Quantitative structural uncertainty analysis of composite honeycomb sandwich using a feedback neural network

Abstract

The quantification of margins and uncertainty (QMU) method was applied using a neural network to inversely determine the strength boundary of a composite honeycomb sandwich structure considering the random uncertainty of its elastic modulus, shear modulus, tensile strength, compressive strength, and shear strength in the warp and fill directions, and panel thickness. An application framework for the deterministic quantitative analysis of strength was proposed based on margin design theory, combining the strength problem with the quantitative analysis of uncertainty. First, the failure criterion of the structure was modeled based on the progressive damage principle using ABAQUS. Next, Latin hypercube sampling was applied to simulate the stochastic uncertainty of the structural parameters, and the resulting sample sets were input into ABAQUS to calculate the corresponding structure lateral bearing capacities. These results were considered to collectively represent the actual strength boundary of the composite structure. A feedback neural network was subsequently trained on these ABAQUS data to obtain the functional mapping relationship between the parameters and capacity. The capacity obtained using this relationship was multiplied by an empirical safety factor to serve as the theoretical strength boundary. Finally, the QMU method was applied using cumulative distribution functions of the theoretical and actual strength boundaries to calculate the confidence factor and thereby evaluate the reliability of the composite honeycomb sandwich structure.