Structural Engineer / Programmer

Deep learning predicted column dimensions.

A feed forward neural networks predicts the required dimensions of square columns.

The neural network has trained on a data set of various cross section calculations. The allowed buckling length of these cross sections were determined with the output of these calculations. The calculations are based on the design requirements described in the NEN-1992.


The data set was acquired by making the following assumptions:

fck = 30

φeff = 2

Kr = 1

MEd is constant over the length of the column

The bending moment due to the second order force distribution was derived by MRd - MEd. The buckling length for each configuration was determined with M2.

Two sided rebar reinforced columns.

At high axial forces the computed MRd is low because of crushing of the concrete. This was probably not learned by the network. The range is approximately 7000 kN for the axial force and 500 kNm for the bending moment.