PhD Oral Exam - Seyed Alireza Moezi, Mechanical Engineering

When studying for a doctoral degree (PhD), candidates submit a thesis that provides a critical review of the current state of knowledge of the thesis subject as well as the student’s own contributions to the subject. The distinguishing criterion of doctoral graduate research is a significant and original contribution to knowledge.

Once accepted, the candidate presents the thesis orally. This oral exam is open to the public.

Abstract

Minimally invasive procedures in the vascular, cardiac, and bronchial systems remain constrained by the limitations of current tools. Navigating the body’s extremely narrow and highly curved anatomical pathways demands a level of precision that conventional catheters and guidewires often cannot deliver, leading to unreliable steering and elevated patient risk. Magnetoactive Soft Continuum Robots (MSCRs) offer a transformative alternative to overcome these inherent limitations. Composed of elastomeric matrices embedded with hard magnetic particles, these millimetre-sized magnetoactive soft robots can undergo large, fully reversible deformations under wireless magnetic actuation, allowing them to conform to complex anatomies and navigate with unprecedented precision. Despite this potential, widespread clinical adoption has been hindered by the absence of a unified framework capable of integrating accurate and computationally efficient simulation of MSCR mechanics, robust real-time control in dynamic environments, and autonomous navigation tools tailored to the constraints of human anatomy. To this end, the present dissertation establishes a comprehensive end-to-end framework addressing these barriers. The work spans the full development pipeline, ranging from fabricating and characterizing tailored adaptive magnetic composites to developing accurate nonlinear magneto-mechanical and viscoelastic dynamic models capable of capturing large deformations and time-dependent behavior. Building on these models, the dissertation introduces advanced model-free and AI-driven model-based control strategies that enable robust, real-time trajectory tracking in dynamic biological environments. The framework is validated through hardware-in-the-loop experimental setups that assess performance under simulated physiological conditions and is further supported by a state-of-the-art in-vitro autonomous dual-arm robotic platform enabling highly accurate navigation in complex anatomies.

This research begins with the fabrication and comprehensive characterization of MSCRs using composite magnetoactive elastomers. It then develops a versatile family of two- and three-dimensional quasi-static and dynamic nonlinear models that capture MSCR multiphysics behavior under uniform and nonuniform magnetic fields in ambient and fluidic environments. These modeling frameworks include: (1) a reduced-order finite element (FE) model employing a switching-trajectory, piecewise-linear architecture that predicts large-deformation response with approximately 1–3% error relative to the tip-deflection amplitude under 5 mT excitation while achieving a 792-fold improvement in computational speed; (2) standard and fractional-order Kelvin–Voigt magneto-viscoelastic formulations that capture rate-dependent and hysteretic behavior, including under biofluid flow; and (3) full 3D quasi-static and dynamic models incorporating magnetic torques, body forces, gravity, axial strain, and hydrodynamic drag to estimate deformation under nonuniform fields with high accuracy. Across all validations against analytical benchmarks and experimental tests, the developed models demonstrated excellent agreement, with errors in the three-dimensional magneto-mechanical model remaining below 3% under uniform magnetic fields and below 1.5% under nonuniform magnetic fields.

Building on these models, several closed-loop control strategies were developed. A model-free feed-forward interval type-2 fractional-order fuzzy-PID controller achieved precise tip-deflection tracking under uniform fields up to 3 Hz, reducing error by 29% at 0.5 Hz and by 80–90% at 2 Hz, while lowering control effort and eliminating chattering. For fluidic navigation, a deep-reinforcement-learning fractional-order sliding-mode controller reduced error by more than 40% at 1160 mL/min and by 33% at 2190 mL/min, with up to 90% lower control effort. Under nonuniform fields, a feed-forward PID (FFPID) strategy maintained milliradian-level accuracy in ambient conditions and exhibited only 2–12% degradation at flow rates up to 2350 mL/min, yielding more than a 75% reduction in tracking error compared with a classical PID controller.

To translate these developments into autonomous navigation, a dedicated dual-arm robotic platform integrating magnetic actuation and deep-learning-based stereo-vision perception was constructed. Autonomous in-vitro experiments in 3D-printed vascular and bronchial phantoms, using the developed deep-learning-based fractional-order sliding-mode method and blood-mimicking flow up to 5000 mL/min, demonstrated reliable navigation with sub-3-mm tracking error. Overall, this research study delivers a unified and experimentally validated framework, establishing a rigorous foundation for the use of MSCRs at clinical scale for minimally invasive interventions.