Developments in the use of the genetic algorithm in engineering deign
What is a Genetic Algorithm?
A Genetic Algorithm (GA) is a powerful nonlinear search and optimization technique that is particularly well suited to the solution of multivariate design problems that are both multimodal and discontinuous in nature. Its use can facilitate rapid identification of good design options whilst avoiding convergence upon local optima.
The general structure of a GA can be considered to be analogous to the process of Darwinian evolutionary theory in which the characteristics are transmitted from one generation to the next by genes and organisms evolve under the pressure of fitness proportionate reproduction.
How a Genetic Algorithm works?
The GA commences its search of the design space not from a point based on an initial guess but from a number of randomly selected points. To commence a global search of the design space, an initial population of design solutions is generated. Each solution is represented as a string of variables (constituent elements) that can acquire different values (alleles). The fitness of each design solution is then determined by evaluating the extent at which they satisfy set of predefined criteria. Designs are then assigned with a “selection probability” that is proportional to their level of fitness. In this way, while any design may be selected as a candidate for the next generation, chances are that the most fit will be selected several times and the least fit not at all.
Whilst some of the selected designs are passed without modification from one generation to the next, some of them are selected to crossover. This means that a portion of the selected designs will be arranged in pairs so they can exchange part of the information contained in their strings. A limited amount of new information is introduced into the process by means of a mutation operator which causes randomly selected digits within the combined strings to change value in accordance with some predetermined level of probability. Mutation can also be applied to uncombined strings (designs that were not subjected to crossover).
While crossover enables for the better adapted traits to spread among the generations, mutation enables the search for good solutions to be extended into regions of the design space for which the genetic code either never existed within the original population or has been eliminated due to its associations with come other unfit characteristics.
