Neural network-based structural optimization of tow-steered composite panels accounting for polymorphic uncertainty

This paper introduces polymorphic uncertainty modeling in the context of structural optimization for tow-steered composite panels. The approach combines random, interval, and fuzzy variables, into advanced models that simultaneously account for both aleatory (inherent variability) and epistemic (lack of knowledge) uncertainty, allowing for more realistic design. Geometric imperfections are represented using random fields, while uncertainty in the fiber path is modeled with fuzzy functions. To reduce the high computational cost in multi-objective design optimization under polymorphic uncertainty, a hierarchical surrogate modeling strategy based on artificial neural networks (ANNs) is presented. Two interconnected ANNs are constructed: the first predicts the buckling load of random imperfect panels, and the second estimates the stochastic output quantities. Fuzzy cumulative distribution functions and a Pareto front are computed to visualize the results. Structural performance and robustness measures are evaluated for a flat and curved tow-steered panel example within a multi-objective buckling design framework under polymorphic uncertainties. The results are compared with deterministic, aleatory-only, and epistemic-only optimization approaches. It is shown that the optimal stacking sequences are similar when maximizing performance, whereas the optima for robust design depend on the uncertainty characteristics.