A Dimensionally-Aware Genetic Programming Architecture for Automated Innovization

Automated innovization is an unsupervised machine learning technique for extracting useful design knowledge from Pareto-optimal solutions in the form of mathematical relationships of a certain structure. These relationships are known as design principles. Past studies have shown the applicability of automated innovization on a number of engineering design optimization problems using a multiplicative form for the design principles. In this paper, we generalize the structure of the obtained principles using a tree-based genetic programming framework. While the underlying innovization algorithm remains the same, evolving multiple trees, each representing a different design principle, is a challenging task. We also propose a method for introducing dimensionality information in the search process to produce design principles that are not just empirical in nature, but also meaningful to the user. The procedure is illustrated for three engineering design problems: two-bar truss design, welded-beam design and metal-cutting process optimization.