A New BCC Lattice Modelling Approach Through a Neural Network Optimisation

Lattice structures are defined by the organised repetition of a unit cell. These complex designs are widely used in engineering due to their lightweight nature and tailored mechanical properties. This study presents a novel 1D numerical method for modelling the standard Body-Centered Cubic (BCC) unit cell, a commonly used lattice structure geometry due to its geometry simplicity. Mechanical analyses were carried out using Ansys 2023 R2. A customised ANSYS Parametric Design Language (APDL) routine was developed to assign specific area and inertia properties to each beam element, independent from the nominal value of the strut diameter. The equivalent engineering constants were evaluated through uniaxial compression and in-plane shear tests. A neural network optimisation technique was then employed to optimise the 1D model’s parameters and improve its accuracy. The neural network was trained using high-fidelity 3D FEM results as the reference data, enabling the identification of the most suitable parametric values for various strut diameters. The proposed 1D method aims to address the computational complexity associated with traditional 3D FEM analyses while maintaining accurate predictions of the mechanical response. By representing the lattice struts as beam elements in the 1D model, the computational cost can be significantly reduced, making it more feasible to analyse large-scale lattice structures or perform extensive parametric studies involving different design configurations and loading scenarios. This formulation has been tested by comparing numerical results coming from standard 3D and 1D elements on a variable span 3-point bending test. The accuracy of results and time-saving demonstrate the validity of the proposed procedure.