A Novel Method to Accelerate the Solution of Compliance Using Deep Learning for Topology Optimization

There are many successful advances in using deep learning techniques to accelerate topology optimization design and solve the problem of high computation costs. However, the finite element method (FEM) is still needed to calculate the compliance for the structure output by the training task that uses convolutional neural networks to accelerate the calculation of optimal structures, which leads to more time spent on model training. Therefore, a deep learning model is proposed to accelerate the solution of compliance instead of FEM. To improve the accuracy of the model prediction, we prepare 50,000 samples which are generated by the solid isotropic material with penalization (SIMP) and build the developed model using the appropriate number of network layers. The results show that the error rate of the predicted compliance is less than 0.01 and the developed model is high-precision. In addition, the training efficiency using the developed model is 3 times higher than that using FEM during the model training.