Design of Anisotropic Patterns in a Continuum Tubular Robot Using Metaheuristics Optimization with a Deep Neural Network

Abstract

Concentric-tube robots have shown potential for minimally invasive surgery (MIS) procedures. However, a major challenge in their practical usage is their instability, which is caused by a high value of bending-to-torsional stiffness ratio (EI/GJ) and potentially leads to the risk of tissue rupture. This study aims to optimize patterned tubes for minimizing EI/GJ while conforming to minimum compliance constraints and geometric restrictions. We propose a novel optimization method using metaheuristic optimization accelerated by a deep-neural-network (DNN)-based surrogate model. First, topology optimization was used to determine the general shape of the pattern to maximize torsional stiffness. Then, the surrogate model based on DNN was constructed and trained using 855 datasets generated from finite element analysis (FEA) for pattern design variables. Finally, optimized pattern parameters were determined for reducing the instability of the tubular structure using metaheuristic-based optimization. Accordingly, optimal designs were obtained and the bending and torsion tests of 14 cases with two optimal cases were validated in both experiments and FEA. Using the proposed optimization method, we found an optimal pattern parameter set that showed the lowest EI/GJ of 0.061.

Keywords: design optimization, metaheuristics, deep neural network, surrogate model, surgical robotics, steerable catheters/needles