Genetic Algorithm-Based Scheme Generation for Mechanical Products in Conceptual Design

Abstract

Conceptual design’s intrinsic uncertainty, imprecision, and lack of information lead to the fact that current conceptual design activities in engineering have not been computerized and very few CAD systems are available to support conceptual design. In most of the current intelligent design systems, approaches of principle synthesis, such as morphology matrix, bond graph, and design catalogues, are usually adopted to deal with the scheme generation in which optional schemes are generally combined and enumerated through function analysis. However, as a large number of schemes are generated, it is difficult to evaluate and optimize these design candidates using regular algorithm. It is necessary to search a new approach or a tool to solve the scheme generation in conceptual design. Generally speaking, scheme generation in conceptual design is a problem of scheme synthesis. In essence, this process of developing design candidate is a combinatorial optimization process, viz., the process of scheme generation can be regarded as a solution for a state-place composed of multi-schemes. In this research, genetic algorithm (GA) is utilized as a feasible tool to solve the problem of combinatorial optimization in scheme generation, in which the encoding method of morphology matrix based on function analysis is applied, and a sequence of optimal schemes are generated through the search and iterative process which is controlled by genetic operators, including selection, crossover, mutation, and reproduction in GA. Several important problems on GA are discussed in this research, such as the calculation of fitness value and the criteria for heredity termination, which have a heavy effect on selection of better schemes. The feasibility and intellectualisation of the proposed approach are demonstrated with an engineering case. In this study scheme generation is implemented using GA, which can facilitate not only generating several better schemes, but also selecting the best candidate. Thus optimal schemes can be conveniently developed and design efficiency can be greatly improved.