Structural design of reinforced concrete buildings based on deep neural networks

•Correct initial prediction of wall dimensions can accelerate the design procedure.•Deep neural networks can provide an adequate regression model for wall dimensions.•Existing architectural and engineering plans are used to define a regression model.•Geometrical and topological features provide good input variables. In shear wall building design, the initial process requires the interaction between the architectural and structural engineering groups to define the adequate wall layout, usually done with a trial-and-error procedure to fulfill architectural and engineering needs, slowing down the design process. For the engineering analysis, first, the wall thickness and length are required to check the building deformation limits, base shear strength, among other parameters. For this reason, the present investigation develops a structural design platform for reinforced concrete wall buildings that uses a deep neural network to predict the wall’s thickness and length based on previous architectural and engineering projects. The study includes, in the first place, the surveying of the architectural and engineering plans for a total of 165 buildings constructed in Chile; the generated database has the geometric and topological definition of the walls and the slabs. As a second stage, a model was trained for the regression of the wall segments’ thickness and length, making use of a feature vector that models the variation between the architectural and the engineering plans for a set of conditions such as the thickness, connectivity (vertical and horizontal), area, wall density, the distance between elements, wall angles, foundation soil type, among other engineering parameters. The regression model results in terms of R2-value are 0.995 and 0.994 for the predicted wall thickness and length, respectively, proving to be a reliable method for the initial engineering wall definition.