Exploring the Adjacent Possible in Engineering Design via Graph Operation Programs (June 2026)

This presentation was given by Babis Peteinarelis at the 2026 IEEE World Congress on Computational Intelligence (WCCI), the world’s leading event for Computational Intelligence research and applications (21–26 June 2026) in Maastricht.

Abstract – Nature-inspired algorithms and mechanisms have been used for many years to assist the design of engineering structures. However, in most fitness landscapes in engineering, Evolutionary Algorithms converge and the result is accepted as-is. Design tasks on the other hand often prioritise novelty and diversity in solutions, hoping for innovation, not simply optimisation. Needs that are satisfied by Generative Design methods are usually at the cost of optimality, transparency or feasibility. Inspired by Kauffman’s Adjacent Possible, this paper attempts to reimagine the popular concept in contexts that evolve structural paradigms, aiming to combine the glass-box rationale of design with the generative power of Evolutionary Methods. The presented formulation uses the graph space to represent structures and the formalism of Graph Automata to express transition rules that can reach adjacent structures based on a set of elementary graph operations. Sequences of such graph operations (programs) are evolved to reveal a set of fit adjacent structures. This evolutionary logic is demonstrated on a Multi-Objective Topology Optimisation scenario where a seed truss structure experiences environmental forces. The evolved programs transform the seed structure into a variety of truss designs that respond to localised strain in different ways in order to address trade-offs between the objectives. The paper illustrates the method on truss (chassis) design and discovers a range of Pareto-optimal adjacent designs within a program length radius, maintaining the main design intent.
Index Terms—Evolutionary Computation, Generative Design, Engineering Design, Graph Operations, Graph Automata.