Exploring the composition-structure-property space for beam lattice archimats using machine learningRoom 2
Owing to the diversity of possible properties, beam lattice archimats (BLA) have a great potential as structural materials. The possible range of their structural characteristics can be broadened by employing beams of different stiffness in the architecture of lattice structures. Furthermore, metamaterials exhibiting such effects as asymmetric response to tension and compression, auxetic properties, etc., can be designed in this way. Here we present our recent results on designing lattice materials through the appropriate choice of the architecture of a set of beams. In this approach, beams of different kinds are blended together to produce the desired outcomes in terms of the properties of the archimat. The composition and structure of such materials is determined by the characteristics of the beam-like elements they consist of, their mutual arrangement, and the nature of their junctions. For a given set of parameters specifying the system, calculating the physico-mechanical properties of BLA is facile and can be conducted using the matrix approach. By contrast, solving an inverse problem, i.e. determining the composition and structure of an archimat that provide it with pre-defined properties, is incommensurably more complex. It involves identification of a region within the parameter space whose dimensionality may be in the tens. We propose a way to solving this problem by employing methods of machine learning, specifically, cluster analysis, and apply it to the multi-dimensional composition-structure-property space for BLA.