A Production Scheduling A Production Scheduling Problem Using Genetic Problem Using Genetic AlgorithmAlgorithm

Presented by: Ken Johnson

R. Knosala, T. WalSilesian Technical University, Konarskiego

Gliwice, Poland

IntroductionIntroductionThe way of Flexible Manufacturing cell work

Scheduling with the aid of genetic algorithm and draft of code strings,

Results obtained by computer program have been presented.

In the first case it has been assumed that the cell works in optional mode (every operation can be done on every machine)

In the second, each works in sequential mode (the first operation is executed on the first machine, the second operation on the second, etc…)

The only criterion of evaluation is the time of work. (shortest for a finite number of jobs and machines).

Genetic AlgorithmsGenetic AlgorithmsSearch algorithms, based on natural

selection mechanisms and heredity. They join the survival principle of

the best fitted strings with systematic information exchange.

In every generation the new group of artificial organisms, made from the fusion of the best fitted representatives fragments of previous generation, come into existence.

GeneticsGenetics

Task Parameters Task Parameters (values of function (values of function domain) must be transformed to the code domain) must be transformed to the code strings. strings.

1. they do not directly transform task parameters, but their coded form.

2. they lead searching, coming out not from one point, but from some population of points.

3. they use only fitness function, but do not use derivative or other auxiliary information.

Design PrinciplesDesign PrinciplesFirst block defines which jobs are

first taken into consideration Within each job are the

operations in order of succession when machining

Program StructureProgram Structure

Program leads operations of genetic algorithm for 600 generations (it is constant, assumed number).

There are 30 individuals (code strings) in every generation.

Fitness FunctionFitness Function

Maximizes work time of longest working machine

Singles out the worst, and gets rid of it

Takes bottle-necking into account

CrossoverCrossover

MutationMutationEnsures ‘natural selection’ is

following the best routeOccurs in both 1st and 2nd blocksIn 2nd block, a ‘double’ mutation

occurs

ModelsModelsScheduling 3 jobs to 2 machines:

ResultsResultsIn the form of Gantt ChartsFor a more complex problem:

ResultsResultsReached “ near optimal ”

solution very fast (by 200 generations)

ConclusionsConclusionsGenetic algorithm has generated correct

schedulesNot sure that the solution is optimal. Number of jobs and their operations

have not had influence on quality of obtained results

Gained schedules have been correct for all cases, that means strings assure right

Applied structure of code string has assured good, but not the best, efficiency of creation and propagation of schemes

Assured high adjustment of strings