Indirect Encoding of Structures for Evolutionary Design

The use of Evolutionary Computations (EC’s) has become one of the primary methods in the field of automated design synthesis. The overwhelming majority of EC’s in use today use a direct encoding, where an individual is described by its gene string. This means that every engineering domain must create its own encoding scheme, making implementation in new fields difficult and slow. Additionally, direct encoding does not produce symmetry or modularity, unless these attributes are written into the encoding scheme ab initio. Direct encoding, however, is not the way that genetic information is used in biological evolution. In nature, each DNA string (genotype) is composed of instructions or rules on how an individual should grow and develop. This has provided nature with a wide array of evolved solutions, based on simple coding blocks. This paper presents a method of indirect (rules-based) encoding for an EC based on the biological analog of development: the gene string no longer describes the individual, but rather contains an instruction set on how to generate the individual. This allows for additional elements to be added by modifying the rule set, rather than re-composing the entire genetic structure. Indirect encoding also can have a built-in response to the environment, and is therefore able to adapt more readily to dynamic situations. Examples are shown to demonstrate the rule-set and its adaptability.

A Novel Synthesis Design Approach for Continuous, Inhomogeneous Structures

This paper presents a novel approach for the design synthesis of continuous inhomogeneous structures. The objective of this research is to mimic biological principles of growth and evolution in order to explore a set of novel design configurations identified by high complexity both in topology and mechanical properties. The ability to synthesize novel structures is explored from an engineering point view, where the use of inhomogeneous properties can increase the ability of a structure to support external loads and minimize weight. Based on the observation that biological structures are inhomogeneous, in the sense that different cells have different properties, an artificial environment has been created which models the biological growth procedure with cells that serve as building blocks of the structure. Cell differentiation is expressed only in the sense of mechanical properties. Each cell contains an identical artificial DNA sequence which is executed during the growth procedure and stops once the structure meets desired engineering requirements, such as supporting loads. The DNA contains sets of rules which are encoded as a gene string. A relatively simple DNA sequence can give rise to complex inhomogeneous structures; small changes in the rules can lead to a significantly different structures with different properties. The representation of these rules is ideally suited for evolution, which will be applied in the future to evolve rule-sets that grow and develop high-performance inhomogeneous structures.