Automated clash resolution for reinforcement steel design in precast concrete wall panels via generative adversarial network and reinforcement learning

Pengkun Liu, Hongtuo Qi, Jiepeng Liu, Liang Feng, Dongsheng Li, Jingjing Guo

Abstract: The increasing adoption of precast concrete elements (PCEs) is evident in civil infrastructure projects. The structural integrity of prefabricated structures relies on the quality of connections between adjacent components, necessitating the design of reinforcing steel bars (rebars) in PCEs. Dealing with PCEsand the complex design codes for rebars' arrangement can be labor-intensive, impractical, and error-prone when using traditional computer software due to their irregular shapes. This often results in frequent rebar clashes on construction sites. On the other hand, Building Information Modeling (BIM) technology, which is widely employed for structural design in the industry, offers potential solutions to these challenges. Several studies have proposed clash resolution strategies for moving components using optimization algorithms; these strategies are only suitable for regular reinforced concrete (RC) structures. The optimized path of rebars cannot be adjusted to avoid obstacles in PCEs. Due to strict design codes and large dimensions, existing studies do not possess the learning knowledge from design codes or drawings needed to generate clash-free rebar arrangements for real-world PCEs automatically. To overcome this limitation, we present a BIM-based framework that utilizes Generative Adversarial Network (GAN) and Deep Reinforcement Learning (DRL) to automatically generate clash-free rebar designs in prefabricated concrete wall panels (PCWPs). Our method employs GAN to learn from designers’ experiences from existing design drawings and generate 2D preliminary rebar designs, which are then transformed into mesh environments for DRL. The DRL engine incorporates state, action, and rewards designed according to buildability constraints and design codes. We conduct extensive experiments on real-world PCWPs to assess the efficacy of the proposed method. The results indicate that our framework can reduce engineering time for rebar designs by up to 80% compared to manual design, while achieving clash-free designs that adhere to design codes.