Reinforcement Learning Framework for Aeroengine Pipe Routing

A novel framework for reinforcement learning, named Frenet-based pipe routing optimization (FPRO), has been introduced to incorporate manufacturing insights into the design of free-form pipes for aeroengines. This framework tackles the separation between pipe routing and subsequent manufacturing processes, which currently necessitates extensive trial-and-error efforts. FPRO redefines the routing challenge as a boundary value problem within the Frenet frame, using cubic Hermite interpolation to create curvature and torsion profiles for pipe paths. Additionally, constraints based on specific manufacturing knowledge are integrated to limit acceptable ranges of curvature and torsion. This research is available on arXiv with the identifier 2605.20644.