Volume 10, Issue 5 p. 236-241
Research Article
Free Access

Fault autonomous model handling through integrated adaptive-filters for eliminating deployment faults in wireless sensor networks

Walaa M. Elsayed

Corresponding Author

Walaa M. Elsayed

Faculty of Computers and Information, Mansoura University, Egypt

Search for more papers by this author
Hazem M. El-Bakry

Hazem M. El-Bakry

Faculty of Computers and Information, Mansoura University, Egypt

Search for more papers by this author
Salah M. El-Sayed

Salah M. El-Sayed

Faculty of Computers and Information, Benha University, Egypt

Search for more papers by this author
First published: 01 October 2020
Citations: 1

Abstract

Wireless Sensor Networks (WSNs) are exposed to various data-deployment faults during the communication action. These faults may impact the behaviour of the sensors that degrade its performance and cuts its life. Therefore, we tend to implement the integration of two independent trends are self-awareness and self-adaptation capabilities along with two integrated adaptive filters, FIR and RLS. The proposed Autonomous Fault-Awareness and Adaptive (AFAA) model composed of three adaptive two-stage executed self-awareness approach to limit the impact of such faults during the propagation process. In this paper, we introduce the operational mechanism of AFAA that manages to identify the failure and aware of the lost signal values autonomously, then filter the perceptive-signals for eliminating the accompanied interference and gaining convergent values. It executed the incorporated autonomous model at the level of Cluster Head (CH) for independent fault-correction using an adaptive feedback model. Compared to the state-of-the-art methods, the proposed model achieved speed in fault diagnosis; also high-accuracy rate in the prediction of the lost signal values as much as 98.63%, thus improving the percentage of performance efficiency to 3:1 times along of duty cycle. Hence, it enhanced the overall network lifetime.

1 Background

In WSNs, sensor device/mote establishes itself per the network topology for broadcasting the monitored events of the surrounding environment to multiple hops. Wireless Sensor Networks (WSNs) usually prone to failures because of their deployment in harsh conditions and risky environments. WSNs exposed to various failures such as software, hardware, and communication failures; additionally the vast amount of data circulating between the network elements which are not kept due to restricted memory [1]. The appearance of hardware and communication faults affects the efficiency of the sensor's dissemination board and measurement unit. This makes WSN produces unfortunate performance and inaccurate measurements that degrade the credibility of the network [2]; thereby it forms an awfully serious challenge. WSN often suffer from poor performance during a broadcast platform that produces from hardware malfunctions, which enough to reduce network life [3]. If CHs are attacked by one of the mentioned failures happens, WSN performance will emerge poor and insufficient. Due to the sensor's limited resources and diverse deployment fields; the mission of fault detection, classification and elimination it in WSNs, has become a daunting function [4]. Therefore, many investigations were directed to study the recognition of malfunctions and fading method throughout deployment via the routing protocol. However, WSNs still require awareness and adaptive systems that offer the answer for overcoming the procedure complexities.

2 Autonomous fault awareness & adaptive model

The proposed technique aims to introduce an integral work composed of three adaptive two-stages are: (i) A self- detection method for the sudden faults that attack WSN, which allows the head of each cluster to autonomous determine a fault according to the proposed fault aware algorithm. (ii) A Self-Awareness method for such faults model that deals with an autonomous method for awarding the lost scales to the defected sensors based on prior-knowledge feeding in feedback adaptive loop, which spot the upcoming reading based on sensor time series. The previous two methodologies act as one-stage in the implementation. (iii) Adaptive-stage cross the integrated adaptive-filters, has studied interference cancellation for reducing the magnitude of the observed data from the above stage, also the transmitted signals; for acquiring the pure and converge signals. Besides, the AFAA mechanism relied on adjusting the weights consensus by a rate of error according to the reference signal during that phase. It used the product of the consensus weights as a feedback loop to feed a filtering stage, for significantly minimizing the error signal and refining transfer function cross iterations effectively. The paper shows new results related to the accuracy and speed of self-awareness model in fault detection and aware of it, also the efficiency of the self-adaptive model in cancelling interferences and convergence the registered measurements under other criteria.

The designate sensor network was designed according to the LEACH routing protocol employed elsewhere [5]. The numbers of clusters (C) were constructed during the setup phase. One representative node was assigned to be a cluster-head in each round, while the rest of the other nodes became ordinary members of the cluster. The cluster members broadcasted the beacon messages at every (K) time on a length of M through the mobility phase, even they declare alive. Generally, the lifespan of a sensor node in WSN involved many duty-cycles. The sensor may fall in one of the cycles under the impact of temporary or permanent failures. More clarifications are presented in the next context.

The proposed model is an adaptive op-application that contributed to cancel the interferences that transferred in the signals and extract the pure-signal. The model was built upon a prior knowledge coming the adaptive cancellation model. It was optimized for reporting the values of lost signals due to sudden WSN defects during deployment at a head level by the following sequence:

2.1 Failure detection

The delay-clock was taken as a basic principle enters in calculating time-delaying for input signal x[k]. The estimation of delayed signal urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0001 was computed at k time, as a linear combination of N previous readings{x[k − 1], x[k − 2], … , x[k − N]} using the following equation:
urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0002(1)
The synchronous threshold is assigned for real sensor readings x[k]. If the threshold is not exceeded, the node reports the reading to the head node. Otherwise, if the threshold is exceeded, the node simply discards the reading x(k) and avoids reporting it to the head node and that interprets the missing reporting for occurring fault. Then, the value u(k) will be assigned instead of its readings vector x(k) and calculated wi referred to the new weight at t+ 1 time on length (M), as given in (2).
urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0003(2)
It was worth mentioning that both w(n) and u(n) have the same filter length M.

This work introduced a simple development for the integration filter structure to cover the measurements of defected sensors by exploiting the delayed signal and using it as the desired reference signal, Fig. 1.

Details are in the caption following the image

Illustrates track of adaptive feed for the proposed fault self-awareness model

2.2 Self-awareness for missing standards & adaptation

The model studied various cases of faults-appearance and showed feedback as:
  • Delay-port transferred a default value estimated by − 1 (u(n)) to FIR filter when the value of input signal x(n) lost. The output of the filter is given by (3):

    urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0004(3)

  • In case of an iterative failure, the desired signal urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0005 is identified by uk(n). While the output of the filter (urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0006 is correlated with a prior-signals of u(n) and is defined by (4):

    urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0007(4)

  • In the temporary occurrence of an error, it is necessary to calculate the modification of weights resulting from the settings calculated at the time (k). These modifications are applied after the sampling period (k) and before their use at the time of sample (k + 1), for minimizing the weighted least squares error as follows:

    urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0008(5)

Then, substitute with the definition of the error signal e(n):
urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0009(6)
By compensating the desired signal urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0010 in the back feeding pattern, urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0011 becomes:
urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0012(7)
It can be said that the filter weights can be identified by the prediction error signal e(n)- the cost function C- and a gain vector as:
urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0013(8)
urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0014(9)
urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0015(10)
Hence,
urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0016(11)
The transfer function of the signal for feedback filters is often described and implemented by the difference equation that defines how the output signal is related to the input signal, as:
urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0017(12)
where,
  • P is the feedforward filter order
  • urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0018 are the feedforward filter coefficient s
  • Q is the feedback filter order
  • urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0019 are the feedback filter coefficient s
  • x(n) is the input signal
  • y(n) is the output signal.
  • A more brief form of the difference equation is:

    urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0020(13)

  • Which, when reshuffled, becomes:

    urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0021(14)

  • To find the transfer function of the filter, it first takes the Z-transform of each both side of the previous equation, where we use the time-shift property to obtain:

    urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0022(15)

  • From the above equation, we can define the transfer function urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0023 of the feedback filters as:

    urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0024(16)
    Fig. 2

    Details are in the caption following the image

    Equality the sides of difference equation for gaining a transfer function

Hence,
urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0025(17)

3 Simulation & performance evaluation

It used Network simulator Ns-2.34 for simulations. The implementation simulated 54 Mica2Dot sensors distributed in 9 clusters. The wireless system of network simulator equipped with the different parameters, i.e. Routing protocol LEACH, time simulation is 0.0003 s, initial energy is 2*AA batteries ∼3 Joule, receiving/transmission power is 3 dBm, data length is 256-bit packets, propagation area is 270 m × 270 m and the observed data type is temperatures, humidity and pressure values stamped with time throughout the 1000 rounds, see Table 1. In the experimental field, it forced sensors to transfer their readings to CH at regular intervals (k) to monitor two conditions:

Table 1. Simulation parameters
Parameter Value
wireless system International Resource Identifiers System (IRIS)
routing protocol LEACH
number of nodes 54
monitoring area 270 m × 270 m
topology type Clustering
simulation time urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0026
fault coverage time urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0027 0.0003 s.
data size 256 1024 bit
initial energy 2 − AA Batteries≅3 J
data type Temperature, Humidity, Pressure
faulty deployments 1. Transient (ft)
scenario 2. Intermittent (fi)
3. Byzantine (fb)
4. Potential (fp)
simulation scenario(s) Communication Environment
Fault Detection
Fault Aware & Adaptive
  • (i) The packet data transfer time has exceeded a predetermined threshold (>emax); emax is the timing threshold known as the delay-port;
  • (ii) CH did not receive the sensor reading during the simulation time which is a timing slot for diagnosing the fault.

CH infers to discover defect by the delay-clock that concludes the presence of malfunction by the sensor when signal arrival is delayed than the presumed time. Then, the delay-clock produces default value of (−1) as an input of the FIR filter, for gaining the desired signal DSP.FIR (d(n)), while retaining the prior knowledge of output signal of DSP.RLS urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0028 as a feed of the RLS filter.

In practical terms, the model exported the filter output by adjustment value, as it modified each weight with their the former value of the filter output and then estimated the inverse feed vector of inputs u(n). Furthermore, minimizing the error function and gaining optimum converges of filter coefficients, was achieved by two factors: (i) Set λ = 1 for specifying the length of the filter weights vector. This specifies how quickly the filter ‘forgets’ past sample information. (ii) Enter the initial filter weights, urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0029, as a scalar vector for the initial value of filter weights parameter. When you enter a scalar, the filter block uses the scalar value to create a vector of filter weights, according to (2). This vector has a length equal to the filter length and all of its values are equal to the scalar value. Furthermore, the filter block shall reset the weights whenever repeated fault detection at the delay-port. Likewise, when the passed output from delay port is −1, thus the Back feeding loop updates the filter weights. Otherwise, the filter weights remain at their current values, as expressed in Algorithm 1 (see Fig. 3).

Details are in the caption following the image

Algorithm of Fault Self-Awareness (FSA)

Further clarification, the proposed model tried to coverage domain of the lost reference signal -fault appearance- continuously and also minimize the sum of squares of the errors in the model through producing the value of input signal defined as the discrete-time signal, by applying a polynomial of degree p-1 [6]. This estimation might be suitable for gaining equally-spaced values of a slowly varying signal with low noise. In practical application, it located the error and covered it in the fault-diagnostic decisions table, as shown in Table 2. Whereas, the prediction procedure (P) carried out for the first-previous four values of the lost measurement, as:
urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0030(18)
where: urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0031 is the predicted signal value; x(n-j) are the previously observed values, with p < n; and aj is the predictor coefficient (aj = −1).
Table 2. Decision-making respect of fault diagnosis observed from node s25 readings throughout 15 h
x(n) U(n) urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0032 urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0033 urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0034 e(n)
R[1] 19.91 19.91 19.91 18.046 18. 019 0.027
R[2] 19.92 19.92 19.92 18.166 18.032 0.134
R[3] 19.90 19.90 19.90 17.891 18.148 −0.257
R[4] 19.90 19.90 19.90 18.447 17.890 0.557
R[5] −1 19.97 −20.438 18.510 −38.948
R[6] 2.67 2.67 2.67 2.624 1.757 0.867
R[7] 2.65 2.65 2.65 1.975 2.609 −0.634
R[8] 2.65 2.65 2.65 1.097 1.024 0.073
R[9] 2.68 2.68 2.68 2.484 2.108 0.376
R[10] 2.67 2.67 2.67 1.871 2.463 −0.592
R[11] 2.67 2.67 2.67 2.506 1.990 0.516
R[12] 2.66 2.66 2.66 1.062 2.435 −1.373
R[13] −1 2.618 −3.475 2.416 −5.891
R[14] −1 2.522 2.605 1.862 0.743
R[15] 19.90 19.90 19.90 18.80 19.587 −0787
Afterwards, it accomplished the adaption and filtering operation of the gained signal. We tried to test the efficiency of the model in eliminating interference under data reduction and failure occurrence, so the proposed adaptive schema attempted to adjust its coefficients by simulation conditions:
  • (i) If the carry-signal x(n) enters to the integrated filter block, both products of DSP.FIR (d(n)) and DSP.RLS (y(n)) will export to adjust the product filter-weight w(n). Then it directs the weighted signal to the RLS filter for re-adjusting the value of y(n) with the existing size of w(n), which will handle in the next step.

  • (ii) If the signal hides and does not enter the filter block, the FIR filter will process the signal coming from delay-port for gaining the desired signal d(n), and the RLS filter will retrieve the processed signal in the previous condition to get y(n) signal. The filters collaborate for gaining the value of the conscious filter-weight urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0035 mentioned in (2). The value of urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0036 transfers into FIR Filter, to extract a noise-free output signal representing the predicted filter outlet is urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0037, see Fig. 4.

    Details are in the caption following the image

    Adaptation of temperature samples picked from several Meteorological sectors

To test the ability of the proposed FSA algorithm in detecting the deployment faults and recognizing the lost standards, it injected numbers of sensors in various clusters within the designate the network throughout 200 iterations. It estimated fault detection accuracy rate through computing a ratio of the number of nodes that have been properly diagnosed by the simulator compared to the total number of the sensors have been injected per the cluster during each round. After that, it calculated the proportion of overall efficiency of the model throughout duty-cycle. Table 3 displays the accuracy rate of fault-aware throughout the mobility/communication mode for the studied samples.
Table 3. Standard database of FSA testing algorithm for the tolerated atmospheric samples
Sample w w2 w3 ʎ ʎ2 ʎ3 ft fi fb fp urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0038 urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0039 Coverage(EC)
temperature 0.3 0.6 0.9 0 0.7 1 98.902 99.913 99.756 99.537 0.0001 s. 0.00017 s. 99.5%
humidity 0.3 0.6 0.9 0 0.7 1 98.439 98.151 98.492 99.526 0.00015 s. 0.00018 s. 98.6%
pressure 0.3 0.6 0.9 0 0.7 1 97.784 98.700 97.280 97.678 0.0001 s. 0.0002 s. 97.8%
network coverage 98.63%
Table 3 showed the proportion of the network coverage through testing the FSA algorithm after 200 iterations. It worth mentioning that coverage was estimated by the number of packets which was covered by destination node (CH) in a specific interval of time. This performance metric demonstrates the total number of packets that have been effectively carried from source to CH, and it could enhance by increasing its speed. The Efficiency of Coverage (EC) was calculated as:
urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0040(19)
urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0041(20)
To test signal convergence and cancellation of noise ratio, firstly the error rate calculated by subtracting the expected signal from the desired signal, which picked from the final self-adaptive stage. Then, it identified the transferred bit error rate and the ratio of lost energy per noise power spectral density along of the observed measurements of the interconnected sensors during the operating rounds using criteria (BER) and (Eb/N0), as follows:
  • (i) Energy per bit of data (Eb):

    urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0042(21)

  • (ii) Noise power (N):

    urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0043(22)

  • (iii) Noise spectral density (N0):

    urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0044(23)

  • (iv) Ratio of Eb/N0:

    urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0045(24)

  • (v) Convert dBm to watt:

    urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0046(25)

  • (vi) Convert watt to joule:

    urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0047(26)
    where: B is the noise bandwidth in hertz (urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0048), fs is the bit rate per second (noise spectral rounding factor (fs = 0.15)), C is the carrier power of a signal (C = −25 dBm) and t (s) is the elapsed simulation time.

Through experiments, Eb/N0 reached to 118 W in temperature, 29.92 W in humidity and 129.51 W in pressure samples. The rate of lost energy per transferred bit data (E(J)) arrived up to 0.226 joules through noise cancellation phase. This means that sensors kept 92.68% of battery-energy. We generalized this at most of the picked samples throughout this phase. It is worth mentioning the values of BER decreased to −0.010319% of temperatures, 0.0198758% of humidity and 0.000294% from pressure picked samples in the same phase. This reveals that the proposed model removed the accompanied interferences and reduced energy and hence attained a better performance of the system. Besides, we evaluated the prediction accuracy of AFAA under the predicted measurement induced from a linear regression between the gained measurements are the desired signal (urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0049) and the filter output (urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0050) signal, and the adjusted weight along the expected input-signal (urn:x-wiley:20436386:media:wss2bf00269:wss2bf00269-math-0051) via a time index of every sample. It tested the accuracy of the predictive quality of the proposed model in fault tolerance and prediction using R-Squared (R2), as that reached the accuracy rate to ∼98% of injected samples throughout the duty-cycles. Also, we assessed the performance efficiency using Mean Absolute Percentage Error (MAPE), as its results pointed to rise as much as 98.7%. This confirmed that QoS of the designate network has upgraded.

4 Comparison with related work

For further evaluation of the two incorporated methodologies, The practicability diagnostic-results of the proposed AFAA model showed that accuracy of the self- aware and efficiency of filtration for the deployment-faults issued sensor via time-varying samples accomplished up to 98.63% (data reliability is 98.63%), compared to the results of FDD scheme [7] that achieved about 90.4% through the time-varying faults detection, and also results of the improved KAF [8] that reported 98% in the ability of fault diagnosis and filter its. Fig. 5 illustrated the localization process of a failure that degraded the node at a certain interval, and also displayed the throughputs scenarios of AFAA filter output, which showed the CH response in delivery the feedback throughput of AFAA for recovering the node failure circumstances observed in that interval.

Details are in the caption following the image

Outlined view of AFAA model coverage for the real-world sensor temperatures

5 Conclusion

The paper aimed to recover the lost readings for the deactivated sensors deployed in harsh environments. Therefore, the main goal of this paper is introducing an incorporated autonomous schema for deployment faults management by identifying the multi-communication failures in the sensor network and replacing the missing scales (fault) with its concurrent eligible candidate ones. We presented a performance optimization algorithm, which based on integrating capabilities of autonomous aware and adapting for allowing a sufficient coverage of missing measurements and eliminating interference accompanied the magnitude of data that needed to be transmitted to BS. The autonomous awareness paradigm in the integrated model managed to rely on the first-degree prior-knowledge of recovering the readings of defective sensors. Also, the autonomous adaptation paradigm succeeded to reduce the volume of the transferred data by depressing the rate of noise related to the expected perceptive-signal and decreasing the volume of the carry-signals throughout communication, hence enhanced the network performance and upgraded WSN lifetime.