Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
44 IEEE TRANSACTIONS ON COMPONENTS AND PACKAGING TECHNOLOGIES, VOL. 32, NO. 1, MARCH 2009 Optimization of Cylindrical Pin-Fin Heat Sinks Using Genetic Algorithms Sajjad Mohsin, Ayesha Maqbool, and Waqar A. Khan Abstract—In this paper, genetic algorithms (GAs) are applied for the optimization of pin-fin heat sinks. GAs are usually considered as a computational method to obtain optimal solution in a very large solution space. Entropy generation rate due to heat transfer and pressure drop across pin-fins is minimized by using GAs. Analytical/empirical correlations for heat transfer coefficients and friction factors are used in the optimization model, where the characteristic length is used as the diameter of the pin and reference velocity used in Reynolds number and pressure drop is based on the minimum free area available for the fluid flow. Both inline and staggered arrangements are studied and their relative performance is compared on the basis of equal overall volume of heat sinks. It is demonstrated that geometric parameters, material properties, and flow conditions can be simultaneously optimized using GA. Index Terms—Genetic algorithm (GA), heat sinks, inline, optimization, pin-fins, pressure drop, staggered, thermal resistance. NOMENCLATURE Total base area . Exposed area of base plate . Cross-section or contact area of a single pin Surface area of a single pin . Surface area of heat sink . Pin diameter . Friction factor. Equality and inequality constraints. Pin height . Resistance . Reynolds number . Entropy generation rate . Dimensionless diagonal pitch . Dimensionless streamwise pitch . Dimensionless spanwise pitch . Diagonal pitch . Pin spacing in streamwise direction . Pin spacing in spanwise direction . Absolute temperature . Thickness . Approach velocity . Maximum velocity in minimum flow area Design variables. Subscripts . Approach. Ambient. Base plate. Contact. Wall. Heat transfer coefficient . Number of imposed constraints. Constant defined in (18) and (19). Thermal conductivity Lagrangian function. Number of rows in streamwise direction. Dimensionless entropy generation rate . Number of rows in spanwise direction. Nusselt number based on pin diameter. Nusselt number based on heat sink length in flow direction. Total base heat flow rate . . Length of heat sink in flow direction Total number of pins in heat sink Number of design variables. . . Manuscript received November 13, 2007; revised May 21, 2008 and August 04, 2008. First published February 06, 2009; current version published February 27, 2009. This work was recommended for publication by Associate Editor R. Prasher upon evaluation of the reviewers comments. S. Mohsin is with the Department of Computer Science, COMSATS Institute of Information Technology, Islamabad 44000, Pakistan. A. Maqbool is with the Automatic Control and System Engineering Department, Sheffield University S1 3JD, U.K. W. A. Khan is with the Department of Engineering Sciences, PN Engineering College, National University of Sciences and Technology, PNS Jauhar, Karachi 75350, Pakistan. Digital Object Identifier 10.1109/TCAPT.2008.2004412 Fluid. Single fin. All fins with exposed base plate area. Heat sink. Bulk material. Greek Symbols Pressure drop across heat sink Slenderness ratio . Absolute viscosity of fluid . Kinematic viscosity of fluid Fluid density . . 1521-3331/$25.00 © 2009 IEEE . . MOHSIN et al.: OPTIMIZATION OF CYLINDRICAL PIN-FIN HEAT SINKS USING GENETIC ALGORITHMS 45 I. INTRODUCTION HE ENETIC algorithm (GA) right from its inception is being utilized as a global optimization technique. It was started by Friedburg [1], [2] with the mutation to induce learning in a small Fortran program. Holland [3] gave a boost to this field by utilizing bit strings representation for GA operators of selection/reproduction, crossover, and mutation. The boom in the evolutionary computing in general and GA in particular is due to the work of Koza [4]. Although GA has been used as an optimization technique for many problems, yet it was not considered for this particular problem. Khan et al. [5] presented a detailed survey of conventional optimization studies related to pin-fin heat sinks. Foli et al. [6] determined the optimal geometric parameters of the microchannels in micro heat exchangers. They used multi-objective genetic algorithms in combination with computational fluid dynamics (CFD) and concluded that the performance of micro heat exchangers depends on the operating conditions and aspect ratio of the microchannels. Jeevan et al. [7] determined the optimal dimensions for a stacked microchannel using the GAs under different flow constraints. In this paper, GA is employed to optimize all relevant design parameters for pin-fin heat sinks, including geometric parameters, material properties, and flow conditions simultaneously by subject to manufacminimizing entropy generation rate turing and design constraints. The potential and feasibility of applying GA as an optimization tool in electronic cooling will be demonstrated in this paper. T A. Genetic Algorithm The idea of applying the biological principle of natural evolution to artificial systems, introduced more than three decades ago, has seen impressive growth in the past few years. Usually grouped under the term evolutionary algorithms or evolutionary computation, we find the domains of genetic algorithms, evolution strategies, evolutionary programming, and genetic programming. Evolutionary algorithms are successfully applied to numerous problems from different domains, including optimization, automatic programming, machine learning, economics, operations research, ecology, population genetics, studies of evolution and learning, and social systems. Sajjad and Mohsin [8] have used GA in optimization of codebook generation through vector quantization in image compression and have shown that GA has given good results. Looking at the world around us, we see a staggering diversity of life. Millions of species, each with its own unique behavior patterns and characteristics, abound. Yet, all of these plants and creatures have evolved, and continue evolving, over millions of years. They have adapted themselves to a constantly shifting and changing environment in order to survive. Those weaker members of a species tend to die away, leaving the stronger and fitter to mate, create offspring, and ensure the continuing survival of the species. The laws of natural selection and Darwinian evolution dictate their lives, and it is upon these ideas that genetic algorithms are based. Goldberg [9] defines it as: Genetic algorithms are search algorithms based on the mechanics of natural selection and natural genetics. Bauer [10] gives a similar definition in his book: Genetic algorithms are software, procedures modeled after genetics and evolution. Fig. 1. Flowchart of Genetic Algorithm. B. Simple GA Algorithm Fig. 1 shows the flow diagram of simple GAs. Genetic algorithms are used to find optimum solutions to any problem. The problem is described by the string (1) we have to determine the way of evaluating the efficacy or fitness . A prototypical GA consists of the following steps: 1) generate initial population; 2) measure fitness; 3) select a mating; 4) pair members of the mating pool and perform crossover; 5) mutate each member of the crossover pool to obtain a new generation; 6) return to (2) until some stopping condition is satisfied. C. Encoding of a Chromosome As for any search and learning method, the way in which candidate solutions are encoded is a central, if not the central, factor in the success of a genetic algorithm. In recent years, there have been many experiments with different kinds of encoding. A chromosome should in some way contain information about solution that it represents. The most used way of encoding is a binary string. A chromosome then could look like this, as follows. Each chromosome is represented by a binary string. Each bit in the string can represent some characteristics of the solution. Another possibility is that the whole string can represent a number. Of course, there are many other ways of encoding. The 46 IEEE TRANSACTIONS ON COMPONENTS AND PACKAGING TECHNOLOGIES, VOL. 32, NO. 1, MARCH 2009 TABLE I BINARY STRING ENCODING a very high probability. Crossover can be one-point or multipoint (Fig. 3). The one-point crossover involves cutting the chromosomes of the parents at a randomly chosen common point and exchanging the right-hand-side sub-chromosomes. 4) Mutation: After a crossover is performed, mutation takes place. Mutation is a random change in the genetic material of a single individual (Fig. 4), and it is applied to prevent the solution to get into local minima. Mutation is applied to genes by changing them with a low probability. In case of binary encoding, we can switch a few randomly chosen bits from 1 to 0 or vice versa. Mutation can be then illustrated as in Fig. 4. E. Difference Between GA and Traditional Methods Fig. 2. Roulette-wheel selection operator. encoding depends mainly on the problem. For example, one can encode directly integer or real numbers (Table I). D. Operators of GA Typical operators for all types of genetic algorithms are: selection, crossover, and mutation. These operators are used consecutively at each step. 1) Selection: The selection operator selects chromosomes according to their fitness function values. In this procedure, the well-fitted individuals have more chances to be selected. The most common two selection operators are roulette-wheel selection and tournament selection. Elite approach is sometimes used in tournament and roulette wheel selections results. When creating a new population by crossover and mutation, we have a big chance, that we will loose the best chromosome. Elitism is the name of the method that first copies the best chromosome (or few best chromosomes) to the new population. The rest of the population is constructed in ways described above. Elitism can rapidly increase the performance of GAs, because it prevents a loss of the best-found solution. 2) Roulette-Wheel Selection: The idea behind the roulette-wheel selection technique is that each individual is given a chance to become a parent in proportion to its fitness. The chances of selecting a parent can be seen as spinning a roulette wheel with the size of the slot for each parent being proportional to its fitness. Obviously, those with the largest fitness (slot sizes) have more chance of being chosen. Thus, it is possible for one member to dominate all the others and get selected a high proportion of the time. Roulette-wheel selection can be implemented as follows. 1) Sum the fitness of all the population members. Call this (total fitness). 2) Generate a random number , between 0 and . 3) Return the first population member whose fitness added to the preceding population members is greater than or equal to . A roulette-wheel proportional to the fitness function can be demonstrated in Fig. 2. 3) Crossover: Crossover is a random exchange of genetic material between two parents to produce a unique offspring with The following list is a very quick look at the essential differences between GAs and other forms of optimization. (Goldberg, [9]). 1) Genetic algorithms a coded form of the function values (parameter set), rather than with the actual values themselves. So, for example, if we want to find the minimum of the , the GA would not deal function directly with or values, but with strings that encode these values. For this case, strings representing the binary values should be used. 2) Genetic algorithms use a set, or population, of points to conduct a search, not just a single point on the problem space. This gives GAs the power to search noisy spaces littered with local optimum points. Instead of relying on a single point to search through the space, the GAs looks at many different areas of the problem space at once, and uses all of this information to guide it. 3) Genetic algorithms use only payoff information to guide themselves through the problem space. Many search techniques need a variety of information to guide themselves. The only information a GA needs is some measure of fitness about a point in the space. Once the GA knows the current measure of “goodness” about a point, it can use this to continue searching for the optimum. 4) GAs are probabilistic in nature, not deterministic. This is a direct result of the randomization techniques used by GAs. However, recently some deterministic methods are also searched and worked on to get better results. 5) GAs are inherently parallel. Here lies one of the most powerful features of genetic algorithms. GAs, by their nature, are very parallel, dealing with a large number of points (strings) simultaneously. II. ANALYSIS The entropy generation rate associated with heat transfer and frictional effects serve as a direct measure of the ability to transfer heat to the surrounding cooling medium. A model that establishes a relationship between entropy generation rate and heat sink design parameters can be optimized using GAs in such a manner that all relevant design conditions combine to produce the best possible heat sink for the given constraints. Following Khan et al. [5], the entropy generation rate for a fluid flowing across a heat sink (Fig. 5) can be written as (2) MOHSIN et al.: OPTIMIZATION OF CYLINDRICAL PIN-FIN HEAT SINKS USING GENETIC ALGORITHMS 47 Fig. 3. Single point crossover. where (6) (7) (8) with (9) (10) Khan [5] has developed the following analytical correlation for the dimensionless heat transfer coefficient for a cylindrical fin in a fin array: Fig. 4. Mutation at a given point. (11) This expression shows that the entropy generation rate depends and the pressure drop across on the thermal resistance the heat sink, provided that the heat load , mass flow rate , and ambient conditions are specified. Assuming the same area of heat source and base plate, the thermal resistance of the heat sink can be written as where is a constant which depends upon the longitudinal and transverse pitch ratios, arrangement of the fins, and thermal boundary conditions. For an isothermal boundary condition, it is given by (12) (3) for inline arrangement, and where is the bulk material resistance, given by (4) (13) and is the overall resistance of the fins and the exposed base plate, which can be written as for staggered arrangement. The heat transfer coefficient for the base plate can be determined by considering a finite plate. Khan [5] has developed the following analytical correlation for the dimensionless heat transfer coefficient for a finite plate (5) (14) 48 IEEE TRANSACTIONS ON COMPONENTS AND PACKAGING TECHNOLOGIES, VOL. 32, NO. 1, MARCH 2009 TABLE II DIMENSIONS USED FOR OPTIMIZATION OF PIN-FIN HEAT SINKS Fig. 5. Geometry of an inline pin-fin heat sink. where is the length of the base plate in the streamwise direction. The mass flow rate through the fins is given by (15) The pressure drop associated with flow across the pin fins is given by (16) where the friction factor depends on the Reynolds number and the array geometry, and can be written as (17) for inline arrangement, and (18) for staggered arrangement, and is a correction factor depending upon the flow geometry and arrangement of the pins. It is given by (19) for inline arrangement, and (20) for staggered arrangement. All the correlations for friction and correction factors are derived from graphs given by ukauskas [11]. The velocity , in (15), represents the maximum average velocity seen by the array as flow accelerates between pins, and is given by (21) where is the dimensionless diagonal pitch. The simplified expression for the dimensionless entropy generation rate can be written as (22) where is a fixed dimensionless duty parameter that accounts for the importance of fluid friction irreversibility relative to heat transfer irreversibility. The duty parameter is fixed as soon as the fluid, fin material, and the base heat transfer rate are specified. The greater the base heat transfer rate, the smaller will be the fluid friction irreversibility. The constants and in (22) are given by (23) (24) (25) MOHSIN et al.: OPTIMIZATION OF CYLINDRICAL PIN-FIN HEAT SINKS USING GENETIC ALGORITHMS III. OPTIMIZATION PROCEDURE The problem considered in this study is to minimize the entropy generation rate using GA, given by (2) or (22), for the optimal performance of cylindrical pin-fin heat sinks. If represent the entropy generation rate that is to be minimized subject to equality constraints and inequality constraints , then the complete mathematical formulation of the optimization problem may be written in the following form: minimize (26) subject to the equality constraints (27) and inequality constraints (28) where and are the imposed equality and inequality constraints and are given by In (26), denotes the vector of the design variables . In this paper, we focus to find optimized values through GA with binary coding having roulette-wheel selection with elitism, single point crossover, and bitwise mutation. For optimization purpose, we chose three design variables, the approach i.e., , the pin height , diameter of pin, and velocity. The other related dimensions and physical properties are given in Table II. The variable is of 10 bit where its range 5.00 mm. Variable of 8 bits are dedicated is 0.7 mm for ranging from 0.8 mm 10.00 mm; here, we have to keep in mind that, in order to select the most optimized value, we have to consider not only the least entropy generation but also the practically feasible solution. Thus, the range for the is kept under than three times of , but care is taken that it must not increase the maximum limit of 10 mm. And the last variable is of 11 bits having range from 1 m/s to maximum 8 m/s is due to acoustic problems. These three variables are then concatenated to form a gene of 29 bits Size of Gene . In this paper, the initial parent population of size 21 is randomly generated. For the selection the roulette-wheel method with elitism is used. Fitness is the dimensionless entropy generation rate . In roulette-wheel selection, each individual is given a chance to become a parent in proportion to its fitness, those with larger fitness have more chance of being selected. In this paper, we have selected only those parents that results in real entropy generation. The discarded parent is replaced by new randomly generated gene that does not yield a complex value. This selection procedure is repeated until 21 offsprings had 49 been generated. Then, the individual showing minimum (i.e., the fitness function matrix), in the population of parents , in the is exchanged with the individual showing maximum population of offsprings. This is elitist preserving strategy. For single-point crossover, two individuals were chosen from the population of the offsprings. It is a random exchange of genetic material between two parents to produce a offspring with a random probability of 0.6. It involves cutting of the chromosome at randomly chosen point and exchanging the sub-chromosomes among two randomly selected genes. This crossover process is carried out until the new population of gene is generated. For mutation, each bit of chromosome is altered with the probability of 0.5. After mutation the fitness, i.e., the entropy generation is computed for all offsprings. This process of selection crosses over, and mutation is repeated for 2000 generations. The values of probability of crossover, mutation, and population size are all obtained empirically. IV. RESULTS AND DISCUSSION This paper shows the implementation of GAs to optimize the performance of a pin-fin heat sink under different operating conditions. GAs are global search techniques based on the mechanism of natural selection and genetics. Unlike the Newton Raphson method, used by Khan et al. [5], GAs do not require detailed mathematical derivations for unknown parameters; instead, the same model of entropy generation minimization can be used for optimization irrespective of the number of unknown parameters. Moreover, GAs do not require an accurate initial guess of the variable. The computational cost does increase whenever GAs are applied instead of any traditional method. This cost is dependent on the population size and the total number of iterations. In brief, GAs are best suited for the optimization of pin-fin heat sinks involving several variables, and they produce a high-quality solution. For optimization, the 25.4 objective is set to select the best heat sink to fit the 25.4 mm footprint but not to exceed a maximum overall height of 12 mm (Table II). The maximum height restriction is selected to represent a typical board pitch found in communications systems. It is also assumed that a total heat dissipation of 10 W is uniformly applied over the entire base plate which has a uniform thickness of 2 mm. The ambient temperature is fixed at 27 C, and the problem is solved for the low and high thermal conductivities 25 and 237 (enhanced plastics composites and aluminum). Parametric variations include the pin diameter approach velocity, , and the pin height . With reference to this, each arrangement for different thermal conductivities is examined for optimal heat sink geometry that leads to overall optimized performance where both heat transfer and viscous effects are considered. Minimum entropy generation rate for different number of generations corresponding to different number of pins for an inline arrangement is shown in Fig. 6. It is clear from figure that as the number of generations increases from 20 to 600, the entropy generation rate decreases with the increasing number of generations and number of pins as well. With the increase in the number of pins, the pressure drop increases, but at the same time thermal resistance decreases. As a result, the entropy generation rate decreases. The number of generations were increased up to 2000 as shown in Fig. 7 for an inline arrangement using low 50 IEEE TRANSACTIONS ON COMPONENTS AND PACKAGING TECHNOLOGIES, VOL. 32, NO. 1, MARCH 2009 Fig. 6. Entropy generation rate versus number of generations. Fig. 7. Entropy generation rate versus number of generations for inline arrangement. and high thermal conductivities. It is clear from Fig. 7 that the entropy generation rate decreases with the increase in thermal conductivity. It is observed that, in each case, it starts from a random minimum and reaches an optimum minimum solution using its basic parameters, selection, crossover, and mutation. It is also observed that the GA has evolved itself and have reached a stable condition as the generation progress. This shows that the 2000 generations is sufficient for this experiment. The results of optimization for low and high thermal conductivities are summarized in Tables III and IV. The same experiment was repeated for different number of generations corresponding to different pin diameters, pin heights, and approach velocities for an inline arrangement, and the results are plotted in Figs. 8–11. MOHSIN et al.: OPTIMIZATION OF CYLINDRICAL PIN-FIN HEAT SINKS USING GENETIC ALGORITHMS 51 TABLE III RESULTS FOR OPTIMIZATION OF THREE PARAMETERS FOR INLINE ARRANGEMENT Fig. 8. Dimensionless entropy generation rate versus pin diameter for different approach velocities. Fig. 9. Dimensionless entropy generation rate versus pin diameter for different number of pin-fins. For each approach velocity, optimum diameter exists that decreases with the increase in approach velocity. The minimum dimensionless entropy generation rate increases with the increase in approach velocity. The effect of number of pins on dimensionless entropy generation rate and optimum diameter for fixed pin height and approach velocity are shown in Fig. 9. It is obvious that optimum pin diameter and minimum dimensionless entropy generation rate decrease with the increase in number of pins. This is due to the fact that for fixed pin height and approach velocity, the thermal resistance decreases with the increase in surface area. The variation of dimensionless entropy generation rate with number of pins for different approach velocities is shown in Fig. 10. The pin diameter and pin height are kept constant in this case. The optimum number of pins exist for each approach velocity and decreases with the increase in approach velocity. The minimum dimensionless entropy generation rate increases with the increase in approach velocity due to higher pressure drop in the heat sink. V. SUMMARY AND CONCLUSION Fig. 10. Dimensionless entropy generation rate versus number of pins for different approach velocities. Fig. 8 shows the variation in dimensionless entropy generation rate versus pin diameter for different approach velocities. The number of pins and pin height are kept constant in this case. GAs are successfully employed for the optimization of pin-fin heat sinks. GAs are usually considered as a computational method to obtain optimal solution in a very large solution space. Entropy generation rate due to heat transfer and pressure drop across pin-fins is minimized by using GAs. The effects of pin diameter, approach velocity, and pin density for low and high thermal conductivities are examined with respect to their role in influencing optimum design conditions and the overall performance of the heat sink. The following is observed. 1) GA has evolved itself and has reached a stable condition as the generation progresses. 52 IEEE TRANSACTIONS ON COMPONENTS AND PACKAGING TECHNOLOGIES, VOL. 32, NO. 1, MARCH 2009 TABLE IV RESULTS FOR OPTIMIZATION OF THREE PARAMETERS FOR STAGGERED ARRANGEMENT [6] K. Foli, T. Okabe, M. Olhofer, Y. Jin, and B. Sendhoff, “Optimization of micro heat exchanger: CFD, analytical approach and multi-objective evolutionary algorithms,” Int. J. Heat and Mass Transfer, vol. 49, no. 5–6, pp. 1090–1099, 2006. [7] K. Jeevan, G. A. Quadir, K. N. Seetharamu, I. A. Azid, and Z. A. Zainal, “Optimization of thermal resistance of stacked micro-channel using genetic algorithms,” Int. J. Numer. Methods Heat Fluid Flow, vol. 15, no. 1, pp. 27–42, 2005. [8] S. Sajjad and S. Mohsin, “Comparative study on VQ with simple GA and ordain GA,” in Proc. 9th WSEAS Int. Conf. Autom. Control, Modeling, Simulation (AMCOS 07), Istanbul, Turkey, May 27–29, 2007, pp. 203–207. [9] D. E. Goldberg, Genetic Algorithms in Search, Optimization and Machine Learning. New York: Addison-Wesley, 1989. [10] R. J. Bauer, Genetic Algorithms and Investment Strategies. New York: Wiley, 1994. [11] A. Zukauskas, “Heat transfer from tubes in crossflow,” Adv. Heat Transfer, vol. 8, pp. 93–160, 1972. Ayesha M. Sheikh received the M.S. degree in computer science from COMSATS Institute of Information Technology, Abbottabad, Pakistan, in 2005. She is currently pursuing the Ph.D. degree in the Automatic Control and System Engineering Department, Sheffield University, Sheffield, U.K. Her research interests include intelligent agents, unmanned autonomous vehicles (UAVs), cooperative planning, and stochastic modeling. 2) Optimum dimensionless entropy generation rate increases with the increase in approach velocity and decreases with the increase in number of pins and pin diameter. 3) Staggered arrangement gives the highest thermal performance for the same material and the pin density. 4) Both inline and staggered arrangements show the same behavior for thermal resistance, pressure drop, and dimensionless entropy generation rate. REFERENCES [1] R. M. Friedberg, “A Learning Machine,” IBM J., pt. Part I, pp. 2–3, 1958. [2] R. M. Friedberg, “A learning machine,” IBM J., pt. Part II, pp. 282–287, 1959. [3] J. Holland, Adaptation in Natural and Artificial Systems. Ann Arbor, MI: Univ. of Michigan Press, 1975. [4] J. Koza, Genetic Programming: On the Programming of Computers by Means of Natural Selection. Cambridge, MA: MIT Press, 1992. [5] W. A. Khan, “Modeling of Fluid Flow and Heat Transfer for Optimization of Pin-Fin Heat Sinks,” Ph.D. dissertation, Univ. of Waterloo, Waterloo, ON, Canada, 2004. Waqar Ahmed Khan is an Associate Professor of mechanical engineering at the National University of Sciences and Technology, Karachi, Pakistan. He has developed several unique analytical models for the fluid flow and heat transfer across single cylinders (circular/elliptical), tube banks, and pin-fin heat sinks to Newtonian and non-Newtonian fluids. His research interests include modeling of forced convection heat transfer from complex geometries, microchannel heat sinks, thermal system optimization using entropy generation minimization, forced and mixed convection, and conjugate heat transfer in air and liquid cooled applications. He has more than 28 publications in refereed journals and international conferences. Dr. Khan is a member of ASME, AIAA and the Pakistan engineering council. Sajjad Mohsin received the M.Sc. computer science degree from Quaid-i-Azam University, Islamabad, Pakistan, in 1987, the M.E. degree in computer science and systems engineering from Muroran Institute of Technology, Japan, in 2002, and Ph.D. degree from the Muroran Institute of Technology, Muroran, Japan, in 2005. He is an Associate Professor in Department of Computer Science, COMSATS Institute of Information Technology, Islamabad, Pakistan. His research interests include neural networks, genetic algorithms, fuzzy logic, soft computing, evolutionary computing, pattern recognition, vector quantization. He has more than 15 refereed journal and conference publications to his credit. He is a member the editorial board of three international journals.