ScholarWorks@경상국립대학교

초록

A neural network (NN)-based approach holds promise for the stacking sequence optimization (SSO) of a carbon-fiber reinforced plastic (CFRP) double roller, though it remains challenging due to the large number of input parameters in the SSO. To address this, we devised an iterative NN-based SSO strategy with a layer information compressing (LIC) technique to efficiently determine the adequate stacking sequence. In the proposed strategy, we derived the LIC technique to reasonably compress the input parameters using the information in the stacking sequences. Next, based on the squeezed input parameters, the NN for SSO is iteratively trained based on sample data. Here, we suggested criteria for moderately constructing the sample data by utilizing very small amounts of data while considering both local and global observations. As a result, the quality of the NN can be significantly enhanced by the reasonably reduced input parameters, even without an additional deliberation of the output data. Furthermore, based on the sample data in this work, the iterative procedure in the NN-based SSO could be effectively accelerated without a heavy computational burden. Consequently, the proposed strategy can efficiently yield a highly accurate stacking sequence for the CFRP double roller, as demonstrated through various numerical examples.

키워드