Magnetic manipulation has the potential to recast the medical field both from an operational and drug delivery point of view as it can provide wireless controlled navigation over surgical devices and drug containers inside a human body. The presented system in this research implements a unique eight-coil configuration, where each coil is designed based on the characterization of the working space, generated force on a milliscale robot, and Fabry factor. A cylindrical iron-core coil with inner and outer diameters and length of 20.5, 66, and 124 mm is the optimized coil. Traditionally, FEM results are adopted from simulation and implemented into the motion logic; however, simulated values are associated with errors; 17% in this study. Instead of regularizing FEM results, for the first time, artificial intelligence has been used to approximate the actual values for manipulation purposes. Regression models for Artificial Neural Network (ANN) and a hybrid method called Artificial Neural Network with Simulated Annealing (ANN/SA) have been created. ANN/SA has shown outstanding performance with an average R2, and a root mean square error of 0.9871 and 0.0153, respectively. Implementation of the regression model into the manipulation logic has provided a motion with 13 μm of accuracy.
Keywords: ANN; ANN/SA; deep learning; electromagnetism; magnetic actuator; magnetic manipulator.