REAL-TIME TOPOLOGY OPTIMIZATION DESIGN UPON PYRAMID-ATTENTION-U-NET DEEP LEARNING MODEL

Topology optimization design provides innovative structures with excellent thermal, mechanical and acoustic performance for modern engineering structures. Moving morphable components (MMC) have gained popularity due to their unique display topology optimization. MMC achieves boundary evolution through migration and superposition of a series of moving morphable display components, resulting in structural optimization. Here instead of the original linear skeleton components with secondary thickness changes, oval initial components are used for topology optimization, which not only reduces the complexity of the variable design but also shows a better fitting effect of boundary evolution, resulting in reduced computation time. However, as the number of initial components increases, the intermediate iterative computation becomes increasingly time-consuming, furthermore, initial components of different parameters generate large differences in the topological configuration. To obtain the optimal topology in an accurate and real-time manner, Pyramid-Attention-U-Net (PA-U-Net) deep learning model is proposed to improve the optimization design and to avoid intermediate iterative computations. According to the results, the method can not only obtain the optimal topology of the initial components under various parameter conditions in an accurate and timely manner in a negligible computing time, but it can also achieve an accuracy of 90.89%, which is better than other deep learning network models. Meanwhile, in the optimization design of large engineering structures, the organic combination of deep learning with topology optimization has many prospects.