A GAN-Based Intelligent Design Algorithm for Diaphragm Wall with Inner Supports

The integration of artificial intelligence with engineering design is one of the future directions of civil engineering. Thus, a generative adversarial network (GAN) based intelligent design algorithm DESDWGAN is proposed, which for the first time enables the use of artificial intelligence to generate preliminary design schemes for diaphragm walls with inner support. The algorithm first defines a complete set of mapping relationships between rock, soil, supporting structures, and colors, mapping the sectional drawings of foundation pits and support into semantic segmentation images. Secondly, the structure of the algorithm is designed and the loss function is constructed. Its core consists of a generator and a discriminator network composed of deep convolutional neural network. Through the confrontation, the similarity between the generated design and the real design is gradually improved. Finally, the algorithm is trained and tested based on the data set of real examples, with the influence of iteration times, Huber and wall length loss function weight being observed. The similarities and differences between the intelligent design and the real design on the test dataset are also analyzed. After training, the DESDWGAN is capable of capturing the patterns of existing designs, and can complete new designs for different conditions. It can be considered that the algorithm has preliminarily formed intelligent design capabilities for diaphragm wall with inner supports.