Imperfection-tolerant design of variable-angle filament-wound composite cylinders using surrogate-based multi-objective optimisation
Uzair, Muhammad (2025)
Diplomityö
2025
School of Energy Systems, Konetekniikka
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi-fe20251113107559
https://urn.fi/URN:NBN:fi-fe20251113107559
Tiivistelmä
The structural efficiency of filament-wound composite cylinders is critically limited by geometric imperfection sensitive buckling. Traditional design methodologies account for this using empirical knockdown factors (KDFs), which, though conservative, result in overweight and suboptimal structures. This work presents an imperfection-tolerant multi-objective optimisation framework that integrates experimentally measured imperfections, surrogate-assisted modeling, and advanced global search algorithms to enable aerospace-relevant design of variable-angle filament-wound cylinders.
Imperfection fields obtained via digital image correlation (DIC) are compressed by principal component analysis (PCA) and expanded with Latin hypercube sampling (LHS) to produce statistically consistent synthetic datasets. Consequently, the resulting fields are mapped onto finite element (FE) models in order to perform an evaluation of mass, nonlinear collapse load, and eigenvalue buckling of the corresponding perfect geometry. Consequently, KDFs are computed to quantify the imperfection sensitivity. GPR surrogates are trained for these responses, with adaptive Kriging enrichment guaranteeing the accuracy of the surrogate models within those regions of high predictive variance. Model generalisation is verified by means of cross-validation (CV), whereas sensitivity analysis identifies the winding angles that dominate structural performance.
The tri-objective optimisation problem of minimising mass while maximising nonlinear collapse loade and KDF is addressed using two complementary multi-objective strategies, independently: the Bayesian optimisation (BO) framework, which exploits surrogate uncertainty to accelerate convergence, and the Non-dominated Sorting Genetic Algorithm II (NSGA-II), which promotes global exploration through evolutionary search. Feasibility constraints on mass and imperfection robustness were enforced, and Pareto-optimal sets were benchmarked using convergence histories and hypervolume indicators, revealing the superior sample efficiency of BO and the strong exploration capacity of NSGA-II.
These results confirm that the adoption of realistic imperfection statistics, uncertainty-aware surrogates, and optimisation strategies allows a unique, imperfection-tolerant methodology for VAFW cylinders. In fact, the framework provides quantitative insight into the tradeoffs among lightweight efficiency, buckling strength, and robustness, thus offering design guidelines for next-generation aerospace structures.
Imperfection fields obtained via digital image correlation (DIC) are compressed by principal component analysis (PCA) and expanded with Latin hypercube sampling (LHS) to produce statistically consistent synthetic datasets. Consequently, the resulting fields are mapped onto finite element (FE) models in order to perform an evaluation of mass, nonlinear collapse load, and eigenvalue buckling of the corresponding perfect geometry. Consequently, KDFs are computed to quantify the imperfection sensitivity. GPR surrogates are trained for these responses, with adaptive Kriging enrichment guaranteeing the accuracy of the surrogate models within those regions of high predictive variance. Model generalisation is verified by means of cross-validation (CV), whereas sensitivity analysis identifies the winding angles that dominate structural performance.
The tri-objective optimisation problem of minimising mass while maximising nonlinear collapse loade and KDF is addressed using two complementary multi-objective strategies, independently: the Bayesian optimisation (BO) framework, which exploits surrogate uncertainty to accelerate convergence, and the Non-dominated Sorting Genetic Algorithm II (NSGA-II), which promotes global exploration through evolutionary search. Feasibility constraints on mass and imperfection robustness were enforced, and Pareto-optimal sets were benchmarked using convergence histories and hypervolume indicators, revealing the superior sample efficiency of BO and the strong exploration capacity of NSGA-II.
These results confirm that the adoption of realistic imperfection statistics, uncertainty-aware surrogates, and optimisation strategies allows a unique, imperfection-tolerant methodology for VAFW cylinders. In fact, the framework provides quantitative insight into the tradeoffs among lightweight efficiency, buckling strength, and robustness, thus offering design guidelines for next-generation aerospace structures.