Analysis and optimization of toroidal composite pressure vessels using data-driven evolutionary algorithms

Toroidal pressure vessels manufactured by filament winding are a promising solution for storing gases or liquids in confined spaces where conventional cylindrical or spherical vessels are unsuitable. This study introduces a hybrid framework that combines analytical modelling of geodesically wound toroids with a data-driven surrogate modelling and multi-objective optimization workflow. The toroidal pressure vessel is analysed under combined internal pressure and thermal loading, while explicitly accounting for manufacturing uncertainty in the fibre volume fraction. Surrogate models constructed with the Evolutionary Deep Neural Networks algorithm are coupled with the constrained Reference Vector Evolutionary Algorithm, which simultaneously maximizes nominal pressure and vessel volume while minimizing weight and the Hoffman failure index within acceptable limits. Four industrially relevant fibre/epoxy systems (E-glass, T300, Boron and Kevlar) were compared across practical fibre volume fractions and toroidal geometry ratios. The obtained results are also validated numerically (Abaqus), with good agreement observed. The results show that optimized T300/epoxy toroids sustain substantially higher allowable internal pressures at similar or higher volume-to-weight ratios, making them attractive for compact gaseous-fuel storage in space-constrained automotive and aerospace applications.