formulas at each stage.
The objective of the optimal operating node subset is selected to satisfy K-coverage with probability pA with most Erest_sum and least ns. Erest_sum is the total residual energy of the operating nodes at the end of this round, and ns is the total number of operating nodes. In this case, it will not only balance the network load energy consumption and allow nodes with the highest energy to work first, but also ensure coverage ratio and avoid triggering too many nodes simultaneously. Therefore, the total energy consumption of the wireless sensor network is reduced.
In the initialization stage, the food source corresponds to a feasible solution to the practical problem. At this stage, each dimension of each feasible solution generates a random number rij (0 < rij <1), and the value of the dimension is determined according to equation (11). Then each feasible solution is assigned to an employed bee:where i = 1, 2, ..., NP, j = 1, 2, ..., Ns, and is the jth dimension of the ith feasible solution. NP is the number of feasible solutions, also represents the number of employed bees. Ns is the number of sensor nodes, also represents the dimension of the feasible solution in this article. The initial feasible solution of the IBABC algorithm is shown in Figure 6.
After allocating the employed bee to each feasible solution in the initialization stage, the fitness value of each solution is calculated by equation (12). The fitness value of the solution corresponds to the quality of the feasible solution:where fiti (k) represents the fitness value of the ith solution in the kth iteration. Erest_sum(k) represents the total predicted residual energy at the end of the current round according to the current topology. COV (k, S) means the network coverage ratio in the kth iteration. ns (k) is the number of operating nodes in the kth iteration. The impact factor of the node numbers, α, is a constant equal to 0.1 in this article. In general, at the end of the current round, the more the residual energy is, the larger the network coverage ratio is, the fewer the number of awakened nodes are, the larger the fitness value and the better the quality of the solution are. The steps for the coverage ratio COV (S) based on the probability sensing model are shown in Table 1.
In the employed bee stage, each employed bee i selects a neighbor u randomly and generates a random number φij (0 < φij < 1). Then the candidate feasible solution is generated according to the following equation:where k represents the current iteration. i ≠ u. That means the selected neighbor cannot be the same as the current feasible solution. is the jth dimension of the ith solution. is the jth dimension of the feasible solution of a selected neighbor, u. is derived from and . The employed bee uses a greedy selection method to judge whether to select the candidate feasible solution. If the candidate feasible solution is better than the original feasible solution, the original feasible solution will be replaced. Otherwise, the original state is unchanged, and the cumulative times of inactivity for a feasible solution increases by 1.
In the follower bee stage, the probability of each feasible solution selected by the follower bee can be calculated by the following equation:where k represents the current iteration. pi(k) is the probability that the ith feasible solution is selected by a follower bee in the kth iteration. If a follower bee selects a feasible solution, a candidate feasible solution is calculated using equation (13) to improve the quality of the current feasible solution. If the quality of the candidate feasible solution is better than that of the current feasible solution, the current feasible solution is replaced. Then the employed bee uses this new candidate feasible solution. Otherwise, the original state remains unchanged, and the cumulative times of inactivity for a feasible solution increases by 1. This is similar to that in the employed bee stage.
In the scout bee stage, the cumulative times of inactivity for a feasible solution is checked with the threshold Limit. If the variable value is larger than Limit, the employed bee for exploring the current feasible solution changes into a scouter. And a new feasible solution is generated by equation (11), and the cumulative times of inactivity for a feasible solution is reset. Then the scouter changes into an employed bee. In the process, only one scouter exists. If the scouter stage is over, the optimal solution of all feasible solutions is recorded.
After selecting the optimal operating node subset SD, it is necessary to find a group of cluster head node subset SDH from SD to act as the cluster heads in this round. Since the dimension of SD is not too large, the fitness value of all feasible solutions can be obtained by the ergodic method. The steps are listed as follows:(1)To predict the energy consumption of each node in the current round according to the energy consumption model and calculate the residual energy of each node.(2)To ensure that the total energy consumption of the operating nodes is the lowest under the SDH network topology, and make sure that the standard deviation of the node energy consumption is the smallest. The fitness value is calculated by using the following equation:
In equation (15), fitness value fiti_DH is used to evaluate the topology quality of the ith feasible solution. If the consumption of energy of the whole WSN and the standard deviation of the node energy consumption are smaller simultaneously, the fitness value is smaller, and the quality of this topology organized by LEACH-IBABC is better, which brings a more balanced energy consumption.
In the aircraft environment, it is necessary to ensure that the operating nodes can meet the coverage index. In this article, the network is considered to get the maximum lifetime when the coverage index cannot be reached due to the exhaustion of some nodes. Figure 7 shows the flow chart of the LEACH-IBABC algorithm.
4. Simulation and Analysis of the Proposed Method
This section presents the simulation validation and compares the lifetime of WSN between the LEACH and LEACH-IBABC algorithms. This article considers a WSN with the K-coverage (K = 3) index deployed in an aircraft cargo model with 80 sensor nodes in total. Each node has the same initial energy. The simulation parameters are set as follows:(1)Cluster head selection probability p = 0.1(2)The initial energy of each sensor node E0 = 0.02 J(3)Transmitting and receiving circuit energy consumption Eelec = 50 pJ/(bit m−2)(4)Transmitter amplifier energy consumption parameters εfs = 10 pJ/(bit m−2) and εmp = 0.0013 pJ/(bit m−4)(5)Data fusion energy consumption EDA = 5 pJ/bit(6)The data packet length Ld = 4000 bit, and the control packet length Lc = 32 bit(7)PSM is used as the sensing model of sensor nodes, and the monitoring accuracy pA = 0.99(8)Coverage degree K = 1
The corresponding simulation results are shown in Figures 8 and 9. As illustrated in Figure 8, the network using the LEACH algorithm cannot meet 2-coverage with probability 0.99 after the 106th round and cannot meet the 2-coverage with probability 0.9 after the 131st round. However, the network with the LEACH-IBABC algorithm cannot meet 2-coverage with probability 0.99 after the 154th round and cannot meet 2-coverage with probability 0.9 after the 189th round. As curves shown in Figure 8, the LEACH-IBABC algorithm prolongs the network lifetime. When the monitoring coverage index is 2-coverage with probability 0.99, the WSN runs 48 more rounds, so the network lifetime is increased by 45.3%. When the monitoring coverage index is 2-coverage with probability 0.9, the WSN runs 58 more rounds, so the network lifetime is increased by 44.3%.
The residual energy comparisons between two algorithms as the number of active rounds change are shown in Figure 9. When it goes to the 106th round, the residual energy of WSN using LEACH-IBABC is 2.43 J more than the WSN using LEACH. This helps the WSN using LEACH-IBABC runs 48 more rounds. When it goes to the 131th round, the residual energy of WSN using LEACH-IBABC is 3.18 J more than the WSN using LEACH. This helps the WSN using LEACH-IBABC runs 58 more rounds. The total residual energy by using the LEACH-IBABC algorithm reduces slower than that of the LEACH algorithm in the early stage.
Therefore, the LEACH-IBABC approach can be used to organize the global optimal topology and reduce the overall energy consumption of the network. On one hand, the base station handles the task of topology allocation of WSN, so the calculation burden of the nodes is reduced greatly. On the other hand, each node only needs to select the different states according to the base station information and send the collected sensing information to the cluster head directly. Therefore, the energy consumption of nodes is reduced, and the lifetime of the network is prolonged.
5. Conclusions
In this article, a dynamic routing approach to WSN, the LEACH-IBABC algorithm, is proposed for the aircraft environment. In this method, the IBABC algorithm is designed to select the operating node subset SD which is a global optimization by using the artificial bee colony algorithm. In this case, the nodes with more residual energy have the priority to participate in the current round and meet the requirements of the K-coverage index with a probability pA. The others are in sleep mode. Then the optimal cluster head node subset SDH is selected with the proposed algorithm to decrease energy consumption and make a better load balance. Therefore, the information selected from the nodes is estimated by using the Kalman filter. Finally, the simulation results prove that the proposed algorithm can reduce energy consumptions and prolong network lifetime based on the constraint of the coverage index.
Data Availability
The figures.zip data used to support the findings of this study are available from the corresponding author upon request.
Conflicts of Interest
The authors declare that there are no conflicts of interest regarding the publication of this paper.
Acknowledgments
The authors would like to extend our sincere gratitude to Prof. Youmin Zhang from Concordia University, Canada, for his instructive advice and suggestions on this paper. This work was supported by Tianjin Natural Science Foundation, 18JCYBJC42300, and Scientific Research Project of Tianjin Education Commission, 2019KJ143.