Effective Techniques Applying Via Genetic Algorithm Approach
Proceedings of the World Congress on Engineering 2008 Vol IWCE 2008, July 2 - 4, 2008, London, U.K.Effective Techniques Applying via GeneticAlgorithm ApproachStefania GALLOVA, Member, IAENGAbstract— The relevant idea is to consider the search for thebest genetic algorithm approach as an optimisation problem anduse another genetic algorithm approach to solve it. Amethodology calculation is based on the idea of measuring theincrease of fitness and fitness quality eva.luating created by twomethodologies with secondary genetic algorithm approachusing. Performance results for finding the best genetic algorithmfor the complex real problem of optimal machinery equipmentoperation and predictive maintenance are presented. Weillustrate two interesting solutions approaches within geneticalgorithm environment.Index Terms— genetic algorithm, fuzzy set, fitness, crossover,mutation.I. INTRODUCTIONThe aim of solved system is to develop an effective system,which is able first of all to provide intelligent advice forsolving complex technical (diagnostic) problems efficientlyand precisely with its relevant domain expertise or skills.Performance of the approach was verified on a real process ofmachine tool conditioning. Real-time diagnostic results showthat each component correctly represents examined states orfaulty states of the system.A dynamic model and dynamic simulation of this system iscapable of investigating the operation of the process undernormal conditions as well as various faulty conditions. Thepurpose of the solved system is also to detect and localize intime possible abnormalities and faults of machineryequipment and thus prevent the damage of technologicalprocess.A solved diagnostic system is very complex. There are thefollowing topics:- equipment condition monitoring- data acquisition for equipment reliability modeling- database for equipment state evaluation- knowledge-based system for equipment fault diagnosis- impact of plant operation on maintenance activities.II. PROBLEM SOLVINGA measurement of the vibrations and temperatures, analysisManuscript received March 22, 2008. This work was supported by theSlovak Republic’s Ministry of Education under Scientific Grant Project:Intelligent Approach to Automated Diagnostic Problem Solving ofMachinery Equipment (Number: VEGA 1/4133/07).Stefania GALLOVA. Author is with the Technical University of Kosice,Letna 9, SK-042 00 Kosice, Slovak Republic; phone: 421-904-584-426;e-mail: stefania.gallova@zoznam.sk; stefania.gallova@tuke.skISBN:978-988-98671-9-5of input signals and their processing is an important part ofthis work. Sensors are connected to a processing unit. In thesecond phase we transform based signal features, which willbe used as the inputs to domain knowledge-based system.There is the problem-dependent data structure representation,and cost function, i.e. fitness, evaluation, and the robustreproduction phase, which are functionally separated and maybe common to each application. That's reason for creating agood and relevant genetic model for clustering and adoptingefficient operators for the optimisation.Set of solutions is represented by a string of numbers. Theuse of a binary representation means that each point on thestring can be occupied by one of only two alleles. There arecan be either a „1“ or „0“ at each point of the string. A solvedgenetic algorithm starts to work with a set of domainknowledge structures that are coded into binary strings. Thediagnostic rules, which are to be evaluated through geneticalgorithm, should then be coded. A solved system used herehas the diagnostic rules (production rules representation).Diagnostic rules are in the following form, which are easilyrealised also within expert (knowledge-based) systemenvironment:IF ( S1 Ù S 2 Ù . Ù S n ) THEN( Fi )(1)This formula states that if symptoms S1 to Sn are presentthen the ith fault (Fi) occurs. The used symptoms S1 to Sncorrespond to n different on-line information sources, whichcould be on-line measurements and controller outputs. Eachsymptom is considered to take one of the following values:-increase, steady, decrease, neutral. Each symptom in thecondition part of rule is coded by a 2-bit binary, where „00“stands for symptom decrease, „01“ stands for symptomsteady, „10“ stands for symptom increase, „11“ stands forsymptom neutral. Symptom neutral means that thecorresponding symptom is not important. For example,a normal rule applying can match with any values. Byintroducing this symptom, the condition parts of all thediagnostic rules will be of the same length and thecorresponding parts of the rules will represent the sameobservations.The quality of a genetic algorithm approach seems to bedependent on a few important parameters and operatorvariants. Fitness function, as well as the parameters of thefitness function, can affect the result of learning process. Thefitness of an individual structure is a measure indicating howfitted the structure is.We illustrate a genetic algorithm scheme:WCE 2008
Proceedings of the World Congress on Engineering 2008 Vol IWCE 2008, July 2 - 4, 2008, London, U.K.Genetic Algorithm ()( Initialize population while Not (Stop condition) do( Fitness evaluation Selection Reproduction and Mutation ) Choose final solution )A genetic algorithm is a stochastic computational modelthat seeks the optimal solution to an objective function. Thesearch is performed through an iterating procedure applied toa „population of individuals“, i.e. a set of feasible solutions.Searching strategy is similar to biological evolution, i.e. bettersolutions are reproduced, whereas worse solutions arediscarded. Thus, the search strategy is based on the possibilityto discriminate between elements in order to resolve which isthe good solution of the fitness function and therefore hasa good chance of reproducing and generating new elementswith its genetic inheritance.Experiments with different approaches to solve fitnessfunction models evaluation lead to the different results.Earlier experiments with a single population and simple(classical) based genetic algorithm approach convergedprematurely to solutions of poor quality. A genetic algorithmhas to maintain a balance between the preservation of goodcombinations of genes, and the exploration of newcombinations. We adopt a successful strategy for achievingthis balance which has been to combine a highly explorative,or disruptive crossover with elitism, in which a fraction of thebest individuals found so far survive into the next generation.Elitism gives better individuals more chances of mating toproduce fit offspring, an advantage when their offspring willfrequently be poor.This (solved) system is a good illustration of how a geneticalgorithm approach to a complex practical combinatorialproblem can provide an extremely robust solution withseveral practical advantages. The main difficulty in theproblem is that there typically is a multitude of local extreme,which happen to be located close to a bounding constraints,conventionally imposed at the given threshold, and thatanyway has to be imposed out of safety considerations.The aim of the research (two methodologies) describedhere was to investigate the factors involved in designinga genetic algorithm with respect to the overall objective ofrobustness and utility as a practical tool. By robustness, wemean the ability of the program to produce good solution inreasonable time, independently of any user interaction in theform of careful set-up or run-time intervention, possiblyrequiring considerable expertise.Given the time constraint within which the program mustwork, within a genetic algorithm approach the problemdevolves into the design of a number of components: geneticrepresentation of candidate solutions, selection of mattingpairs and recombination of their genetic material, randommutation of genetic material, measure of fitness for a givenpopulation, and population distribution.We solve a non-linear fitness function, which has beenconstructed (and optimized) to direct the search efficiently inISBN:978-988-98671-9-5the presence of the many local optima that result for theconstraint on solutions. The contribution describes the designof an efficient and robust genetic algorithm for the diagnosticsystem problem – a complex combinatorial, multimodeloptimisation. A technical computing environment is Matlabprogramming tool.A. First methodology approachThe primary genetic algorithm is applied to find the bestgenetic algorithm for a secondary problem, which closelyresembles the properties of the primary problem, see Fig.1.Each of the secondary runs of genetic algorithmindependently to produce a solution of the problemconsidered. The fitness of the solution influences theoperation of the primary genetic algorithm. Such a problemwould require (for example) real valued genes, the possibilityto treat logical subgroups of genes as an atomic unit anda sufficiently complex search space with multiple suboptimalpeaks. The values obtained for the best secondary geneticalgorithm in this scenario are then copied and used as theparameter settings of the primary genetic algorithm foroptimizing the secondary genetic algorithm for the particularproblem to be solved.On the bottom level, the secondary genetic algorithmapproach operates on a population of gene strings thatrepresent possible solutions of the problem to be solved. Onthe top level, the primary genetic algorithm approach workson a population of secondary genetic algorithm approach,each of which is represented as a separate gene string.Each of the secondary genetic algorithm approach runsindependently to produce a solution of the problemconsidered, and the fitness of the solution influences theoperation of the primary genetic algorithm approach. Thenumber of generations created on the two levels isindependent of each other. The string with the highest fitnessin the last primary generation is expected to be the bestgenetic algorithm for the original problem. Generally, a prioriinformation determines the kinds of genetic operators andtheir parameters settings for the primary genetic algorithmapproach. The genetic operators and parameters values areinitially determined by the designer (domain expert) and bythe Internet technology approach.The genes as a decision in the string are numericallyrepresented by the probability that a particular variant ofa genetic operator is selected among a limited number ofvariants of that operator. The genes as parameters in the stringspecify a real-coded value associated with the selectedvariant.We define many parameters and operators within solvedgenetic algorithm approach. There are some more relevantparameters. Elitist component decides if the best stringgenerated up to time t should be included in the population ofgeneration t 1. The distance between two strings is measuredand their fitness is modified in the sense that strings in thesame neighbourhood are forced to share their fitness amonganother, which effectively limits the uncontrolled growth ofparticular species within a population. We use three variouscrossover operators: order crossover, cycle crossover andpartially matched crossover. We have also a distance ofcrossover points parameter, which determines the maximalWCE 2008
Proceedings of the World Congress on Engineering 2008 Vol IWCE 2008, July 2 - 4, 2008, London, U.K.distance between two crossover points in the used crossoveroperator for reordering problems. The crossover unit reflectsthe decision if the crossover operator (as usual) shouldconsider genes as the smallest atomic entity. Anotherpossibility considers if it should be applied such that logicalsubgroups of genes stay together as a structural unit [2], [3].The crowding technology has been introduced to induceniche like behaviour in genetic algorithm search in order tomaintain diversity in the population. Mutation probabilityparameter determines the probability of applying the mutationoperator to a gene. We also realise the decision if the mutationamount of real-valued genes is determined according toa normal distribution or an exponential distribution. Specialparameters determine the density function of the normaldistribution of the mutation amount and the mean value of theexponential distribution of the mutation amount. Mutationvalue replacement reflects the decision of the mutationoperator should overwrite the old gene [4].We explicitly distinguish between decisions andparameters values and provide several alternatives for thedecision components. In the absence of an a priori known bestor worst fitness of a string, the solved methodology of fitnesscalculation is based on the idea of measuring the increase offitness created by the secondary genetic algorithm approach.We have a parameter Fmn , which is the best fitness of a stringin the secondary population for test problem example m aftergeneration n.A parameter pt is the number of test problems.The fitness of the secondary genetic algorithm GAFs aftergeneration n is given by:Fitness[GAFS ] 1ptmax(0, Fnm - F0m )1 - F0mm 1ptå(2)By this way, we will avoid distorted fitness values arisingfrom possibly different degrees of complexity of test problemexamples. The fitness of an individual structure is a measureindicating how fitted the structure is [5].We realize many experiments with various changes ofgenetic algorithms operators. For example, we implementedmutation as follows. A single mutate step with two new(created) children consists of randomly choosing two bits inthe string of length 1,660 and interchanging their values. Theprobability that a single mutate will actually modify anindividual is 0.29.Fig.2 shows the performance profile with indicators of themost significant decisions. They become important after the70th primary generation, when other decisions and parametershave already been appropriately determined. The optimalcrossover probability evolved as 0.51. This value remainednearly constant for all strings lengths investigated. Theoptimal mutation probability value increases with increasingnumbers of genes in a string.The fitness of the best secondary genetic algorithmapproach increases with increasing primary generations. Thebest secondary genetic algorithm is selected after every 6thprimary generation, applied it to 60 randomly generated testproblems with different numbers of weights parameters tooptimize (for 120 secondary generations) and measured itperformance.ISBN:978-988-98671-9-5DIAGNOSTIC SYSTEM PROBLEMPRIMARY PROBLEMParameters of Primary GeneticAlgorithmControl ProcedurePRIMARY GENETICALGORITHMThe BestStringPrimary PopulationEvaluation ProcedureObjective Function(Primary Problem)Parameters of Secondary GeneticAlgorithmControl ProcedureSECONDARY GENETICALGORITHMThe BestStringSecondary PopulationEvaluation ProcedureObjective Function(Secondary Problem)F ITN E SSSECONDARY PROBLEMFig. 1 Optimization problem architecture (two-levelapproach)The total computation time required for performing thewhole cycle was 38 hours. It involves 115 primarygenerations and a total of 79000 secondary generations forthe 42 test problems every 6th primary generation. ThisWCE 2008
Proceedings of the World Congress on Engineering 2008 Vol IWCE 2008, July 2 - 4, 2008, London, U.K.complex procedure was repeated six times. The final values offitness performance profile are averages over the sixexperiments, see Fig.6. Up to primary generation 22, theincrease in quality is slow, but then it significantly gets largerduring a few primary generations. In the third phase, theincrease again slows down. The main reason for this type ofbehaviour is that the quality of a genetic algorithmmethodology seems to be dependent on a few importantparameters and operator variants [5], [6].the optimal decision boundaries between the different faultystates with as many details as possible or needed even in thepresence of noise and uncertainties.InputsCrisp emberships of Fuzzy SetFuzzy rulesMemberships of Fuzzy SetDefuzzificationCrisp ValuesFig. 2 The Most Significant Decisions Indicators:y1 crossover units course; y2 selection method course;y3 elitist model course; y4 mutation function course;y5 value replacement courseThe solved genetic algorithm approach is feasible byimplementing it in a multi-transputer environment.Performance results for finding the best genetic algorithm forthe problem of optimal operation parameters in solveddiagnostic system have demonstrated the quality of ourimplementation.B. Second methodology approachIf we have a solved diagnostic signal, which is morecomplex, vague, uncertain, we realize a second methodologyapproach (we observe and analyze more signal parameters).The reason for this approach is to implement a real predictivemaintenance system.Fuzzy rules based system with an optimization by geneticalgorithms approach can be effectively used to obtain relevantresults in real diagnostic system problem solving. Fig.3illustrates a fuzzy rules handling procedure.The fuzzy theory may be combined with the genetic model,for instance by putting a value between „0“ and „1“ in theBoolean cluster code to act as object belonging to probability.We have a primary expert system (Spel-expert 4.0) andsecondary GA approach.The aim is to detect changes of the current processbehaviour and to generate analytical symptoms. Thediagnosis task is accomplished by fuzzy evidentialapproximate reasoning scheme to handle different kinds ofuncertainty that are inherently present in many real worldprocesses, and to make decision under conflicting data orknowledge.The diagnostic system serves several purposes. It identifiesISBN:978-988-98671-9-5OutputsFig. 3 A Fuzzy Rules Handling ProcedureThe diagnostic system can be improved and can be moreeffective when taking into account aspects like simplicity,transparency, uncertainty and conflict management. Thediagnostic process includes first of all trees of fault symptomsand fuzzy rule basis that use explicit knowledge to treatheuristic symptoms, which are mostly obtained by inspectionsthrough a human operator and by stating the statistics historyof the monitored process. The appropriate heuristic rules forthe fault diagnosis are then gathered through consultingexperts [1].Diagnostic process includes also measurement data togenerate analytical symptoms. We have to solve the lack ofsufficient information and the existence of uncertainty. Wesolve an iterative hierarchical optimization problem.Procedure of parameter optimization is realised byfollowing way. The identification procedure of obtainedparameters leads to optimization and tuning procedures. Wesolve the forward procedure. The functional models FMij,i 1,.,m; j 1,.,l are identified by solving a least squareproblem. The backward procedures fix the functional models.The parameters of the membership functions pik, qik,;i 1,.,m; k 1,.,n are updated by an effective non-lineargradient descent optimization technique. It requires thecomputation of derivates of the objective function to beminimized with respect to the parameters pik, qik,. We applythe optimization algorithm with variable learning ratesprocess using.We have a set S (xp, sp)Np 1 , such that xp Î X Ì Rr; sp ÎY Ì Rl, the objective is to find subsystem yj(xp) in the form:WCE 2008
Proceedings of the World Congress on Engineering 2008 Vol IWCE 2008, July 2 - 4, 2008, London, U.K.nmå FM ij Õ ei 1y j ( x) -( xk - pik ) 2qik2k 1( x - p )2m n - k 2 ikqik(3)åÕ ei 1 k 1We minimize the function of the mean squared error:Er 1 lå ( y j - s jp ) 22 j 1(4)where x p Î S .We solve the mean pik , variance qik (i.e. ellipsoidalfunctions) and the adjustment of the FMij. We also assumethat pik Î Xi; qik 0; FMij Î Yj. We solve a complex non-linearmulti-input and multi-output relationship withx (x1,Tnx2,.,xn) Î X Ì R . Parameter x is the vector of inputvariables. We have also yÎ Y Ì Rl. Parameter y
best genetic algorithm approach as an optimisation problem and use another genetic algorithm approach to solve it. A methodology calculation is based on the idea of measuring the increase of fitness and fitness quality eva.luating created by two methodologies with secondary genetic algorithm approach using.
The Genetic Code and DNA The genetic code is found in a acid called DNA. DNA stands for . DNA is the genetic material that is passed from parent to and affects the of the offspring. The Discovery of the Genetic Code FRIEDRICH MIESCHER Friedrich Miescher discovered in white blood . The Discovery of the Genetic Code MAURICE WILKINS
genetic algorithms, namely, representation, genetic operators, fitness evaluation, and selection. We discuss several advanced genetic algorithms that have proved to be efficient in solving difficult design problems. We then give an overview of applications of genetic algorithms to different domains of engineering design.
An Introduction to Genetic Genealogy Overview Genetic Genealogy using genetic analysis as a genealogical tool relies on two special types of DNA (one for direct male line and one for direct female line) Some of my experiences with genetic genealogy Pike Surname DNA Project started in summer of 2004 currently has 24 participants (2 from Newfoundland)
n 33: ponticelli – via stazione n 36: via bergullo n 49: via punta n 50: montericco – rio palazzi n 52a: montericco nord n 52b: montericco nord n 54: via dal monte n 56: via del santo n57: lughese nord n58: lughese sud n 60: via i maggio n 64: via montanara – pedagna sud n 68: via molino rosso - h.e.r.a. n 69: via gardi
3.1 Otolith imaging via CT scanning 5 3.2 Age estimation via machine learning 6 4. RESULTS 8 4.1 Otolith imaging via CT scanning 8 4.2 Age estimation via machine learning 13 5. DISCUSSION 15 5.1 Otolith imaging via CT scanning 15 5.2 Age estimation via machine learning 16 5.3 Potential for applying machine learning to otolith CT im ages 17 6.
SolidWorks (CAD CAM CAE) Address: B-125, Sec-2, Noida Web: www.multisoftsystems.com Contact: 91-9810306956 Landline: 91-1202540300/400 . Applying the Cam Mate Applying the Gear Mate Applying the Rack Pinion Mate Applying the Screw Mate Applying the Hinge Mate
NETWORK. Genetic diversity, population differentiation, and analysis of molecular variance (AMOVA) were used to determine genetic structure. MEGA was used to construct phylogenetic trees. Genetic diversity of J. hopeiensis was moderate based on nuclear DNA, but low based on unipa-rentally inherited mitochondrial DNA and chloroplast DNA.
N. Sharma, M. P. Rai* Amity Institute of Biotechnology (J-3 block), Amity University Uttar Pradesh, Sector-125, Noida, India A B S T R A C T P A P E R I N F O Media requirement for microalgae cultivation adds most to the cost of biodiesel production at Paper history: Received 30 July 2015 Accepted in revised form 6 September 2015 Keywords: - Chlorella pyrenoidosa Cattle waste Lipid content .
