Khan et al. / Front Inform Technol Electron Eng in press 1 Frontiers of Information Technology & Electronic Engineering www.zju.edu.cn/jzus; engineering.cae.cn; www.springerlink.com ISSN 2095-9184 (print); ISSN 2095-9230 (online) E-mail: jzus@zju.edu.cn Detection of faulty sensor in array using symmetrical structure and cultural algorithm hybridized with differential evolution* d Shafqat Ullah KHAN†1, Ijaz Mansoor QURESHI2, Fawad ZAMAN3, Wasim KHAN4 (1School of Engineering & Applied Sciences, ISRA University, Islamabad 44000, Pakistan) (2Electrical Department, Air University, Islamabad 44000, Pakistan) 3 ( Electrical Department, COMSAT Institute of information Technology Attock 44000, Pakistan) ite (4Electronic Engineering Department, IIU, H-10, Islamabad 44000, Pakistan) † E-mail: shafqatphy@yahoo.com Received Sept. 26, 2015; Revision accepted Jan. 9, 2016; Crosschecked ed Abstract: In this work, the detection of fully as well as partially defective sensors in a linear array composed of N sensors is addressed. In the first step, the symmetrical structure of a linear array is proposed, while in the second step a hybrid technique based on cultural algorithm with differential evolution is developed. The symmetrical structure has two advantages: instead of finding all damaged patterns, only (N – 1)/2 patterns are needed, and we are required to scan the region from 0 to 90 instead of 0 to 180, which obviously reduces the computational complexity. Monte Carlo simulations are carried out to validate the performance of proposed scheme, which is compared with already available methods in terms of computational time and mean square error. Key words: Cultural algorithm, Differential evolution, Linear symmetrical sensor array http://dx.doi.org/10.1631/FITEE.1500315 CLC number: 1 Introduction un In the literature, several techniques are available that address the issue of defective sensors in an array antenna. Rodríguez et al. (2000, 2009) diagnosed the defective sensor using the genetic algorithm (GA), where the fitness function compares the measured radiation pattern with the given configuration of failed/unfailed sensors. Patnaik et al. (2007) used a neural network (NN) approach to detect a maximum of three defective sensors in a small array of 16 sensors, while Bucci et al. (2000) considered the ambiguity of the result in the continuous and discrete on–off cases. Xu et al. (2007) used the support vector * Project supported by the Higher Education Commission of Pakistan. ORCID: Shafqat Ullah KHAN, http://orcid.org/0000-00031969-1289 © Zhejiang University and Springer-Verlag Berlin Heidelberg 2016 machine (SVM) to diagnose the defective sensors in a small array of four sensors, but this technique is not applicable to large arrays because of the possible number of combination boosts. Moreover, the available techniques are computationally expensive, as they not only require storing the patterns of all defective sensors in the array but also need to scan the entire region from 0 to 180. Today, biologically inspired techniques, especially differential evolution (DE) and cultural algorithm (CA), are considered efficient and reliable tools for optimization in the field of science and engineering ( Zaman et al., 2012). CA and DE are optimization techniques that include domain knowledge obtained during the evolutionary process. In many optimization problems, both CA and DE have successfully overcome the shortcomings of conventional optimization techniques due to their suppleness and effectiveness (Reynolds and 2 Khan et al. / Front Inform Technol Electron Eng in press where wn is the nonuniform weight of the nth sensor, d is the spacing between the adjacent sensors, and is the angle from broadside. Similarly, k 2 / is the wave number with as the wavelength, and kd cos s is the progressive phase shift, where s is the steering angle for the main beam. For a nonhealthy setup, as shown in Fig. 1, AF can be written as M AF i wn exp jn kd cos i (2) d n M nm If either wm or w m is damaged, i.e., by putting wm or w m equal to zero, the array factor for the mth ite Peng, 2005; Reynolds and Chung, 1996; Becerra, 2006; Jin X and Reynolds 1999). DE and CA are stochastic-based search algorithms, in which the function parameters are programmed as floating-point variables. They are simple in structure, converge fast, and are robust against noise. Fonollosa et al. (2013) developed a more reliable e-nose and robust system in which machine learning based on multiple kernels was generated to overcome sensor failures. The outcome confirmed that multi-kernel models are more robust to sensor failures when the sub-kernels are trained with small sets of sensors. In this paper, the detection of fully as well as partially defective sensors in a linear array composed of N sensors is addressed. In the first step, the symmetrical structure of a linear array is proposed, while in the second step a hybrid technique based on CA with DE is developed. In this hybrid process, the results achieved through CA are given to DE for further tuning. Mean-squared error (MSE) is used as an objective evaluation function that defines the error between the responses of desired and estimated ones. The symmetrical structure has two advantages: instead of finding all damaged patterns, only patterns are needed, and we are required to scan the region from 0 to 90 instead of 0 to 180, which obviously reduces the computational complexity. The proposed method outperforms the conventional method of Choudhury et al. (2013) in terms of computational time and MSE. Monte Carlo simulations are carried out to validate the performance of proposed scheme, which is compared with the already available methods in terms of computational time and MSE. damaged sensor in a noisy environment is given by M AF i wn exp jn kd cos i i (3) n M nm where i is the additive zero mean complex Gaussian noise with the variance at the nth sensor. AFm (i ) un ed is the pattern when wm or w m is fully faulty, as depicted in Fig. 1. Mathematically, the measurement noise (in dB) can be expressed as K 2 AF i i 1 SNR K 2 i 1 i We assume that the sensor w4 fails in the array. The method of locating a faulty element in a linear 2 Problem formulation Consider a uniform linear array (ULA) of N 2M 1 sensors along the x-axis with respect to the origin. The far-field array factor (AF) for a healthy setup of equally spaced sensors of nonuniform amplitude and progressive phase excitation can be given as (Wolf, 1937) M AF i wn exp jn kd cos i n M (4) (1) Y w M w2 d w1 w0 w1 × w2 X wM Fig. 1 Nonuniform amplitude array of 2 M 1 sensors with w2 sensor defective 3 Khan et al. / Front Inform Technol Electron Eng in press array starts with the measurement of several samples of the faulty pattern. The damaged array pattern for the w4 sensor is shown in Fig. 2, where one can whether the sensor is fully or partially defective. The value of the threshold has been found on the basis of the MSE. If the lowest error is < 0.5 = Eth , the weight clearly observe that the pattern is symmetrical about 90 . wm or w m is fully faulty. If the lowest error is >0.5 CADE technique to find the weights for a partial detective sensor. The fitness function is given by 0 Original w4 element faulty -10 = Eth , then the weight is partially faulty. We use the 90 G PF i PCADE i -30 -40 (6) i 1 -50 -60 where -70 PF i is the desired response, and PCADE i is the value of the pattern obtained by -80 0 Fig. 2 20 40 60 80 100 Angle (Degrees) 120 140 160 180 N 1 Original Chebyshev array and the w4 sensor damage pattern 3 Proposed methodology using the CADE technique. The proposed method starts with tabulating both the faulty patterns ite -90 -100 2 d Far-Field Pattern(dB) -20 FF i F 21 and the single defective pattern Pm i evaluated in the range 0 to 90 . Then the computation of Cm in Equation (5) is performed by finding the faulty sensor that minimizes Cm faulty patterns and the single defective pattern when wm or w m is fully faulty. ed In this section, we develop the proposed method based on faulty patterns that are symmetric about 90 . Due to the symmetric structure, the patterns are the same if either wm or w m is dam- 0 -20 90 Cm PF i Pm i i 1 2 (5) Equation (5) compares the faulty patterns with a given configuration of a fully faulty sensor, and the minimum result will give us the location of a faulty sensor. In Equation (5), PF (i ) is the faulty pattern, Pm (i ) is the pattern when wm or w m is fully faulty, and m is given by 1 m ( N 1) / 2 . Then, on the basis of another fitness function, we will decide Far-Field Pattern(dB) un patterns. The other advantage is that we have only to scan the damage pattern from 0 to 90 , as the damaged pattern is symmetrical about 90 . The location of the faulty sensor can be found as Original w-4 element failure -10 aged. The failure of w4 or w4 gives the same pattern as shown in Figs. 2 and 3. For the detection of the faulty sensor, we have to tabulate half the number of faulty patterns; that is, we require ( N 1) / 2 faulty between the -30 -40 -50 -60 -70 -80 -90 -100 0 20 40 60 80 100 Angle (Degrees) 120 140 160 180 Fig. 3 Original Chebyshev array and w4 sensor failure pattern The proposed method is computationally efficient, as we require half the number of samples 90 i i 0 ; also we have to tabulate ( N 1) / 2 faulty patterns for the detection of the faulty sensors. The method of finding the faulty sensor in an array starts with the measurement of faulty patterns. If the defective sensor fails but radiates some power, i.e., the defective array pattern can be obtained from Equa- 4 Khan et al. / Front Inform Technol Electron Eng in press tion (2), the defective weight is a fraction of the original one. A. Differential evolution f hq , ge if rand () CR or h hrand rkq , ge q , g e d h otherwise where 0.5 CR 1 and h rand (8) is chosen randomly. DE was developed by Storn and Price and is used to solve real-valued optimization problems (Storn and Price, 1995). DE is a stochastic-based search algorithm that has a simple structure, fast convergence, and robustness against noise. It has shown good results for multimodel and non-differential fitness functions. DE is based on a mutation operator, which adds an amount obtained by the difference of two randomly chosen individuals of the present population. Thus, it has got tremendous applications in the field of science and engineering ( Das and Konar, 2006; Rogalsky, 1999). The basic steps are given in the form of a pseudo-code as follows: Step 1. Initialization: First we initialize Q c) Selection operation: The next-generation chromosome is generated by the following expression: chromosomes randomly, where the length of each chromosome is 1 P . The P genes in each chromosome represent the weights of the array antenna, given as B: Cultural algorithm d rq, ge if error r q, ge error d q, ge CRor j jrand d q, ge 1 q, g d e otherwise where error r q , ge and error d q , ge are defined in ite matrix S . Step 3. Stopping criterion: The stopping criterion is based on the following condition: iteration has reached. CA was developed by Reynolds (1994) to model the evolution of the cultural component of an evolutionary computational system over time. The main idea behind CA is to clearly attain problem solving knowledge from the growing population, and apply this knowledge to guide the search space (Robert and Chung, 1996). CA utilizes culture as a van for storing information that is available to the entire population over many generations. The parameter setting is given in Table 1, while the flow diagram for CA is shown in Fig. 4. CA consists of three components: a populated space, a belief space, and a communication protocol. The first one contains the population to be evolved and the mechanisms for its estimate. The population space consists of a set of possible solutions to the problem. In this work, the population space is DE. The second one is a belief space that represents the bias that has been acquired ed S If error d q , ge 1 , the maximum number of lb wi , k ub i 1, 2, 3,...Q k 1, 2, 3,...P un wi , k R w1,1 w1,2 , ...., w1, p w2,1 w2,2 ,...., w2, p w w ,...., w Q, p Q,1 Q ,2 Step 2. Updating: All the chromosomes form 1 to Q of the current generation are updated. Choose d hq, ge from the matrix in which ge and h represent the particular generation and length of chromosome, respectively. Our main task is to find the chromosome of the next generation, i.e., eq , ge 1 , by using mutation, crossover, and selection operations. a) Mutation: To perform the mutation process, we select randomly three different chromosomes from matrix S : 1 c1 , c2 , c3 Q with c1 c2 c3 j f q , ge d c1 , ge F (d c2 , ge d c3 , ge ) 0.5 F 1 (7) b) Crossover: Crossover is performed on the expression Table 1 Parameters used in the CADE algorithm Parameter Setting Population size 500 No of generation 500 Value of F 0.5 Value of CR 0.5 < CR < 1 5 Khan et al. / Front Inform Technol Electron Eng in press knowledge. CA is used as an optimization algorithm, its belief space is stored and updated with the population space during each generation, and the knowledge in the belief space is used to guide the search space toward the required solution. The differential evolution variation operators are influenced in the following way: n ,g n ,g y i , g Qi F w 2 w 3 (9) where Qi is the ith component of the individual stored d by the population during its problem-solving process. The belief space is the information depository in which the individuals can store their experiences for other individuals to learn them ultimately. These two spaces are connected to each other through the communication protocol composed of two functions, i.e., the acceptance function and the influence function. The acceptance function is used to accept the experience of the best individuals from the population space and store them in the belief space, and then the knowledge in the belief space can be updated through the update function. The influence function can guide the search space. In the present work, the belief space is divided into two knowledge components, i.e., situational knowledge and normative in the situational knowledge. The normative knowledge includes a scaling factor dsi to influence the mutation operator adopted ite in DE. The following expression shows the influence of the normative knowledge of the variation operators: un ed zij, g wn1 , g F wn2 , g wn3 , g if wn1 , g l i n1 , g n3 , g n2 , g n1 , g if w ui w F w w (10) n1 , g ui li n3 , g n2 , g w F w w otherwise * ds i where li and ui are the lower and upper bounds for the n ,g ith decision variable, respectively. w 1 represents the jth component of the ith individual that is selected from the gth generation by the acceptance function i, i 1, 2,...naccepted . naccepted is the number of best individuals at the gth generation. The values dsi are updated with the difference w n2 , g w n3 , g found of the variation operators of the prior generation. Normative knowledge leads individuals to jump into the good range if they are not already there. The normative knowledge is updated as follows: Let us consider xa1 , xa 2 , xa 3 ,..., xan to be the accepted individuaccepted als in the current generation. Here wmini , wmaxi NO YES a , a , a ,..., n 1 2 3 accepted belong to and the accepted individuals with minimum and maximum values for the parameter i. Fig. 4 Generic flow diagram of the cultural algorithm 6 Khan et al. / Front Inform Technol Electron Eng in press i and i (11) (12) otherwise wi ,min i li l i If li if wi,max ui or f wmax Ui if wi ,mini li or f wmini Li otherwise ui are updated, the values of Li and U i will be updated in the same way. dsi are updated with the greatest difference of wi , r1 wi , r 2 found during the variation operators at the previous generation. faulty, and for partial fault we will use the CADE technique to find the weights using Equation (6). Simulation results for completeness as well as for partial fault have been checked, which show the validity of the proposed method. Figure 5 shows the behavior of the diagnostic error for different values of the signal-to-noise ratio (SNR). We examined different values of SNR versus the MSE; the smaller the value of SNR, the greater the MSE. As one can clearly see from Fig. 5, the MSE decreases as the value of SNR increases. ite 4 Simulation results and discussion faulty. If Cm > 0.5 = Eth , then weight is partially d w i,maxi ui ui Fig. 5 Mean-squared error versus signal-to-noise ratio un ed In this section we discuss several cases based on different numbers of defective sensors in array. Case 1: Consider that a Chebyshev linear array of 51 sensors with / 2 intersensor spacing is used as the test antenna. The array sensors are placed symmetrically along the x-axis and excited around the center of the array. An analytical technique is used to find the nonuniform weights for a –30 dB constant side lobes level (SLL) in the Chebyshev array. To diagnose the faulty sensor in the linear symmetrical array, the radiation patterns for the fully and partial faulty sensors were generated. The samples were taken from the patterns in the region from 0 to 90 . In the interval of 6, 15 samples were taken from each pattern to scan the region from 0 to 90 for the detection of fully and partially faulty sensors. It is clear from Figs. 2 and 3 that the failure of either w4 or w4 Case 2: In this case we consider that a sensor fails partially and radiates some power, i.e., its weight is not zero but a fraction of the original one. First we consider that w2 sensor is 50% damaged and this damage pattern is created by making the weight of the sensor to half of the original weight in the original Chebyshev array. Now the CADE technique is used to locate its position. This array factor is obtained by making the weight of the w2 sensor equal to half of gives the same pattern, which is symmetrical about 90 if either w4 or w4 fails. For the detection of the original weight in Equation (1) and is represented by PF i in the fitness function of Equation (6). the faulty sensor, we have to tabulate both the faulty Then CADE technique is used to minimize the fitness function, which in turn gives the weight of the defected sensor. MSE is used as a fitness function, which is given by Equation (6). To check the performance of the CADE technique, these weights are compared to the weights obtained by the Chebyshev method in which the weight of the sensor equals half of original weight. The weight obtained by CADE is given in Table 2, and the recovered pattern by CADE technique is shown in Fig. 6. patterns FF i F 1 and the single fault pattern 25 Pm i if either wm or w m is damaged. The cost function in Equation (5) is minimized for a given 90 sample i i 0 . The decision for the detection of faulty sensor will be made on the basis of the cost function. The minimization of cost functions in Equation (5) will give us the location of a faulty sensor. If Cm < 0.5 = Eth , the weight wm is fully 7 Khan et al. / Front Inform Technol Electron Eng in press 1 1 Original w2 50% faulty Corrected by CADE 0.9 Original w10 25% faulty Corrected by CADE 0.9 0.8 0.8 0.7 0.7 0.6 0.6 0.5 0.5 0.4 0.4 0.3 0.3 0.2 -20 -15 -10 -5 0 5 10 15 20 25 Fig. 6 Weight distribution of the original , w2 defected, and obtained by CADE 0.1 -25 -20 -15 -10 -5 0 5 10 15 20 25 Fig. 8 Weight distribution of the original, w10 defected, and corrected by CADE Now we suppose that the sensor is 50% faulty by making w5 weights half of the original value. The of the partial fault. The weights for the original Chebyshev array and the obtained weights for different cases by CADE technique are given in Table 2. Now we assume that the w sensor has 25% failed. ite CADE technique is used to locate its position. The array factor is obtained by making the weight of the w5 sensor equal to half of the original value in Equa- d 0.2 -25 10 The CADE algorithm is run to locate the partial fault of the sensor. The weights obtained by CADE for the detection of partial faults (25%) for sensor are given Table 2 and shown in Fig. 8. Table 2 CADE Sensor position w25 w24 w23 w22 w21 w20 w19 w18 w17 w16 w15 w14 w13 w12 w11 w10 w9 w8 w7 w6 w5 w4 w3 w2 w1 w0 Chebyshev and normalized weights obtained from un ed tion (1). Then the CADE technique is used to minimize the fitness function in Equation (6), which in turn gives the weight of the faulty array. The weights obtained by using the CADE technique for partial failure are given in Table 2. To check the validity, the weights obtained by CADE technique are compared to the weights obtained from the Chebyshev method of damage patterns for 50% fault. The comparison of the weights amplitude of the defective array with the weights obtained by CADE technique is given in Table 2. The weight distributions of original, partially faulty, and that obtained by CADE are depicted in Fig. 7. From this comparison, the partial fault can be clearly identified. Comparison of the obtained weights by CADE with that of the defected array shows the level 1 Original w5 50% faulty Corrected by CADE 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 -25 -20 -15 -10 -5 0 5 10 15 20 25 Fig. 7 Weight distribution of original, w5 defected, and obtained by CADE Chebyshev w2 sensor weights 50% failed 0.6426 0.6475 0.2200 0.2234 0.2554 0.2593 0.2928 0.2959 0.3321 0.3402 0.373 0.3754 0.4152 0.4197 0.4584 0.461 0.5023 0.5076 0.5465 0.5485 0.5907 0.5934 0.6344 0.6431 0.6773 0.6796 0.719 0.7203 0.7591 0.7612 0.7973 0.7995 0.833 0.8413 0.8661 0.8676 0.8961 0.8987 0.9228 0.9264 0.9459 0.9476 0.9651 0.9685 0.9802 0.9843 0.9912 0.4879 0.9978 0.9986 1.0000 0.9994 w5 sensor 50% failed 0.6437 0.2252 0.2584 0.2969 0.3451 0.3758 0.4247 0.4635 0.5085 0.5496 0.5952 0.6417 0.6815 0.7351 0.7652 0.7987 0.8418 0.8673 0.8984 0.9378 0.4857 0.9721 0.9873 0.9992 0.9983 0.9987 w10 sensor 25% failed 0.6453 0.2317 0.2643 0.2975 0.3429 0.3813 0.4341 0.4618 0.5163 0.5538 0.5916 0.6384 0.6794 0.7265 0.7631 0.1987 0.8491 0.8752 0.8987 0.9376 0.9518 0.9687 0.9885 0.9932 0.9986 0.9975 8 0.9995 0.9945 0.9835 0.9676 0.9513 0.9301 0.8978 0.8751 0.8421 0.7985 0.7584 0.7542 0.6833 0.6461 0.5932 0.5496 0.5159 0.4613 0.4271 0.3742 0.3364 0.2953 0.2571 0.2262 0.6432 0.9993 0.9928 0.9834 0.9676 0.9473 0.9362 0.8988 0.8676 0.8414 0.7993 0.7632 0.7531 0.6795 0.6384 0.5951 0.5485 0.5046 0.4597 0.4174 0.3819 0.3357 0.2961 0.2589 0.2317 0.6478 0.9996 0.9934 0.9827 0.9675 0.9486 0.9264 0.8975 0.8676 0.8451 0.7998 0.7685 0.7326 0.6891 0.6454 0.5937 0.5579 0.5163 0.4627 0.4256 0.3828 0.3482 0.2952 0.2617 0.2356 0.6513 proposed method. We assume a maximum of three defective sensors, which yields a total of 3 N! 2324 different patterns by the f 1 f !( N f )! conventional method but our proposed method requires 1162 (i.e., half the number of patterns of the conventional method) pattern, which requires fewer computations and less cost. To locate the faulty sensors in an array antenna, the weight of each sensor was considered as the optimizing parameter for the BFO and CADE algorithm. CADE will converge to the optimum solution. In an array diagnosis, 35 samples were taken in the range 0 to 180 in intervals of 5. We suppose that the 4th and 10th sensors are partially faulty and the 17th sensor is fully faulty. We run the BFO and the proposed method to diagnose the faulty sensors. Figure 11 shows the faulty pattern with the position of 35 samples. The fault diagnosed by the conventional method is shown in Fig. 12. Now the same process is repeated using the proposed method, which d 0.9978 0.9912 0.9802 0.9651 0.9459 0.9228 0.8961 0.8661 0.833 0.7973 0.7591 0.719 0.6773 0.6344 0.5907 0.5465 0.5023 0.4584 0.4152 0.373 0.3321 0.2928 0.2554 0.2200 0.6426 ite w1 w2 w3 w4 w5 w6 w7 w8 w9 w10 w11 w12 w13 w14 w15 w16 w17 w18 w19 w20 w21 w22 w23 w24 w25 Khan et al. / Front Inform Technol Electron Eng in press 0 Original array Faulty array -10 random samples to validate the performance of the Far-Field Pattern(dB) -20 -30 -40 -50 -60 -70 0 20 40 60 80 100 Angle (Degrees) 120 140 160 180 Fig. 9 Chebyshev array and the 4th, 10th (50%), and 17th (100%) sensor faulty pattern 0 Original array Faulty array -10 Far-Field Pattern(dB) un ed Case 3: In this section we consider a linear Chebyshev array of 24 sensors taken as the reference antenna to execute the method of fault diagnosis developed by Choudhury et al. (2013). The proposed method is compared with the conventional method. The fully and partial faulty patterns were generated by making their weights either equal to zero or some fraction of the original weights. We assume that the 4th, 10th (50%), and 17th (100%) sensor in the array have become faulty. The faulty pattern of the 4th, 10th (50%), and 17th (100%) sensor is shown in Fig. 9. Now we consider the symmetrical counterpart failure of 4th, 10th (50%), and 17th (100%), which is depicted in Fig. 10. It is clear from Figs. 9 and 10 that both failures give the same power pattern. Therefore we have to tabulate half of the faulty patterns. The second advantage of using a symmetrical linear array is that the faulty pattern is symmetrical about 90 , i.e., we need half the number of samples to scan the pattern. First we simulated the pattern with 1–3 elements faulty. We have used a set of 1162 patterns. In each case, the faulty pattern contains the M samples as input to check the diagnosis of fault. In this case, we take samples of M 18, M 35 and -20 -30 -40 -50 -60 -70 0 20 40 60 80 100 Angle (Degrees) 120 140 160 180 Fig. 10 Chebyshev array and the symmetrical counterpart of (4th, 10th (50%), and 17th (100%)) sensor faulty pattern 9 Khan et al. / Front Inform Technol Electron Eng in press Original weights Defected weights 1 Normalized weight excitation is shown in Figs. 13 and 14. By the proposed method, we diagnosed the fault by half the number of samples. Similarly, the same fault is repeated for 18 and some random samples for the conventional and proposed method. The results obtained by conventional and proposed method for 18 and random numbers of samples are shown in Figs. 15–18. By the proposed method even with fewer sample points, one can detect the faulty sensor accurately. Figure 19 shows the MSE plots for conventional and the proposed method. 0.8 0.6 0.4 0.2 5 10 15 20 Element Number Fig. 14 Performance of the proposed method with 19 sample points d 0 0 -10 -20 Far-Field Pattern(dB) -30 -30 ite Far-Field Pattern(dB) -10 -20 -40 -50 -60 -40 -50 -60 -70 0 20 40 60 80 100 Angle (Degrees) 120 140 160 180 Fig. 11 Defective array pattern with fault at 4th, 10th (50%), and 17th (100%) sensor with 35 sample points -70 0 20 40 60 80 100 Angle (Degrees) 120 140 160 180 Fig. 15 Defective array pattern with fault at 4th, 10th (50%), and 17th (100%) sensor with 10 sample points ed 1 Original weights Defected weights Normalized weight excitation Normalized weight excitation 1 0.8 0.6 0.4 0.2 5 10 15 0.8 0.6 0.4 0.2 20 un Element Number 5 0 -10 -10 -20 -20 -40 -50 -60 -70 15 20 Fig. 16 Performance of proposed method with 10 sample points 0 -30 10 Element Number Far-Field Pattern(dB) Far-Field Pattern(dB) Fig. 12 Performance of the conventional method (Choudhury et al. 2013) with 35 sample points -30 -40 -50 -60 0 20 40 60 80 100 Angle (Degrees) 120 140 160 180 Fig. 13 Defective pattern with fault at 4th, 10th (50%), and 17th (100%) sensor with 19 points -70 0 20 40 60 80 100 Angle (Degrees) 120 140 160 180 Fig. 17 Defective array pattern with fault at 4th, 10th (50%), and 17th (100%) sensor with random sample points 10 Khan et al. / Front Inform Technol Electron Eng in press 16000 Original weights Defected by BFO MSE by conventional method MSE by proposed method 14000 0.8 12000 Cost function Normalized weight excitation 1 0.6 0.4 10000 8000 6000 4000 0.2 2000 5 10 15 0 20 Element Number 500 1000 Number of iteration 1500 Fig. 19 Mean square error performance of the conventional (Choudhury et al., 2013) and the proposed method Table 3 Time comparison analysis for six configurations of one defective sensor Proposed method Fault place Conventional method 18 samples 56.72 58.53 59.65 60.87 54.47 52.89 35 samples 117.46 121.69 115.78 109.95 108.6 107.95 Average time 57.18 18 samples Average time 113.57 108.42 113.73 114.56 117.35 107.5 105.45 35 samples Average time 111.16 ed 1 2 5 10 20 23 Time (s) ite Time (s) 201.71 206.5 213.31 205.37 203.95 204.87 Average time 205.95 Table 4 Time comparison analysis for six configurations of two defective sensors Fault place Proposed method Conventional method Time (s) Time (s) 18 samples 67.7 66.4 72.3 74.8 76.2 78.9 35 samples Average time 72.71 139.7 141.4 153.6 155.1 161.9 165.2 18 samples Average time152 .81 un 5, 7 20, 23 18, 20 11, 21 7, 10 8, 12 123.5 121.7 137.1 139.3 143.6 144.1 35 samples Average time 134.88 254.3 237.4 238.75 241.5 231.94 235.5 Average time 240.48 Table 5 Time comparison analysis for six random configurations of three defective sensors Proposed method Conventional method Time (s) Time (s) Fault place 18 samples 5, 7, 12 1, 7, 12 1, 9, 11 13, 17, 24 15, 19, 22 14, 21, 24 81.7 85.3 79.3 82.6 85.5 77.3 Average time 81.95 35 samples 169.4 173.7 167.3 169.9 175.5 164.8 2000 d Fig. 18 Performance of the proposed method with random sample points 0 Average time 170.1 18 samples 144.1 157.3 142.9 153.5 151.3 145.6 Average time 149.11 35 samples 231.4 274.1 230.9 227.2 237.6 241.7 Average time 240.48 11 Khan et al. / Front Inform Technol Electron Eng in press Table 6 Time comparison for six random configurations of fully and partially faulty sensors Proposed method Conventional method Time (s) Time (s) 18 samples 85.7 84.4 82.8 Average time 84.3 18 samples 35 samples 182. 6 171. 6 141.5 241.2 152.9 251.7 175. 80.9 Average time 177.9 179. 86.7 181. 85.1 176. 147.7 Average time 148.2 143.8 249.1 Average time 250.01 252.4 151.3 246.8 152.3 258.9 ite 8, 12 (100%) 2(50%) 13, 18(50%) 24(100%) 14, 20 (100%) 15(50%) 1, 8 (100) 2(50%) 15, 23 (100%) 24(50%) 4, 10 (50%) 17 (100%) 35 samples d Fault place References Bucci, O. M., Capozzoli, A., & D'elia, G. (2000). Diagnosis of array faults from far-field amplitude-only data. Antennas and Propagation, IEEE Transactions on, 48(5), 647-652. Becerra, R. L., & Coello, C. A. C. (2006). Cultured differential evolution for constrained optimization. Computer Methods in Applied Mechanics and Engineering, 195(33), 4303-4322. Choudhury, B., Acharya, O. P., & Patnaik, A. (2013). Bacteria foraging optimization in antenna engineering: An application to array fault finding. International Journal of RF and Microwave Computer‐Aided Engineering, 23(2), 141-148. Das, S., & Konar, A. (2006). Two-dimensional IIR filter design with modern search heuristics: A comparative study. International Journal of Computational Intelligence and Applications, 6(03), 329-355. Fonollosa, J., Vergara, A., & Huerta, R. (2013). Algorithmic mitigation of sensor failure: Is sensor replacement really necessary?. Sensors and Actuators B: Chemical, 183, 211-221. Jin, X., & Reynolds, R. G. (1999). Using knowledge-based evolutionary computation to solve nonlinear constraint optimization problems: a cultural algorithm approach. In Evolutionary Computation, 1999. CEC 99. Proceedings of the 1999 Congress on (Vol. 3). IEEE. Khan, S. U., Qureshi, I. M., Zaman, F., & Naveed, A. (2013). Null placement and sidelobe suppression in failed array using symmetrical element failure technique and hybrid heuristic computation. Progress in Electromagnetics Research B, 52, 165-184. Khan, S. U., Qureshi, I. M., Naveed, A., Shoaib, B., & Basit, A. (2015). Detection of Defective Sensors in Phased Array Using Compressed Sensing and Hybrid Genetic Algorithm. Journal of Sensors, 501, 718914. ed For the few initial iterations, the value of MSE is high, but after some iterations it goes down. BFO and the proposed method were run to detect the location of faults for six random cases and find out the average time of faulty sensor for various scenarios. The results are given in Tables 3–6. From the simulation results, it is clear that the computation time increases as the number of faulty sensors increases. 5 Conclusions un We have proposed a computationally efficient technique of finding fully as well as a partially defective sensors in a linear array. Using the approach of a symmetrical linear array brings two advantages. The failure of wm or w m gives the same pattern, i.e., we require ( N 1) / 2 patterns instead of finding all damaged patterns and, second, we need to scan half of 90 the damage pattern i i 0 , as the pattern is symmetrical about 90 . The decision of fully or partially faulty sensor is made on the basis of the cost function. If the value of Cm is >0.5, the sensor is fully faulty and if the value of Cm is <0.5, then the sensor is partially faulty. For partial fault we used the CADE technique to locate the defective sensor. This method can be extended to planar arrays and L-type arrays. Khan et al. / Front Inform Technol Electron Eng in press Yeo, B. K., & Lu, Y. (1999). Array failure correction with a genetic algorithm. Antennas and Propagation, IEEE Transactions on, 47(5), 823-828. Zaman, F., Qureshi, I. M., Naveed, A., Khan, J. A., & Asif Zahoor, R. M. (2012). Amplitude and directional of arrival estimation: comparison between different techniques. Progress in Electromagnetics Research B, 39, 319-335. Zaman, F., Qureshi, I. M., Naveed, A., & Khan, Z. U. (2012). Joint estimation of amplitude, direction of arrival and range of near field sources using memetic computing. Progress in Electromagnetics Research C, 31, 199-213. un ed ite Khan, S. U., Qureshi, I. M., Zaman, F., Shoaib, B., Naveed, A., & Basit, A. (2014). Correction of Faulty Sensors in Phased Array Radars Using Symmetrical Sensor Failure Technique and Cultural Algorithm with Differential Evolution. The Scientific World Journal, 2014. Mailloux, R. J. (1996). Array failure correction with a digitally beamformed array. Antennas and Propagation, IEEE Transactions on, 44(12), 1543-1550. Oliveri, G., Donelli, M., & Massa, A. (2009). Linear array thinning exploiting almost difference sets. Antennas and Propagation, IEEE Transactions on, 57(12), 3800-3812. Oliveri, G., Manica, L., & Massa, A. (2010). ADS-based guidelines for thinned planar arrays. Antennas and Propagation, IEEE Transactions on, 58(6), 1935-1948. Patnaik, A., Choudhury, B., Pradhan, P., Mishra, R. K., & Christodoulou, C. (2007). An ANN application for fault finding in antenna arrays. Antennas and Propagation, IEEE Transactions on, 55(3), 775-777. Reynolds, R. G. (1994, February). An introduction to cultural algorithms. In Proceedings of the third annual conference on evolutionary programming (pp. 131-139). Singapore. Reynolds, R. G., & Chung, C. (1996, May). A self-adaptive approach to representation shifts in cultural algorithms. In Evolutionary Computation, 1996., Proceedings of IEEE International Conference on (pp. 94-99). IEEE. Reynolds, R. G., & Peng, B. (2005). Knowledge learning and social swarms in cultural systems. Journal of Mathematical Sociology, 29(2), 115-132. Reynolds, R. G., & Chung, C. (1996, November). The use of cultural algorithms to evolve multiagent cooperation. In Proc. Micro-Robot World Cup Soccer Tournament (pp. 53-56). Rodriguez, J. A., Ares, F., Palacios, H., & Vassal'lo, J. (2000). Finding defective elements in planar arrays using genetic algorithms. Progress In Electromagnetics Research, 29, 25-37. Rodríguez-González, J. A., Ares-Pena, F., Fernández-Delgado, M., Iglesias, R., & Barro, S. (2009). Rapid method for finding faulty elements in antenna arrays using far field pattern samples. Antennas and Propagation, IEEE Transactions on, 57(6), 1679-1683. Rogalsky, T., Kocabiyik, S., & Derksen, R. W. (2000). Differential evolution in aerodynamic optimization. Canadian Aeronautics and Space Journal, 46(4), 183-190. Storn, R., & Price, K. (1995). Differential evolution-a simple and efficient adaptive scheme for global optimization over continuous spaces (Vol. 3). Berkeley: ICSI. Wolff, I. (1937). Determination of the radiating system which will produce a specified directional characteristic. Radio Engineers, Proceedings of the Institute of, 25(5), 630-643. Xu, N. (2007). Detecting failure of antenna array elements using machine learning optimization. In 2007 IEEE antennas and propagation society international symposium (pp. 5753-5756). d 12