A neural network-based approximation model for HTCC composite laminates.

摘要

In many aerospace applications of composite structures, mechanically induced out-of-plane deformations subjected to in-plane loads are often demanded. Hygrothermal curvature-stable coupling (HTCC) laminate is such a laminate possessing the unique mechanical and hygrothermal properties. This type of laminates consists of a sequence of sublaminate layup, each with given relations among their fiber orientations. For such laminates, the mechanical coupling properties are preserved without inducing hygrothermal curvature instability. An approximation model of the response functions of HTCC laminates based on the backpropagation neural network (BPN) has been developed. The response functions of the HTCC laminates are taken as the tip displacement and twisting angle of a cantilever composite beam subjected to a transverse load and a twisting moment. The training data of BPN are obtained from an analytic analysis and the results have been verified using a finite elements analysis. Six variables have been designed to define the HTCC laminates. Among the 6 input components, 4 are discrete integer variables. Because the response is highly sensitive to the integer inputs, a composite BPN composed of 64 sub-neural networks have been developed. It is found that such a composite BPN can simulate the responses very accurately.