Pipe Route Design Optimization Using Genetic Algorithms

This paper describes a study of automatic uid pipe route design using Genetic Algorithms (GAs). The pipe route generation process is deened as an optimization problem, based on viewing it as a variant of the Minimal Rectilinear Steiner Tree Problem (MRStT), Gib84], Jul93]. GAs were used for the optimization, and gave promising results. Other optimization methods were used for comparison. Result from this research is expected to be used in automating the pipe route design process. Fluid pipe route design is an area which requires experienced designers because of its complexity and the large search space involved. It is one of the stages of ship design that remains unautomated. Genetic Algorithms are an optimization method taking inspiration from natural evolution. Solutions with good performance are selected from a pool of feasible solutions to participate in reproduction to produce oosprings. OOsprings are expected to have better performance than ancestors, which is the process of getting closer to the optimal solution for a given problem. Genetic Algorithms require a representation of a solution to the problem, which can undergo the reproduction process and produce valid oospring. For this study, three diierent representation methods for pipe routes were devised and their performance was compared. A binary representation method with heuristics to reduce the search space worked well. High cardinality integer representation, BBM93], without any assumption also showed reasonable performance in spite of the huge search space. Simulated Annealing (SA), Stochastic Hill Climbing (SHC), and Random Search (RS) were used for the performance comparison with the GA. The GA performed better than the other techniques across a wide range of parameter settings. SA showed the second best performance.