3. Las colocaciones españolas para los alumnos chinos
3.2. Tres dimensiones de las colocaciones españolas para los alumnos chinos
3.2.2. La dimensión léxica de las colocaciones españolas
3.2.2.2. De la selección léxica a la selección cognitiva
With the development of new materials, such as FRP, for the construction of ship structures, the requirement for structural connections for the construction of large vessels regardless of the material being used and the emphasis on short term and long term behaviour characterisation of structures for use in the marine industry has placed a high level of importance on efficient design. In the context of the present research efficient design is defined as the most efficient solution to meet the design requirements, for example low resistance for hull-form design or maximum pay load capacity for a given displacement. The present research concentrates on detailed design rather than global design issues. However, the methods of optimisation for both global and detailed design problems are the same.
Optimisation methods come in many forms, however the choice of method depends greatly on the problem being optimised. Secant or Newton-Raphson methods [59] are efficient ways of finding minimum or maximum values of a one-dimensional function and hill- climbing methods can be used to find minimum or maximum values of multi-dimensional problem. However, engineering design problems such as joint design or hull-form design often consist of large discrete, non-linear and often discontinuous design spaces. In this
case more complex optimisation techniques such as evolutionary algorithms can be applied. This method of optimisation is based on the traditional Darwinian theory of “survival of the fittest” found in nature. Competition in nature for finite resources results in the fittest individuals dominating over the weaker ones.
Genetic algorithms (GAs) form one of the evolutionary optimisation techniques and is inspired by natural selection and genetics. The method is very general and is capable of being applied to a wide variety of problems. However, its use in the optimisation of engineering problems is not as widespread as one may expect. The use of genetic algorithms began in the 1960’s by John Holland on the subject of cellular automation. This is a discrete model studied in mathematics and theoretical biology and consists of an infinite, regular grid of cells (similar to a page of graph paper). A typical example is that each cell can have two possible states, black and white, and has 8 neighbours (those cells touching it) therefore there are 512 possible patterns for the cell and its neighbours. The most common reference to cellular automaton is the “game of life” invented by John Horton Conway. Games such as this form the building blocks for what is now called genetic algorithms. Fogel [60] describes simulated evolutionary optimisation in terms of genetic algorithms, evolution strategies and evolutionary programming and provides a detailed review of simulated evolutionary techniques and identifies that it will form an important method of solution to real world problems but its advancement will come from “careful observation and abstraction of the natural process of evolution”.
Four main steps can be used to define evolutionary optimisation methods: 1. A number or population of guesses of the solution to the problem; 2. A way of judging the quality of each individual in the population;
3. A method of mixing the fragments of the better solutions to form a new population; 4. A mutation operator to ensure diversity of the population.
The question remains why use a genetic algorithm to solve a particular problem rather than more traditional methods? One of the most common reasons is that an exhaustive search of the problem may require calculations that may take many years to complete. In this case there is good reason to use GAs as a method of efficiently searching a large problem. There
are relatively few examples in the literature of the use of GAs to solve particular design problems.
Day and Doctors [61], used evolutionary algorithms to assess the hydrodynamic performance of ship hull forms. The hull forms are described by a fixed number of variables based on mathematical curves. The variables are given a number of limitations to ensure realistic hull forms are created. The variables are then chosen at random to generate an initial population. Hydrodynamic resistance prediction based on the Michell integral [62] was used to assess each individual of the population. Selection was based on minimum resistance and “breeding” was carried out using crossover and mutations to create the new generation. The work showed that the use of GAs provided a robust and rapid first principles technique to assess the design space for minimum resistance hull forms.
Genetic algorithms were used along with a gradient method which used a truncated Newton algorithm [63] and a direct search approach to investigate two problems, a two-dimensional diffuser and a drag minimisation problem of a fixed area body in flow which can be defined by two and four variables respectively [64]. For the diffuser the goal was to maximise the static pressure recovery coefficient and for the drag minimisation the goal was minimum drag coefficient. In both tests it was found that the more traditional direct approaches of Newton and the direct search produced results in fewer evaluations than with GAs. However, it is agreed by the authors that the GA would be more suited to a large problem that contained a number of local optima in the design space. Neither of the problems examined in this work contained local optima. This is highlighted by the better optimum solution found in the higher dimensional problem of drag minimisation by the GA.
A number of other investigations have utilised genetic algorithms for the design of various engineering structures including, minimum weight design of composite plates and stiffened panels [65], Design of steel structures in tall buildings [66] and design for operation of large container ships [67]. These references are specifically concerned with design. However, their application is to a very large component, i.e. a building, ship or entire system. In the present research the genetic algorithm will be applied to the design of a specific component of the system, the joint between the hangar and deck of a naval vessel.