基于生成对抗网络的叠合板拆分

为了解决叠合板拆分受设计师与预制厂的差异容易出现难以设计的问题，提出了基于pix2pix算法构建叠合板拆分预测模型，通过输入支座图进而生成对应的叠合板拆分图，从而实现对叠合板快速拆分设计。探讨利用规则约束下机器学习的方法和理念，优化叠合板拆分方案，为预制厂提供了模型训练方法。为评估模型，建立基于拆分尺寸、方向与顺序掌握的评价体系。结果表明，训练后的模型可在2s内快速绘制叠合板拆分图，且模型生成思维与人工设计思维基本吻合。

Composite plate splitting based on generative adversarial network

Aiming at the problem that the separation of composite slabs is easy to be difficult to design due to the difference between designers and prefabricated factories, a prediction model of composite slab separation based on pix2pix algorithm is proposed. By inputting the bearing diagram, the corresponding composite slab separation diagram is generated, so as to realize the rapid separation design of composite slabs. The method and idea of machine learning under rule constraint are discussed to optimize the splitting scheme of laminated board and provide model training method for prefabricated factory. In order to evaluate the model, an evaluation system based on split size, direction and order is established. The results show that the trained model can quickly draw the composite plate split graph within 2 s, and the model generation thinking is basically consistent with the manual design thinking.