Machine learning based lattice generation method derived from topology optimisation
Jier Wang, Ajit Panesar
Abstract
Lattice structures are widely being employed in various lightweighting and multifunctional applications. With the developments in Additive Manufacturing (AM), lattice structures can now be fabricated with limited manufacturing constraints, which facilitates the design of lattice structures to become more flexible. In this paper, a novel lattice generation strategy is proposed to design graded lattice structures with the assistance of Machine Learning (ML). A Neural Network (NN)-based inverse lattice generator is trained to output lattice unit cells from the input of target mechanical properties. The proposed ML-based lattice generation method utilises the density distribution and stress field of low-resolution Topology Optimisation (TO) design to inform the inverse generator and produce lattice cells. The efficiency and efficacy of this method and the influence of cell types are demonstrated with the MBB-beam design case. Furthermore, the developed ML-based method is also applicable to multiple cell types.