Simultaneous optimization of fiber paths and geometric dimensions for fiber composite laminates using double neural network-surrogate model and genetic algorithm

CFRP is characterized by its anisotropic properties. Therefore, further structural weight reduction might be achieved by introducing the fiber direction as a design variable. For this study, steering fiber laminates defined by a Bezier curve are applied to a skin panel of CFRP stiffened panels. Then, simultaneous optimization of the fiber paths and stringer dimensions is performed to minimize the panel weight. We propose a new surrogate model that combines two neural networks and Latin hypercube sampling. Stiffness, manufacturability, and buckling load are introduced as constraints. Also, optimization is performed using the surrogate model and a genetic algorithm (GA). Furthermore, the stringer dimensions and fiber directions are optimized for straight fiber laminates. Comparison of steering fiber laminates and straight fiber laminates demonstrated that the introduction of steering layup into skin plates can reduce the stringer weight by up to 25.6%. An effective method for designing the geometric dimensions and layup path is provided.