Optimising the stacking sequence of Type IV composite overwrapped pressure vessels (COPVs) is challenging due to dome geometric complexity, progressive damage behaviour, and high computational cost. This study proposes an efficient optimisation framework for compressed natural gas (CNG) COPVs by coupling experimentally validated finite element modelling with machine learning. A high-fidelity finite element model incorporating dome-specific variations in fibre angle and layer thickness and progressive damage evolution via a VUMAT subroutine was used to generate a training database. Validation against four experimental prototypes yielded relative errors of 1.3–11.7% in burst pressure and 0.8–7.2% in composite mass, with good agreement in failure modes and dome thickness distribution. An artificial neural network (ANN) trained using physically informed secondary variables achieved high predictive accuracy, with a minimum mean squared error of 1.84 × 10 -3 and Pearson correlation coefficients between 0.9505 and 0.9987 for mass, burst pressure, and stress ratio. The ANN surrogate was integrated into particle swarm optimisation and genetic algorithm frameworks, identifying an optimal stacking sequence that reduced composite mass by 18.3% while satisfying the minimum burst pressure requirement and increasing dome-failure safety by 32.6%. The proposed ANN-assisted optimisation reduced computational cost by 96.28%, enabling practical and scalable optimisation of filament-wound COPVs with progressive damage modelling. Highlights : • Artificial neural networks enable optimisation of pressure vessel stacking sequences • Dome angle and thickness variations are captured using progressive damage modelling • Mass, burst pressure and failure location are predicted with high accuracy • Particle swarm optimisation reduces composite mass by 18% while improving safety • Machine-learning-assisted optimisation reduces computational cost by over 96%
Agne et al. (2026) studied this question.