Fuzzy logic based nodes distributed clustering for energy efficient fault tolerance in IoT-enabled WSN

Abstract

The Internet of Things (IoT) enabled wireless sensor network (WSN) is now widely employed in various sectors like smart city and vehicle transportation for their expanded capabilities such as data storage, access, and monitoring. The use of smart sensors that continuously collect data from the smart environment makes these possible. Furthermore, these facilitate the easy access of stored data over a secure IoT-gateway for mobile users. This device mobility that allows shifting to multiple locations, makes it challenging to route data across many access points. In this regard, it induces packet loss and improper node selection, which could result in connection failure and network unreliability. This study proposes a new data routing protocol called as Fuzzy Logic Nodes Distributed Clustering for Energy-Efficient Fault Tolerance (F-NDC-EEFT). It can be deployed on any network platform, including mobile and non-mobile nodes. It considers performance metrics such as delivery rate, withstand node aliveness, communication delay, and energy efficiency to find an optimized path for the better performance of IoT enabled WSNs. The clustering approach is applied to the instant data load, which divides it into the distinct node groups. When proposed algorithm is tested alongside existing routing protocols for performance, it is found to save energy, minimize the number of connection failures, boost the throughput, and increase the network’s lifetime.

Keywords

CH eligibility energy efficiency fuzzy modules energy aware routing protocol IoT enabled WSN

1 Introduction

The Internet of Things (IoT) enabled wireless sensor networks (WSNs) are networks of sensors deployed in different locations for collecting and transmitting data on environmental conditions over wireless links to a central server. These enable continuous environmental monitoring by transmitting the sensed data to the Base Station (BS) for taking apt management decisions [1]. IoT enabled WSNs are used in applications including healthcare [2], home automation [3], smart cities [4], and surveillance services [5]. Because of the limited power levels of the sensors, direct connection is not possible. They rely on multi-hop communication to reach their target [6]. As a result, it costs more energy to keep all sensors active and there are communication delays between sensor nodes. To address this issue, a clustering method is used to construct sensor node groups of varied sizes. Each group has a designated cluster head (CH) for regularizing energy limitations so that the network lifetime is increased [7]. This way, the sensed data is collected and transferred to the next hop node, and then to the next, until it reaches the BS. There are various clustering algorithms available for performing this task in various ways to accomplish specific purposes such as avoiding high latency, preventing low throughput, and lowering energy wastage.

However, a greater control overhead (OH) is required to provide a realistic flow of the sensing function through all the sensor node groups to equalize the work process [8]. This causes the accumulation of massive OH within the network, potentially causing energy imbalances between the sensor nodes. Reinforcement learning has facilitated a significant improvement in terms of latency, throughput, and energy by reducing the number of messages sent.

Feedback Routing for Optimizing Multiple Sinks (FROMS) [9] and Q-learning for role free clustering (CLIQUE) [10] are the most commonly used methods in IoT enabled WSNs for introducing the subspace scaling of the sensed data into a high dimension projection to trace out the required field. However, to keep the clustering technique generic, these are accompanied by big state action pairs. Deep reinforcement learning (DRL), [11] on the other hand, has exploded in popularity due to the synergistic benefits of deep neural network and traditional reinforcement learning. IoT-enabled WSNs are increasingly being used for a variety of services like air quality monitoring [12], data flow scheduling [13] and disaster modeling [14], among others.

1.1 Motivation of the research

Even when some sensor nodes in a WSN fail, the network as a whole can still function normally, a feature known as fault tolerance. Owing to their limited battery life and difficulty of recharging or replacing failure nodes, sensor nodes are deployed densely in the target area to increase coverage and connectivity. It is difficult to route data across many access points when devices are mobile and locations change frequently [15]. For, this could lead to connection failure and unreliability in the network as packet loss and incorrect node selection are induced. In these situations, the ratio of active to inactive nodes is indicative of the presence of faulty nodes in the network. Even with the dead faulty nodes, the network still gets the packets to the base station. End-to-end delay quantifies the typical time it takes for a packet to travel from its sending node to its receiving node in the base station [16]. Despite the discovery of the dead nodes in the network, the packets were delivered to the base station with a noticeable lag. The simulation results demonstrated the strong evidence that the proposed model is highly fault-tolerant in comparison to the existing methods, demonstrating the high degree of reliability of the approach.

1.2 Problem identification

Fault-tolerance in WSNs is what enables a network to continue providing its services despite the failure of some of its sensor nodes. Sensor nodes are deployed in a dense manner in the target area to increase coverage and connectivity [17]. Even with the presence of some dead faulty nodes, the network still manages to deliver the packets to the base station. Therefore, in this work, we focus primarily on ensuring that the node selected for cluster head has a high potential for data transfer over minimal communication delay, optimal energy utility, and an effective network lifetime, even under any data routing circumstance. Based on this concept, the present study has come up with the Fuzzy Logic Nodes Distributed Clustering for Energy-Efficient Fault Tolerance (F-NDC-EEFT). The following steps are employed to provide minimum delay, greater energy savings, and a longer lifespan:

Using three sets of Fuzzy modules, determine which CHs are suitable for cluster formation (I, II and III).

Using additional parameters, evaluate the iteration-specific eligibility criteria for nodes.

The result will be a reduction in energy consumption, network throughput, and lifespan.

The remainder of the paper is structured as follows: Section II reviews literature related to the previous works done in the field of employing fuzzy clustering strategy and reinforcement learning for effective data routing in IoT enabled WSNs. Section III describes the proposed model of cluster formation using Fuzzy modules (I, II and III) for energy-efficient fault tolerance. Section IV discusses the simulation results obtained therefrom (I, II and III). Finally, Section V concludes the study and outlines the future research directions to build on it.

2 Related works

This section focuses on the advancement of reinforcement learning in relation to data routing protocols in IoT-enabled WSNs. Routing, clustering, and deep reinforcement learning in relation to energy conservation are the three basic sections of the study. First, routing protocols are meant to find the next hop and ensure that the data reaches its destination in the simplest and most energy-efficient manner possible. Based on its current location and prior delivery time, we can infer the location of the next hop node. To begin with, the routing protocol uses the Q-learning concept to choose the next hop nodes based on delivery time [18]. There is a problem with network lifespan because of not laying restrictions on the battery usage for long data queues. In [19], a routing technique based on reinforcement learning that follows a spinning tree for minimizing traffic congestion was proposed. However, when used in large-scale WSNs, it suffers from a significant lag in communication. Therefore, the current study focuses on Q-learning based routing protocols that focus on latency, distance, possible hop count, and residual energy for identifying an effective path for data transfer between the source and the destination [20].

For a metaheuristic search, a WSN needs more control messages transmitted across nodes, which results in energy wastage. In [21] adaptive cooperative systems that promote network resilience and reduce transmission delays are discussed. Individual nodes in the cluster group have not been able to maintain their balance of work. Using Multi-Agent Reinforcement Learning (MRL) [22], the buffering interval for the data in a queue is changed so that it reaches its lowest energy level. Due to these issues, the packet delivery rates and latency are affected.

The second section comprises applying a clustering-based routing protocol based on reinforcement learning to establish cluster groups and select cluster heads for each group. For clustering the nodes, it employs hop count, and to identify the CHs, it uses residual energy [23]. Despite their large-scale, IoT-enabled WSN’s consume a lot of energy. Consequently, they suffer greatly in terms of latency, throughput, and energy consumption [24]. As a part of this routing protocol, deep reinforcement learning algorithms are utilized to enable relay node selection. To estimate the best hop node and count, you only need to collect the necessary field parameters. Routing protocol selection, which is used to determine the next hop, employs Q-learning. Q-learning considers factors such as the distance between nodes and their energy levels [25]. However, there is a significant loss in packet delivery. An exception is the DRL-based routing protocol, in which the hop node selection is excellent, without any packet loss or energy consumption difficulties [26]. The control OH message is reduced because it does not consider communication delays when selecting a hop node. This research focuses on ensuring that the selected node for clustering has high data transmission potential, a low communication delay, and maximum network longevity. In this regard, a Fuzzy Logic Nodes Distributed Clustering for Energy-Efficient Fault Tolerance (F-NDC-EEFT) is proposed. The proposed method reduces the delay, saves energy, and extends its lifetime.

Mobile devices, such as smartphones, rely on 3 G and 4 G networks to send data and video, which adds to the congestion. Also, other issues like poor quality of service (QoS) are created by this extensive usage of smart phones [27]. The protocol uses a specific routing method to send data packets from one location to another. Routing rules are part of the protocol specifications. Layers are used in wireless communications to carry out protocols and transfer data [28]. In mobile wireless communication, a data transmission protocol is launched by the transport layer. Congestion control is a key part of the Transport Layer Protocol’s resource allocation strategy. The entire wireless network will come to a halt if 5 G mobile wireless communication is subjected to congestion control. To tackle this issue, routing protocols are developed. WSN-aided IoT and machine learning algorithms moderate and examine complex routing and energy management decisions. Algorithms based on learning produce the most efficient routes [29]. A deep learning approach for dynamic cluster-based routing in WSN-assisted IoT that considers QoS was proposed in [30]. Because WSN-assisted IoT is open and resource constrained, security and energy efficiency were the issues faced. To handle this issue, a hybrid WSN-IoT network was developed using the SecDL approach, and it is proposed in this paper. Mobile sink technology and Bi-Centric Hexagons were used to increase efficiency [31]. Cluster data aggregation was handled by enabling two-way data elimination and reduction. A high level of security was provided for clustered data thanks to cryptography. The cipher-text was converted to a mobile sink to ensure high quality of service. A route selection protocol called crossover based Fitted Deep Neural Network (Co-FITDNN) was proposed to provide the most accurate route detection possible. The current work focuses on ensuring user security because IoT users have access to sensory data [32]. An efficient clustering and routing method was introduced in [33]. Clustering was made more efficient by the introduction of Brainstorm Optimization with Levy Distribution (BSO-LD). After that, a Water Wave Optimization approach based on Hill Climbing (WWO-HC) was presented.

Selecting CH with an efficient algorithm and crucial parameters was the technique used by [34]. Performance was enhanced by routing through the CH of one’s choice. For the CH routing and selection process, Genetic Algorithm based Particle Swarm Optimization (GA-PSO) was described. Particle Swarm Optimization (PSO) helped in determining the optimal sink mobility route. IDF-FIT (IDAF-FIT) clustering algorithm was introduced in [35]. Using the if-then-then rule, the CH was grouped together. To select the best possible route, the present study used the ASLPP-RR algorithm [36]. The authors have summarized the benefits and limitations of the existing methods using a table format. This is shown in Table 1.

Table 1
Compares Reinforcement Learning-based routing protocols for fault tolerance based IoT-enabled WSNs

Ref Metrics Focus Benefits Limitations

[12] Packet delivery ratio, energy conservation • To support full duplex mode for connecting two nodes • To reduce overhead issues of the IoT enabled WSNs • Withstand in multi-sink architecture • Faster in route node selection process • Reasonable data rate is achieved even node failure is occurred • Node failure occurs frequently • Identification of sink node is difficult happened in the 3.6% error induced in the execution time •

[16] Latency, reliability, and packet delivery ratio • Concentrating on adaptive network topology change • Ensure synchronization between processing time and route shortest • • Frequently switching route node to recovery network lifetime • No central routing controls • Service processing time is optimistic • Adjustable in varying traffic condition • Average processing time is too high as compared to other standard protocol due to understand entire network • Routing table is not synchronous with agent learning rate

[18] End-to-end delay, residual energy, throughput • Keeps updating routing table on time with state value function • Load balancing and find optimum route node • Ensure data packet header contain correct route node details • Fulfils QoS requirement • Efficient to handle failure and node mobility • Supportive to available sink node • Avoid unnecessary node failure • Exploration scenario is not considered for justifying the improvement of the proposed routing protocol • Computational time is more complex • Few nodes are isolated from the network due to lack of updating routing table • Result shows that some half of the node to be static and others are dynamic which is unable to predict the node state value function

[24] Network lifetime, energy, data rate, failure error • Cooperative in manner • Adaptive dynamic changes made in the network • Minimum overhead exists • Reduce number of delays and enhance the reliability • QoS requirement is fulfilled by minimizing overhead and route node is proactive in nature • Network lifetime is improved significantly • Perfect synchronization exists • Spatial diversity and increases data transmission • Computational Complexity • Lack of reactive process over Q-value due to absent of sufficient memory • Relay selection is purely depending on lexicographic optimization •

[29] Energy consumption and network lifetime • Suitable for exploration scenario • Utilize satellite communication to enhance advance node tracking • Adaptive node sensing data to satellite processing service over finding best route node. • Battery energy properly utilized • Maximize the network lifetime • Support grid-based WSNs • • More control overheads are considered • System performance is arrested until the frequent updating of Q-value is needed • Delay is always accompanied and do nowhere to send data and processing time too high.

[33] Network lifetime and packet delivery ratio • Global optimization is achieved as superior to other DRL based routing protocol • Consider hop count before select to route node to reach sink node • State action is approximately depending on rewards and agent learning rate • Energy and workload are balanced • Without extra cost to get global optimization • High flexibility to topology change • Isolation of node and reduce energy as per packet transaction progressively • Node send the data to sink node if it is closer to existing data having node • Network isolation is significantly too high •

Ref	Metrics	Focus	Benefits	Limitations
[12]	Packet delivery ratio, energy conservation	• To support full duplex mode for connecting two nodes • To reduce overhead issues of the IoT enabled WSNs	• Withstand in multi-sink architecture • Faster in route node selection process • Reasonable data rate is achieved even node failure is occurred	• Node failure occurs frequently • Identification of sink node is difficult happened in the 3.6% error induced in the execution time •
[16]	Latency, reliability, and packet delivery ratio	• Concentrating on adaptive network topology change • Ensure synchronization between processing time and route shortest •	• Frequently switching route node to recovery network lifetime • No central routing controls • Service processing time is optimistic • Adjustable in varying traffic condition	• Average processing time is too high as compared to other standard protocol due to understand entire network • Routing table is not synchronous with agent learning rate
[18]	End-to-end delay, residual energy, throughput	• Keeps updating routing table on time with state value function • Load balancing and find optimum route node • Ensure data packet header contain correct route node details	• Fulfils QoS requirement • Efficient to handle failure and node mobility • Supportive to available sink node • Avoid unnecessary node failure	• Exploration scenario is not considered for justifying the improvement of the proposed routing protocol • Computational time is more complex • Few nodes are isolated from the network due to lack of updating routing table • Result shows that some half of the node to be static and others are dynamic which is unable to predict the node state value function
[24]	Network lifetime, energy, data rate, failure error	• Cooperative in manner • Adaptive dynamic changes made in the network • Minimum overhead exists	• Reduce number of delays and enhance the reliability • QoS requirement is fulfilled by minimizing overhead and route node is proactive in nature • Network lifetime is improved significantly • Perfect synchronization exists • Spatial diversity and increases data transmission	• Computational Complexity • Lack of reactive process over Q-value due to absent of sufficient memory • Relay selection is purely depending on lexicographic optimization •
[29]	Energy consumption and network lifetime	• Suitable for exploration scenario • Utilize satellite communication to enhance advance node tracking • Adaptive node sensing data to satellite processing service over finding best route node.	• Battery energy properly utilized • Maximize the network lifetime • Support grid-based WSNs •	• More control overheads are considered • System performance is arrested until the frequent updating of Q-value is needed • Delay is always accompanied and do nowhere to send data and processing time too high.
[33]	Network lifetime and packet delivery ratio	• Global optimization is achieved as superior to other DRL based routing protocol • Consider hop count before select to route node to reach sink node • State action is approximately depending on rewards and agent learning rate	• Energy and workload are balanced • Without extra cost to get global optimization • High flexibility to topology change	• Isolation of node and reduce energy as per packet transaction progressively • Node send the data to sink node if it is closer to existing data having node • Network isolation is significantly too high •

3 Materials and methods

3.1 System model

A network is created by placing smart sensors at random distances around the N × N (m²) region. In this case, two types of sensors are used-static sensors (inbuilt advanced energy-driven model) and dynamic sensors (restricted power accessibility). When compared to dynamic sensors, static sensors have a much lower occupancy. The energy distribution of each node for the length of n-bit data to reach the required distance D is estimated using the [25] approach. The transmitter energy is provided by the following equation (1) $E_{tx}^{i} = {\begin{matrix} (E_{e} + ɛ_{f} \times D^{2}) n_{i} & if & D < D_{0} \\ (E_{e} + ɛ_{mf} \times D^{4}) n_{i} & if & D ⩾ D_{0} \end{matrix}$ (1)

Where, E_e is the energy utilized by the circuit itself and D₀ is the threshold distance. An ɛ_f and ɛ_mf are the model of free space and multipath fading used, respectively.

At the receiver end, processing energy that can be recovered for a data length of n-bits can be calculated [23] as, $E_{rx}^{i} = n \times E_{e}$ (2) $D_{0} = \sqrt{ɛ_{f} / ɛ_{mf}}$ (3)

The energy correlation between two distinct nodes at the final data transfer can be calculated by using Equations (2). Likewise, the energy consumed by the cluster node E_cn in each cluster group is evaluated [21] by using the following equation (4) $E_{cn} = n \times E_{e} + n \times ɛ_{mf} \times d_{CH}^{2}$ (4)

Where, d_CH represents the distance between a cluster node and its cluster head (CH). The energy distribution of each cluster node is solely determined by its instant load data, resulting in varied energy consumption. As a result, the unequal clustering approach is adequate for computing two distinct energy constraints of the inter and the intra cluster groups [29], which are given as follows.

$\begin{matrix} E_{C H_{i}}^{int ra} = & c n_{i} \times E_{rx}^{i} + (c n_{i} + 1) \\ \times E_{ag} + E (C H_{i}, {CH}_{i}^{next}) \end{matrix}$ (5) $E_{C H_{i}}^{int er} = p (i) \times E_{rx}^{i} + p (i) \times E (C H_{i}, {CH}_{i}^{next})$ (6)

Where, CH_i and cn_i are the cluster head and cluster node of i^th cluster group, respectively. And p (i) represents the data packets received from the other cluster head, and ${CH}_{i}^{next}$ is the cluster head of the next hop.

3.2 Methodology

The proposed work has two phases: (i) identifying the eligible cluster heads and (ii) obtaining the shortest path for data routing. An eligible CH list from the base station is defined using three Fuzzy modules. To create an uneven cluster, nodes are placed in descending order of their energy, distance, density, mobility, and velocity. Therefore, all the nodes within a cluster have their energy consumption set to a moderate level. After the cluster heads have been identified, reinforcement learning is used to find the shortest path with the least number of intermediate hops. As a result, the network’s throughput improves due to faster communication between and within nodes. A detailed description of the two phases is given below.

3.2.1 CH eligibility estimation

Initially, the Base station (BS) sends a hello message to all the deployed sensor nodes. And whoever wants to join immediately sends a positive reply to this message. This message contains the node ID, location, energy, density, mobility, and velocity. These details are often utilized during CH selection, and thus, efficient CHs are elected for each round. Cluster formation is done with respect to a node distance. The radius varies as per the change is node distance. Then, immediately another CH is selected by computing their cluster radius using Equation (7), which includes node distance and iteration [24]. $C_{i + 1}^{r} = C_{i}^{r} + R_{0} / i + 1, \forall i \in (1, 2, . . . n)$ (7)

Where, $C_{i}^{r}$ represents the cluster radius with instance distance covers for every iteration rounds. That means that the Cluster Formation Request Message (CFRM) is circulated within the cluster’s radius. A node can join by sending a message with the node’s ID, location, energy, density, mobility, and velocity as a part of the JOIN message. It’s possible for a node to join a CH cluster radius even if it receives multiple CFRM messages from within and around the CHs in which it is located. So, clusters that are closer to BS have a smaller cluster radius than clusters that are farther away. As a result, the cluster sizes are unequal, ensuring that the load distribution is optimized as shown in Fig. 1.

Fig. 1

Unequal cluster formation based on node Fuzzy logic.

Therefore, the total energy consumption of a profile is reduced. CH selection begins by taking into consideration the input parameters such as node energy, node distance, and BS distance. Cluster formation in the proposed model is dependent on node Fuzzy modules (I, II and III). Additional input parameters are included in each module to fine-tune the CH’s eligibility estimation. So, for evaluating the first-level, Fuzzy Module-I employs mobility factor. The output duration of the mobility factor in conjunction with the velocity factor is used in Fuzzy Module-II. Fuzzy Module-III considers the average node distance to select the CHs eligible for cluster formation. The clustering group and its radius are adjusted based on the Doppler shift, average fade duration (AFD), and level crossing rate (LCR). Based on this, the JOIN message, energy, node position, and distance are updated. This will be repeated during the re-clustering phase until the entire packet transmission is completed. Figure 4 illustrates the overall clustering and re-clustering strategy based on the eligible CHs.

Fig. 4

Overall clustering and re-clustering strategy based on the eligible CHs.

Based on the fuzzy logic node, it is assumed that any node lying in between two cluster group radiuses. It is considered a free node if it is free to join any of the groups. Usually, the CH group with the maximum number of clusters is preferred by nodes because it has the highest energy density. The node Fuzzy module is used to determine whether a node is eligible to serve as the CH. For this, Fuzzy Module-I and Fuzzy Module-II take the node’s mobility and its velocity as their inputs, respectively.

3.2.2 Mobility factor (MF)

Figure 2 shows the estimation of the individual node’s mobility and relative velocity. When calculating each node’s mobility, the aggregate relative mobility metric is considered. Typically, it computes based on the strength of the receiver signal and the power level of the node. However, in ideal conditions, it estimates directly from the inverse square of the distance between the transmitting and receiving nodes. For, the JOIN message contains additional information about these. Computing the relative mobility metric $M_{y}^{rel}$ is done by using the given below equation [3]:

Fig. 2

Mobility factor estimation. (a) Node mobility, and (b) Relative mobility of node.

$M_{y}^{rel} = 10 log (\frac{P_{r_{x \to y}}^{new}}{P_{r_{x \to y}}^{old}})$ (8)

Where, $P_{r_{x \to y}}^{old}$ and $P_{r_{x \to y}}^{new}$ respectively are the old and new power levels at node Y with respect to X. The aggregate local mobility at any node M_y is done by calculating the variance (with respect to zero) of the entire set of mobility values, $M_{y}^{rel} (X_{i})$ , where X_i is the neighbor of Y.

$M_{y} = E [{(M_{y}^{rel})}^{2}]$ (9)

The Fuzzy Module-I uses this initial mobility factor to determine the time when the network gets initially established. It is also possible to use Equation (10) to figure out the mobility factor. $MF = \frac{N_{i + 1}^{C^{r}}}{N_{i}^{C^{r}}}$ (10)

Where, $N_{i + 1}^{C^{r}}$ is the identified moveable node which enters another cluster radius and $N_{i}^{C^{r}}$ is the non- moveable node within the cluster radius.

3.2.3 Velocity factors

For CH selection, each node’s velocity must be considered, as this affects the stability of the cluster as a whole. A node with a high velocity is more likely to leave the cluster than a node with a low velocity. Thereby, the CH selection process can be improved by incorporating this into the eligibility check Fuzzy module-II. Figure 3(a) depicts the CHs elected by BS to form unequal clustering group using the node Fuzzy logic algorithm.

Fig. 3

Operational flows of Fuzzy modules (I, II and III). (a) CH eligibility estimation, and (b) Re-clustering process.

The eligibility check of the available cluster nodes is carried out by the fuzzy modules. When the network is first set up, all the nodes in the system calculate this mobility factor. Fuzzy Module-I takes mobility as one of its inputs to compute the node’s CH eligibility. Other inputs of Fuzzy Module-I are the energy of the node, density of the nodes, their distance from the base station, and their mobility factor. Timer switches are then activated in each node based on the fuzzy module output. Next, the Fuzzy module 2 receives the timer length and velocity factor as inputs. Each node is assigned an eligibility score based on this. To be nominated as a candidate in the BS CH election, a node must maintain a high level of eligibility. Fuzzy module-III decision is made on the remaining CHs in the network based on the average distance between the eligible nodes that have reported the BS. The data routing protocol is still energy efficient because nodes that are not eligible cannot communicate with the BS.

3.2.4 Pseudo code of the proposed F-NDC-EEFT

The execution of each stage is precisely programmed and carried out in a linear fashion for evaluating the node’s aliveness, delivery rate, communication delay, and energy efficiency. The evaluation parameters are correctly defined in accordance with the simulation scenario. They are employed to sequentially process the three fuzzy clustering models.

Input: Deploy_Nodes over 100 x 100 area

Output: CH Selection

1. While (num_node_alive< =100)

2. For i = 1:100 nodes

3. Node(i)

4. Get Nodes energy (N_eg),

5. Calibrate Node Density (N_den)

6. Calibrate Distance from Base Station (N_dist)

7. Calibrate Mobility factor (M_f)

8. Load rules fuzzy_Module_1.fiz

9. Node_eligibity_Score_1(i)=fuzzy_Module_1{N_eg(i), N_den(i)_,N_dist(i)_,M_f(i)}

10. If Node_eligibity_Score_1(i) >eligibility_threshold

11. Select{i,} = Node(j)

12. End

13. End

14. For k=1:j

15. Node(j)

16. Calibrate Node_Timer = N_tim,

17. Calibrate Velocity_factor = V_fact,

18. Load rules fuzzy_Module_2.fiz

19. Node_eligibity_Score_2(j)=fuzzy_Module_2{N_tim(j), V_fact(j)}

20. If Node_eligibity_Score_2(j) >eligibility_threshold

21. Select{j,} = Node(h)

22. End

23. End

24. For g=1:k

25. Calibrate Doppler_Shift (D_shift)

26. Calibrate Level_Crossing_Rate (L_cross)

27. Calibrate Average_fade_Duration (A_fade)

28. Load rules fuzzy_Module_3.fiz

29. Node_eligibity_Score_3(j)=fuzzy_Module_3{D_shift(g),L_cross(g), A_fade(g)}

30. If Node_eligibity_Score_3(k) >eligibility_threshold

31. Select{g,} = Node(m)

32. End

33. End

34. Node(m) Selected CH Nodes

35. End

4 Simulation results

The network is simulated using a tool known as MATLAB R2019 and the results are compared with those of existing algorithms including Threshold Sensitive Stable Election Protocol (TSEP [18], stable Election Protocol (SEP) [29], Distributed Energy. Efficient Clustering (DEEC) [14] and Low-energy adaptive clustering hierarchy (LEACH) [9]. The network architecture is initially based on a random distribution across a 100×100 square meter region. 100 nodes with unique node IDs are designed to be distributed around the environment. The Mamdani fuzzy tool in MATLAB is used to construct Fuzzy modules for various needs. Between (0,0) are the cluster nodes and between (100,100) are the CH nodes. BS is placed at the center of the networks in the first simulation scenario, where 100 sensor nodes are placed in random locations throughout the monitoring area (60, 60).

BS’s coordinates are recorded by 200 sensor nodes randomly placed in the monitoring area, which is not connected to the network in the second simulated scenario (120, 60). One hundred sensor nodes with six CHs nodes, and another 120 sensor nodes with 12 gateways, are used to test the suggested model’s efficacy. Additionally, the values shown in Table 2 are utilized to simulate the proposed approach and the already existing state of the art algorithms.

Table 2
Simulation parameter

Parameters Value

Field Dimension 200 x 200

Number of Nodes deployed 100, 200

Cluster Head Selection Probability 0.1

Sensor Node Date Rate 256kbps

Packet Length 4000 bits

Initial Energy per Node 0.5 Joule

Transmission Energy (E_Tx) 50nJ

Reception Energy (E_Rx) 50nJ

Data Aggregation Energy (EDA) 5nJ

Pathloss 0.13pJ

Maximum Number of Simulation Rounds 5000

Simulation Result Parameters Alive Nodes Statistics, Statistics, Energy Consumption, Dead Nodes, Packet Delivery Ratio, End to End delay.

Parameters	Value
Field Dimension	200 x 200
Number of Nodes deployed	100, 200
Cluster Head Selection Probability	0.1
Sensor Node Date Rate	256kbps
Packet Length	4000 bits
Initial Energy per Node	0.5 Joule
Transmission Energy (E_Tx)	50nJ
Reception Energy (E_Rx)	50nJ
Data Aggregation Energy (EDA)	5nJ
Pathloss	0.13pJ
Maximum Number of Simulation Rounds	5000
Simulation Result Parameters	Alive Nodes Statistics, Statistics, Energy Consumption, Dead Nodes, Packet Delivery Ratio, End to End delay.

Figure 5 shows the node eligibility output of Fuzzy Module-I. The self-eligibility of a node to act as a cluster head is checked using the Fuzzy Module-I. This Fuzzy module determines a timer length based on 81 rules and on inputs including energy, distance from the BS, density of nodes, and mobility.

Fig. 5

Node eligibility evaluation using Fuzzy Module-I phase. (a) Fuzzy system for node eligibility: 4 inputs, 1 output, 81 rules, (b) Input membership functions and (c) Output membership function, and (4) Node Eligibility.

The Fuzzy Module-II, which is simply an extension of module 1, now includes the velocity factor as an enhancement parameter. Other than the outputs from Module-I, another input parameter is also included in this Fuzzy Module. Figure 6 depicts the node eligibility output from Fuzzy Module-II in detail.

Fig. 6

Node eligibility evaluation using Fuzzy Module-II phase. (a) Fuzzy system for node eligibility: 2 inputs, 1 output, 9 rules, (b) Input and Output membership functions and (c) Node Eligibility.

At the base station, the Fuzzy module-III determines the network’s cluster heads using the average distance between the eligible nodes reporting at the BS. This is clearly given in Fig. 7. Based on the Doppler shift, average fade duration (AFD), and level crossing rate, the clustering group and its radius are adjusted (LCR). The energy, node position, and distance are updated in the join message. This process is repeated until all the packets have been transmitted and re-clustering is complete. This is shown in Fig. 8.

Fig. 7

Node eligibility evaluation using Fuzzy Module-III phase. (a) Fuzzy system for node eligibility: 1 inputs, 1 output, 2 rules, (b) Input membership functions and (c) Output membership function, and (4) Cluster Head Eligibility.

Fig. 8

Re-clustering phase. (a) Fuzzy system for node eligibility: 3 inputs, 1 output, 27 rules, (b) Input and Output membership functions and (c) Demand Vector.

Figure 9 depicts the network’s active alive nodes and dead nodes after several rounds of deployment in the case of 100 and 200 sensor nodes. According to Fig. 9(a), the proposed algorithm increases the number of alive nodes by 88.3% when compared to LEACH (65.6%) [9], DEEC (75.4%) [14], TSEP (84.1%) [18], and SEP (86.3%) [29]. Figure 9(b) shows that the proposed algorithm reduces the number of dead nodes by 11.7% when compared to LEACH (34.4%) [9], DEEC (24.6%) [14], TSEP (15.9%) [18], and SEP (13.7%) [29].

Fig. 9

Simulated curve of active nodes and dead nodes with respect to number of rounds. Comparative analysis of the evaluation parameters with 100 and 200 nodes deployment (a)-(b), and (c)-(d), respectively.

Similarly, Fig. 9(c) conveys that the proposed algorithm increases the number of alive nodes by 87.9%. This is higher than that of LEACH (63.2%) [9], DEEC (74.4%) [14], TSEP (83.1%) [18], and SEP (85.3%) [29]. According to Fig. 9(d), the proposed algorithm reduces the number of dead nodes by 12.1%. This reduction is higher when compared to LEACH (36.8%) [9], DEEC (25.6%) [14], TSEP (16.9%) [18], and SEP (14.7%) [29], respectively.

Figure 10 depicts a comparison of the evaluation parameters with 100 and 200 node participation. Figures 10 (a) and (b) respectively show the end-to-end delay of the entire network analysis with 100 and 200 sensor nodes participating. It states that the proposed algorithm reduces the overall network end-to-end delay by 42.3% when compared to LEACH (52.25%) [9], DEEC (47.45%) [14], TSEP (44.1%) [18], and SEP (43.3%) [29].

Fig. 10

Comparative analysis of the evaluation parameters under 100, and 200 nodes participation. (a)-(b) End-to-end delay (ms), (c)-(d) Energy consumption (Joules), and (e)-(f) Cumulative sum of packet delivered to BS.

Figure 10 (c) and (d) show that the proposed algorithm reduces the overall node energy consumption by 78.2% when compared to LEACH (86.2%) [9], DEEC (84.4%) [14], TSEP (82.1%) [18], and SEP (81.3%) [29]. Following that, Figure 10 (e) and (f) conveys the network’s active packet delivery over multiple rounds when 100 or 200 sensor nodes are deployed in parallel. It states that the proposed algorithm improves overall node packet delivery ratio by 89.3% when compared to LEACH (68.2%) [9], DEEC (76.4%) [14], TSEP (85.1%) [18], and SEP (87.3%) [29]. These improvements are due to the correct identification of the eligible CHs by the proposed F-NDC-EEFT protocol, which calculates the optimal path for data routing based on multiple parameters such as residual energy, queuing time, traffic rate, and packet size, etc.

4.1 Discussion

Fault-tolerance in WSNs is the ability of a network to continue providing services even when some sensor nodes fail. Sensor nodes are deployed densely in the target area to increase coverage and connectivity due to limited battery life and the difficulty of recharging or replacing failure nodes. It is difficult to route data across many access points due to device mobility and shifting multiple locations. Accordingly, the connection failure and network unreliability may occur due to packet loss and incorrect node selection. In such cases, the total number of nodes alive or dead serves as an indicator that the network is experiencing some faulty nodes. The network manages to deliver packets to the base station even with dead faulty nodes. Figure 9 (a), (b), (c), and (d) depict this phenomenon graphically. The end-to-end delay parameter specifies the average time required to deliver a packet from the source node to the base station. Even after discovering some dead nodes, the network was able to deliver packets to the base station with a significant delay. Figure 10 (a) and 10 (b) depict the average end-to-end delay parameter as a function of the number of rounds. When compared to other existing works, the proposed model has the least number of delays. Figure 10 (c) and 10 (d) show that the proposed model uses lesser energy to deliver packets to the base station than existing models, even after the network begins to detect dead nodes. Figure 10 (e) and 10 (f) show that, when compared to the existing models, the proposed model has the highest cumulative sum of packets delivered to the base station even with faulty nodes in the network. The facts stated above demonstrate that the proposed model is highly fault-tolerant when compared to the popular existing methods. Time complexity is often reduced because of the withstand-ability, of proposed model. This is clearly shown in Fig. 11.

Fig. 11

Comparative analysis of the Time complexity.

Two different node counts (100 and 200) are considered for conducting a statistical analysis of the WSN’s fault tolerance parameters, such as end-to-end delay, packet delivery to BS, and energy consumption. In all the cases, the proposed method proves itself to be a major improvement over the state-of-the-art alternatives. The estimated end-to-end delay, packet delivery to BS, and power consumption outcomes are shown in Tables 3 –5.

Table 3

Statistical analysis of the proposed methods’ end-to-end delay (ms) in comparison to other existing methods

No. of nodes	100 nodes deployment					200 nodes deployment
	[9]	[14]	[18]	[29]	[Proposed]	[9]	[14]	[18]	[29]	[Proposed]
20	42	37	39	35	33	34	33	30	29	27
40	53	48	50	46	41	47	46	34	33	31
60	60	55	57	53	50	53	52	38	37	35
80	80	75	77	73	72	67	66	42	41	39
100	95	90	92	88	83	82	81	45	44	42
120	112	107	109	105	100	102	101	48	47	45
140	124	119	121	117	112	116	115	54	53	51
160	128	123	125	121	118	20	19	58	57	55
180	136	131	133	129	121	137	136	62	61	59
200	144	139	141	137	135	149	148	65	64	62

Table 4

Statistical analysis of the proposed methods’ energy consumption (milli joules) in comparison to other existing methods

No. of nodes	100 nodes deployment					200 nodes deployment
	[9]	[14]	[18]	[29]	[Proposed]	[9]	[14]	[18]	[29]	[Proposed]
5	203	326	95	72	20	179	309	78	55	3
10	400	418	192	120	82	376	401	175	103	65
15	600	526	215	174	112	576	509	198	157	95
20	800	588	256	192	134	776	571	239	175	117
25	920	600	320	234	189	896	583	303	217	172
30	1000	626	356	289	210	976	609	339	272	193
35	1284	645	384	324	235	1260	628	367	307	218
40	1357	654	402	246	245	1333	637	385	229	228
45	1400	662	413	368	262	1376	645	396	351	245

Table 5

Statistical analysis of the proposed methods’ cumulative sum of packet delivered to BS (x10⁶ and x10⁵, for 100 and 200 nodes, respectively) in comparison to other existing methods

No. of nodes	100 nodes deployment					200 nodes deployment
	[9]	[14]	[18]	[29]	[Proposed]	[9]	[14]	[18]	[29]	[Proposed]
1000	0.2	0.28	0.4	0.6	0.8	0.64	0.66	0.72	0.78	0.82
2000	0.82	0.9	1.02	1.22	1.42	2.38	2.4	2.46	2.52	2.56
3000	1.63	1.71	1.83	2.03	2.23	3.94	3.96	4.02	4.08	4.12
4000	1.86	1.94	2.06	2.26	2.46	4.12	6.4	6.46	6.52	6.56
5000	1.93	2.01	2.13	2.33	2.53	4.67	7.12	8.24	8.3	8.34
6000	1.65	2.09	2.21	2.41	2.61	4.92	7.34	10.24	10.3	10.34
7000	1.17	1.19	2.31	2.56	2.76	5.34	7.89	11.76	11.82	11.86
8000	1.23	1.25	2.37	2.62	2.82	5.78	8.23	12.74	12.8	12.84
9000	1.33	1.35	2.47	2.72	2.92	5.88	8.56	12.84	12.9	12.94
10000	1.39	1.41	2.53	2.78	2.98	6.12	8.9	12.9	12.96	13

5 Conclusion

In this paper, a new data routing protocol called as Fuzzy Logic Nodes Distributed Clustering for Energy-Efficient Fault Tolerance (F-NDC-EEFT) has been proposed. It uses three Fuzzy modules to obtain the list of eligible Cluster Heads, which forwards information to the Base Station, where the actual decision about the final cluster head selection is made. The proposed algorithm has achieved better performance over some of the existing algorithms in terms of withstand node aliveness, delivery rate, energy efficiency, and communication delay. The existing algorithms analyzed in this study are Threshold Sensitive Stable Election Protocol (TSEP), stable Election Protocol (SEP), Distributed Energy, Efficient Clustering (DEEC), and Low-energy adaptive clustering hierarchy (LEACH). In addition, the clustering approach is used to divide the instant data load into distinct node groups. All network platforms, including mobile and non-mobile devices, are supported by the algorithm. Simulation results show that it reduces connection failures, increases network throughput, extends the network’s lifespan, and reduces energy consumption. Thus, the technique is found to be promising for effective clustering of nodes in Internet of Things (IoT) enabled wireless sensor networks (WSNs). The authors suggest that future study in this field should aim at proposing a smart routing protocol that takes into consideration message overhead, time complexity, and maximum data sum rate for improving the performance of IoT-enabled WSNs.

References

Mohit Mittal and Celestine Iwendi , A survey on energy-aware wireless sensor routing protocols, EAI Endorsed Transactions on Energy 6 (2019), 062019–102019. doi: 10.4108/eai.11-6-2019.160835

Zagrouba

and Kardi

, Comparative study of energy efficient routing techniques in wireless sensor networks, Information 12 (2021), 42.

Gharamaleki

M.M.

and Babaie

, A new distributed fault detection method for wireless sensor networks, IEEE Syst. J. 14 (2020), 4883–4890.

Jaiswal

and Anand

, FAGWO-H: A hybrid method towards fault-tolerant cluster-based routing in wireless sensor network for IoT applications, J. Super Comput. 78 (2022), 11195–11227.

Keswani

, Mohapatra

A.G.

, Mohanty

, Khanna

, Rodrigues

J.J.

, Gupta

and de Albuquerque

V.H.C.

, Adapting weather conditions based IOT enabled smart irrigation technique in precision agriculture mechanisms, Neural Computing and Applications 31(1), (2019), 277–292.

Dowlatshahi

M.B.

, Rafsanjani

M.K.

and Gupta

B.B.

, An energy aware grouping memetic algorithm to schedule the sensing activity in WSNs-based IoT for smart cities, Appl. Soft Comput. 108 (2021), 107473.

Idrees

A.K.

and Al-Qurabat

A.K.M.

, Energy-efficient data transmission and aggregation protocol in periodic sensor networks based fog computing, J. Netw. Syst. Manag. 29 (2021), 1–24.

Zeb

, Islam

A.M.

, Zareei

, Al Mamoon

, Mansoor

, Baharun

, Katayama

and Komaki

, Clustering analysis in wireless sensor networks: The ambit of performance metrics and schemes taxonomy, International Journal of Distributed Sensor Networks 12(7) (2016), 4979142.

Gao

, Xiao

, Liu

and Wang

, Distributed soft fault detection for interval type-2 fuzzy-model-based stochastic systems with wireless sensor networks, IEEE Trans. Ind. Inform 15 (2019), 334–347.

10.

Abdulrab

, Hussin

F.A.

, Aziz

A.A.

, Awang

, Ismail

and Devan

P.A.M.

, Reliable fault tolerant-based multipath routing model for industrial wireless control systems, Appl. Sci. 12 (2022), 544.

11.

Menaria

V.K.

, Jain

S.C.

, Nagaraju

, Kumari

, Nayyar

and Hosain

, NLFFT: A novel fault tolerance model using artificial intelligence to improve performance in wireless sensor networks, IEEE Access 8 (2020), 149231–149254.

12.

Gharamaleki

M.M.

and Babaie

, A new distributed fault detection method for wireless sensor networks, IEEE Syst. J. 14 (2020), 4883–4890.

13.

Boukerche

, Pazzi

R.W.N.

and Araujo

R.B.

, Fault-tolerant wireless sensor network routing protocols for the supervision of context-aware physical environments, J. Parallel Distrib. Comput. 66 (2006), 586–599.

14.

Zagrouba

and Kardi

, Comparative study of energy efficient routing techniques in wireless sensor networks, Information 12 (2021), 42.

15.

Ruisong Han , Wei Yang , Yipeng Wang and Kaiming You , DCE; a distributed energy-efficient clustering protocol for wireless sensor network based on double phase clustering head election, Sensors 17 (2017), 998. doi: 10.3390/s17050998

16.

Moridi

, Haghparast

, Hosseinzadeh

and Jassbi

S.J.

, Fault management frameworks in wireless sensor networks: A survey, Comput. Commun. 155 (2020), 205–226.

17.

Effah

and Thiare

, Survey: faults, fault detection and fault tolerance techniques in wireless sensor networks, Int. J. Comput. Sci. Inf. Secure 16 (2018), 1–14.

18.

Ponnan

, Saravanan

A.K.

, Iwendi

, Ibeke

and Srivastava

, An artificial intelligence-based quorum system for the improvement of the lifespan of sensor networks, IEEE Sensors Journal 21(15) (2021), 17373–17385. doi: 10.1109/jsen.2021.3080217

19.

Chouikhi

, El Korbi

, Ghamri-Doudane

and Saidane

L.A.

, A survey on fault tolerance in small and large scale wireless sensor networks, Comput. Commun. 69 (2015), 22–37.

20.

Zhang

, Mehmood

, Shu

, Huo

, Zhang

and Mukherjee

, A survey on fault diagnosis in wireless sensor networks, IEEE Access 6 (2018), 11349–11364.

21.

Jiang

, A new method for node fault detection in wireless sensor networks, Sensors 9 (2009), 1282–1294.

22.

Agarwal

, Tapaswi

and Chanak

, Intelligent fault-tolerance data routing scheme for IoT-enabled WSNs, IEEE Int. Things J. (2022), 4662.

23.

Saeed

, Lee

Y.-D.

, Jan

and Koo

, CAFD: Context-aware fault diagnostic scheme towards sensor faults utilizing machine learning, Sensors 21 (2021), 617.

24.

Moridi

, Haghparast

, Hosseinzadeh

and Jassbi

S.J.

, Novel fault management framework using markov chain in wireless sensor networks: FMMC,. , Wirel. Pers. Commun 114 (2020), 583–608.

25.

Chen

, Li

and Huang

, A hypergrid based adaptive learning method for detecting data faults in wireless sensor networks, Inf. Sci. 553 (2021), 49–65.

26.

Cheraghlou

M.N.

, Khadem-Zadeh

and Haghparast

, Increasing lifetime and fault tolerance capability in wireless sensor networks by providing a novel management framework, Wirel. Pers. Commun. 92 (2017), 603–622.

27.

Panda

and Khilar

, Distributed byzantine fault detection technique in wireless sensor networks based on hypothesis testing, Comput. Electr. Eng. 48 (2015), 270–285.

28.

Srivastava

, Tripathi

, Singh

and Son

L.H.

, Energy efficient optimized rate based congestion control routing in wireless sensor network, J. Ambient Intell. Humaniz. Comput. 11 (2020), 1325–1338.

29.

Shankar

, Sivakumar

N.R.

, Sivaram

, Ambikapathy

, Nguyen

T.K.

and Dhasarathan

, Increasing fault tolerance ability and network lifetime with clustered pollination in wireless sensor networks, J. Ambient Intell. Humaniz. Comput. 12 (2021), 2285–2298.

30.

Feng

, Shi

, Huang

, Shen

, Yu

, Peng

and Wu

, Unknown hostile environment-oriented autonomous WSN deployment using a mobile robot, J. Netw. Comput. Appl. 182 (2021), 103053.

31.

Wang

, Zhang

, Xiong

N.N.

, Wan

, Shen

and Huang

, An effective edge-intelligent service placement technology for 5G-and-beyond industrial IoT, IEEE Trans. Ind. Inform. 18 (2022), 4148–4157.

32.

Tong

, Tian

, Lin

and Wang

, Fault tolerance mechanism combining static backup and dynamic timing monitoring for cluster heads, IEEE Access 8 (2020), 43277–43288.

33.

Thomas

, Orgun

, Hitchens

, Shankaran

, Mukhopadhyay

S.C.

and Ni

, A graph-based fault-tolerant approach to modeling QoS for IoT-based surveillance applications, IEEE Int. Things J. 8 (2021), 3587–3604.

34.

Al-Otaibi

, Al-Rasheed

, Mansour

R.F.

, Yang

, Joshi

G.P.

and Cho

, Hybridization of metaheuristic algorithm for dynamic cluster-based routing protocol in wireless sensor networks, IEEE Access 9 (2021), 83751–83761.

35.

Kaiwartya

, Abdullah

A.H.

, Cao

, Lloret

, Kumar

, Shah

R.R.

, Prasad

and Prakash

, Virtualization in wireless sensor networks: fault tolerant embedding for internet of things, IEEE Int. Things J. 5 (2018), 571–580.

36.

Babu

M.V.

, Alzubi

J.A.

, Sekaran

, Patan

, Ramachandran

and Gupta

, An improved IDAF-FIT clustering based ASLPP-RR routing with secure data aggregation in wireless sensor network, Mobile Netw. Appl. 26(3) (2021), 1059–1067.