Accelerating robotic assembly parameter optimization through the generation of internal models

As time progresses, it is likely that the demand for the ability of robotic assembly solutions to self-tune will steadily grow, and companies will invest more in automation technologies for manufacturing. In this study we present a method that employs computational learning to generate internal models for online optimization acceleration of a genetic algorithms approach of exploring a parameter space for process optimization. By randomly sampling the ldquogene poolrdquo parameter space, it is possible to successfully generate a mapping of high-dimensional input parameters to their respective resulting performances in the presence of noise, and then use this same map to improve the performance of the learning process by acting as a predictive filter for selectively choosing the child gene sequences most likely to produce improved assembly results. Results are given that demonstrate the advantages of internal model building as it relates to both virtual and physical trials.