Novel hybrid machine learning and multi-material topology optimization for composite compliance structures

The inverse multi-material topology optimization (MMTO) for obtaining the optimal composite materials under minimizing compliance condition is a challenge for computational burden and capable methods of MMTO. The computational burden of inverse MMTO can be enhanced by the hybrid machine learning methods. In this current work, the artificial intelligent model (AIM) given by the various machine learning models named artificial neural network (ANN), extreme learning machine (ELM), radial basis neural network (RBNN) and deep neural network (DNN) are discussed for inverse MMTO of composite structures subjected to compliance constraints. The data for training AIMs are computed based on moving iso-surface threshold (MIST) topology optimization method which employ twelve cases of volume fractions (Vf) applied in four compliance composite structures. The results of AIM are compared with statistical surrogate models of response surface method and Kriging through several comparative matrices for evaluating tendency, accuracy and agreement of models. Results indicated that the predictions of optimal volume fraction under the minimum compliance condition using DNN, ANN, and RBNN outperform those of ELM and statistical surrogate models. Moreover, the DNN demonstrates superior performance compared to other artificial intelligence models in the MMTO of composite compliance structures.