抄録
This paper presents an automatic design method for pipe arrangement. A pipe arrangement design problem is proposed for a space in which many pipes and objects co-exist. This problem includes large-scale numerical optimization and combinatorial optimization problems, as well as two criteria. For these reasons, it is difficult to optimize the problem using usual optimization techniques such as Random Search. Therefore, multi-objective genetic algorithms (GAs) suitable for this problem are developed.
A pipe is characterized by both a pattern of generation and numerical parameters. The former describes the way the pipe bends and the latter details the length of the straight parts. For this reason, a combination of the pattern of generation and the numerical parameters is used for the solution representation and a new method of crossover is proposed that takes into account interference with obstacles. As the number of pipes increases, it becomes rapidly more difficult to find feasible solutions where pipes do not interfere with each other. Therefore, two modification operators that transform infeasible solution candidates into feasible ones are introduced. One operator modifies the pipe having a lot of interferences with other pipes so that it will not interfere with them, and the other is related with the operation that modifies the pipe that travels through obstacles. Although there are cases in which pipes cannot completely avoid obstacles in practical designs, this situation is taken into consideration by this design process. The proposed method for optimizing a pipe arrangement efficiently is demonstrated through several experiments, and remarks are provided for applying this methodology to a practical pipe arrangement design.
