Session: 02-03-02: Optimization

Paper Number: 144011

144011 - Automatic Piping Routing Applying Multi-Objective Genetic Algorithm 

Piping for fluid transport is essential in industrial plants. However, the design of piping requires an enormous amount of time. This is because, in a series of piping designs, it is necessary to plan the equipment layout in the facility, design the piping layout (hereinafter referred to as piping routing), analyze and evaluate the seismic resistance of pipes and equipment, etc., in sequence, and redesign them depending on the evaluation of the analysis results. Depending on the scale of the project, a vast number of pipelines (more than 20,000) may be designed. Piping routing has many requirements, such as keeping costs down (e.g., shortening the total pipe length), not interfering with other equipment, and having sufficient seismic performance. Furthermore, the piping design is complex and difficult because it is carried out after the location of plant equipment has been decided while satisfying these items. The design is currently carried out by skilled engineers using analysis tools. In the future, the development of automatic pipe routing tools is desired to improve work efficiency.

Our research group has developed an automatic piping routing tool by applying a multi-objective optimization genetic algorithm with the abovementioned requirements as objective functions. This paper presents an example of tool construction focusing on the total pipe length, interference with obstacles, seismic resistance, and the number of bends. The total pipe length, seismic resistance, and number of bends were used as objective functions. In contrast, interference with obstacles was used as a penalty function, and the sum of these functions was used as the overall evaluation value. The tool minimizes this overall evaluation value by MOGA.

In this tool, evolutionary calculations were performed for up to 100 generations. In this case, 100 initial generations were generated from the A* search algorithm, and the next generation was obtained from A* search algorithm and MOGA with this as the parent. The piping layout information is expressed as a piping gene, representing the position of the piping bend coordinates. The piping genes are evolved through crossover and mutation by combining the A* search algorithm and MOGA to obtain a layout that minimizes the overall evaluation value. However, since gene crossover and mutation are generated stochastically, different final layouts are obtained from trial to trial. This variation in the final layout leads to polysemy in the design solution in industrial terms, which needs to be quantitatively evaluated for quality control of the tool.

This paper performed several examples with fixed starting and ending points in the piping routing space. Several obstructions were placed in the routing space, and a "designable space ratio" was defined to represent whether piping routing could be performed in an area that did not interfere with the obstructions. Variations in the final layout obtained were evaluated when this index was varied.

Presenting Author: Akane Uemichi Yamaguchi University

Presenting Author Biography: Dr. Uemichi studied mechanical engineering at University of Tsukuba since 2005. She received her doctoral degree from University of Tsukuba in 2013. Her doctoral thesis dealt with combustion fundamentals. After that, she spent almost five years as an Assistant Professor at the University of Tokyo, working on the application aspect of combustion fundamentals and engineering and starting fluid-induced vibration research. She then spent nearly one year as an Assistant Professor at Tokyo University of Agriculture and Technology since 2019 and four years as an Associate Professor at Waseda University since 2020. Currently, Dr. Uemichi is working as an associate professor at Yamaguchi University.

Authors:

Shunta Mogi Waseda University
Akane Uemichi Yamaguchi University