Multi-material compositional pattern-producing networks for form optimisation

Abstract

CPPN-NEAT (Compositional Pattern Producing Networks and NeuroEvolution for Augmented Topologies) is a representation and optimisation approach that can generate and optimise complex forms without any pre-defined structure by using indirect, implicit representations. CPPN is based on an analogy to embryonic development; NEAT is based on an analogy to neural evolution. We present new developments that extend the approach to include multi-material objects, where the material distribution must be optimised in parallel with the form. Results are given for a simple problem concerning PV panels to validate the method. This approach is applicable to a large number of problems concerning the design of complex forms. There are many such problems in the field of energy saving and generation, particularly those areas concerned with solar gain. This work forms a first step in exploring the potential of this approach.