Intelligent layout of nuclear pipeline system with mechanical constraint

The layout of pipeline systems in nuclear plant including path planning and support arrangement is achieved manually. Trials and iterations are inevitable in the design process to meet mechanical constraints, such as the strength criterion. The approach suffers from heavy labor intensity, long design cycles, and high costs. In this study, we propose an intelligent layout method that combines the reinforcement learning with an optimization algorithm to satisfy both spatial and mechanical constraints in nuclear plant design. We first employ the Q-learning algorithm to solve the path planning of pipeline in a determined space with obstacles. The Q-learning algorithm makes a pipeline layout with a total length of 22,000 mm and 5 bends, overweighing the traditional Ant-colony algorithm (26,000 mm-length and 13 bends) and NSGA-II (26,000 mm-length and 12 bends). When the stress of the arranged pipeline exceeds the threshold value, the algorithm combining Particle Swarm Optimization (PSO) and Genetic Algorithm (GA) is used for the intelligent layout of supports, where the optimal number and positions of supports are automatically determined with a balance between meeting economic costs and design criteria for mechanical responses. The layout example shows the optimal arrangement of three supports. Further increasing the number of supports, the maximum stress stabilizes. Increasing the cost with additional supports becomes ineffective. The proposed intelligent method reduces the expertise dependence of pipeline layout and can be easily extended to other industrial fields.