Synthesising evolvable smart manufacturing scenarios

Smart manufacturing contexts have been experiencing an increasing complexity over the years, leveraged by higher computational power (CPU/GPU), increasing speed connections (5G/6G), AI transformation from use case to mass adoption and robotic systems enhancements. Research has been focusing on architectural issues to improve disposition of the components involved within a smart manufacturing scenario and find the most effective configuration to provide the most efficient production environment. We introduce in this paper an improved version of our ”maze generator”, a navigation scenario generator that uses a machine learning approach from the Evolutionary Computing branch called Grammatical Evolution (GE) to automatically generate different scenarios with different configurations. GE takes advantage of a BNF grammar (rules describing the experiment) to define the search space, and genetic operations (crossover, mutation, selection, etc) to provide novelty to the solutions found. This not only enables the possibility to test autonomy, self-sufficiency and performance on a simplified model, but also to determine levels of difficulty to test the simulated navigation model under specific conditions.