A Deep Learning Approach to Automated Structural Engineering of Prestressed

Document Type


Publication Date



In this paper, an implementation is presented of deep learning on the structural engineering of prestressed concrete members. Prestressed concrete beams and slabs are essential structural members supporting the floors of buildings, yet their optimum design is still a challenge for engineers as they struggle to design sections that adhere to serviceability and economical needs. Recently, the advancement of artificial neural networks has managed to propose more optimum solutions to general engineering applications with ease. Deep learning and grid search available hyperparameters can be utilized to predict optimum prestressing of members, without the need for structural engineers to produce countless analysis and design iterations. A simple prestressed beam is presented as an initial example to show the viability of neural networks against the traditional approaches. Two industrial examples of a continuous beam and a slab-beam type are added to demonstrate scalability of the design.