Design optimisation of braided composite beams for lightweight rail structures using machine learning methods

Braided composites have seen substantial industrial uptake for structural applications in the past decade. The dependence of their properties on braid angle provides opportunities for lightweighting through structure-specific optimisation. This paper presents an integrated approach, combining finite element (FE) simulations and a genetic algorithm (GA) to optimise braided beam structures in the spaceframe chassis of a rail vehicle. The braid angle and number of layers for each beam were considered as design variables. A set of 200 combinations of these variables were identified using a sampling strategy for FE simulations. The results were utilised to develop a surrogate model using genetic programming (GP) to correlate the design variables with structural mass and FE-predicted chassis displacements under standard loads. The surrogate model was then used to optimise the design variables using GA to minimise mass without compromising mechanical performance. The optimised design rendered approximately 15.7% weight saving compared to benchmark design.