Abstract

We optimized the design parameters of three auxetic patterns to minimize the bending stiffness-to-torsional stiffness ratios (EI/GJ) in concentric-tube robots while maintaining minimal compliance. We proposed a deep neural network-metaheuristics optimization framework that incorporates active data generation close to the Pareto front and retraining of the surrogate model. Iterative procedure of data generation and surrogate model retraining yielded improved optimal solutions due to enhanced prediction accuracy of the surrogate model near the Pareto front, with minimal added data. The auxetic patterns optimized using our method achieved lower EI/GJ values compared to the recently reported design with identical tube specifications.