Keywords: data-driven, surrogate model, stacking sequence optimisation, genetic algorithm, manufacturing constraints, composite material.

This study addresses the computational bottleneck and manufacturing constraint integration challenges inherent in conventional genetic algorithm (GA) based stacking sequence optimisation for carbon fibre reinforced polymer (CFRP) laminates, where fitness evaluation traditionally relies on finite element analysis (FEA). A surrogate-assisted optimisation framework integrating one-dimensional convolutional neural networks (1D-CNN) with evolutionary algorithms is proposed to enable efficient stiffness optimisation while ensuring manufacturability. An automated batch processing pipeline combining Python scripting with Abaqus commercial software was developed to generate a high-fidelity dataset comprising 30,185 samples. The trained CNN surrogate achieves millisecond-level prediction of structural compliance with a coefficient of determination (R²) of 0.9846 on the test set. To handle manufacturing constraints without sacrificing boundary search efficiency, a stepped soft-penalty fitness function was devised, quantifying violations of balance, contiguity, adjacency, and minimum proportion rules as continuous penalty metrics rather than binary hard penalties. Validation on a 40-layer cantilever beam demonstrates that the design feasibility rate reaches 80% within 20 generations, while the optimised solution surpasses the best feasible design in the original database by 6.43% in compliance reduction, with FEA verification errors below 1%. The proposed CNN-GA coupled framework effectively balances computational efficiency with engineering practicality, offering a viable pathway toward intelligent composite design.

download the full-text of this paper (PDF, 14 pages, 740 Kb)

go to the previous paper
go to the next paper
return to the table of contents
return to the volume description