Neural Network Surrogate-Based Optimization of Electromagnetic Coils for Magnetic Endoscopy

Magnetic actuation offers a minimally invasive approach for active navigation of capsule endoscopic systems, overcoming the limitations of passive wireless motion. This paper presents a surrogate-assisted optimization framework for the design of a triad electromagnetic coil array exhibiting ۱۲۰° rotational symmetry. The coil geometry, consisting of stepped windings and a ferromagnetic core, is parameterized by ten variables subject to geometric and functional constraints. High‑fidelity ۳D magnetostatic FEM simulations are used to generate ۶,۵۶۳ feasible coil configurations within the manufacturing bounds. Independent single‑output multilayer perceptron (MLP) surrogates are trained using an ۸۵/۱۵ data split and z‑score normalization to predict magnetic field magnitude, axial field gradient, and coil mass at a clinically motivated distance of z = ۱۵۰ mm. The surrogate models achieve high accuracy (R² ≥ ۰.۹۲) and replace direct FEM evaluations in a constrained optimization solved using Genetic Algorithm (GA), Particle Swarm Optimization (PSO), and Differential Evolution (DE). Among them, DE provides the most robust convergence and highest objective values. The final optimized coil architecture is validated through full ۳D FEM analysis, confirming strong field intensity and symmetry without flux saturation, thereby demonstrating the efficiency and reliability of the proposed NN‑assisted optimization framework.

کلیدواژه ها:

Magnetic Actuation ، Electromagnetic Coil Optimization ، Surrogate Modeling ، Neural Network–Assisted Design ، Capsule Endoscopy