💻 computer science

Model-Informed Joint Material-Structural Optimization of Hard-Magnetic Soft Materials

This paper presents a model-informed framework that unifies multiple constitutive models to accurately predict the behavior of hard-magnetic soft materials and enables their simultaneous structural and material optimization for achieving prescribed deformation responses in soft robotics and adaptive systems.

Original authors: Ian Galloway, Prashant K. Jha

Published 2026-07-17

📖 3 min read☕ Coffee break read

CC BY 4.0

Original authors: Ian Galloway, Prashant K. Jha

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine a world where robots aren't made of stiff metal gears and rigid joints, but are instead soft, squishy, and alive with movement, like jellyfish or octopuses. This is the realm of soft robotics, a field that dreams of machines that can squeeze through tiny cracks, hug delicate objects, or even swim inside the human body without causing harm. To make these robots move without wires or motors, scientists use smart materials that change shape when triggered by invisible forces, like a magnetic field. Think of it like a puppet whose strings are pulled not by a hand, but by a magnet.

One particularly exciting type of these materials is called hard-magnetic soft materials. You can picture them as a soft, rubbery sponge (the "soft" part) that is packed with tiny, super-strong magnets (the "hard" part). When you bring a big magnet near them, the tiny magnets inside want to line up with the big one, and because they are stuck inside the rubber, they drag the rubber along with them, causing the whole object to bend, twist, or stretch. The big question for engineers is: How do we design these squishy magnets so they move exactly the way we want? It's a tricky puzzle because the material's stiffness changes depending on how many magnets you pack inside, and the way the rubber stretches changes how the magnets behave. If you get the recipe wrong, your robot might just flop over instead of doing a backflip.

This paper is like a master chef's guide to perfecting that recipe. The authors, Ian Galloway and Prashant K. Jha, set out to solve two main problems. First, they wanted to figure out the best mathematical "rulebook" to predict how these materials will behave. There are many different theories about how adding magnets to rubber changes its stiffness, and the authors tested seven different theories against real-world data to see which one was the most accurate. They found that while the type of rubber formula didn't matter much, the rule for how magnets stiffen the rubber was crucial, especially when the magnets were placed right where the bending happened. They discovered that the Mooney relation was the best rulebook to use.

Second, they built a powerful new design tool—a "digital sculptor"—that can simultaneously figure out three things at once: where to put the structural material (so the robot is strong enough), how many magnetic particles to pack into each spot (to control how strong the pull is), and which direction the tiny magnets should face (to control the direction of the bend). Using this tool, they created some very clever designs. For example, they designed a wheel that spins much faster than a standard one just by rearranging the magnets inside, and a beam that can bend down under a heavy weight and then magically snap back to straight when a magnetic field is turned on. The paper shows that by treating the material composition and the shape as one big, connected puzzle, we can create soft robots that move in surprising and highly efficient ways, all while using a computer model that has been carefully checked against real experiments to ensure it's telling the truth.

Technical Summary: Model-Informed Joint Material-Structural Optimization of Hard-Magnetic Soft Materials

Problem Statement
Hard-magnetic soft materials (hMSMs) are composites consisting of hard-magnetic particles embedded in a compliant elastomeric matrix. They enable contact-free, field-driven deformation suitable for soft robotics and adaptive structures. However, the predictive design of hMSMs faces two interconnected challenges:

Constitutive Modeling: Accurate prediction requires effective structure–property relations that describe how magnetic particle volume fraction modifies the effective elastic moduli. Existing literature offers various models (rigid-inclusion, self-consistent, constrained-kinematics) that can predict significantly different stiffness values for the same particle concentration. Furthermore, the choice of strain-energy density function may influence numerical robustness and predicted deformation.
Joint Optimization: Optimal design requires the simultaneous control of structural density, magnetic particle distribution, and remanent magnetization direction. Prior topology optimization approaches for hMSMs often treat magnetic material presence as a discrete choice (magnetized vs. non-magnetized) rather than a continuous volume fraction field that modulates both stiffness and actuation strength. Additionally, coupling these fields physically is difficult because magnetic variables are only meaningful where structural material exists.

Methodology
The authors develop a unified framework addressing both constitutive model selection and joint optimization.

Unified Constitutive Framework: The authors recast classical rigid-inclusion relations (Guth, Mooney, Kerner), Hill's self-consistent theory (specialized for rigid inclusions and near-incompressible matrices), and constrained-kinematics models (Lopez-Pamies, Alshammari) into a common effective shear-modulus form, $G(\phi)$ , where $\phi$ is the particle volume fraction. This yields seven candidate shear-modulus relations. These are combined with three strain-energy density functions (two compressible neo-Hookean forms, NH1 and NH2, and the St. Venant–Kirchhoff model, SVK) to create 21 candidate constitutive models.
Model Sensitivity Analysis: Numerical benchmarks (cantilever beams, wheels, grippers, and 3D beams) are used to assess the sensitivity of predicted deformation to the choice of strain-energy function versus the choice of $G(\phi)$ .
Constitutive Model Selection: Experimental uniaxial stress–strain data from Garcia-Gonzalez et al. (2021) are used to select the most representative $G(\phi)$ relation. Two metrics are employed: (1) relative error against extracted shear-modulus values, and (2) relative root-mean-square error (rRMSE) against full true stress–strain curves.
Joint Material–Structural Optimization Formulation: A topology optimization framework is developed where three design fields are parameterized: structural density ( $\rho$ $ρ$ ), magnetic particle volume fraction ( $\phi$ $ϕ$ ), and remanent magnetization direction ( $\theta_{Br}$ $θ_{B r}$ ).
- Coupling Mechanism: To ensure physical consistency, the magnetic energy and effective moduli are interpolated such that magnetic contributions vanish in void regions ( $\rho \approx 0$ ). The magnetic particle volume fraction $\phi$ acts as a continuous field that scales both the magnetic interaction energy and the effective shear modulus via the selected $G(\phi)$ relation.
- Implementation: The framework utilizes the FEniTop library and FEniCSx for finite-element implementation, employing a SIMP-type penalization for density, Helmholtz filtering for all fields, and an adjoint-based gradient computation for the Method of Moving Asymptotes (MMA) optimizer.

Key Results

Constitutive Sensitivity: The strain-energy density function (NH1, NH2, SVK) has a relatively small effect on predicted deformation for the actuation problems considered. In contrast, the choice of the effective shear-modulus relation $G(\phi)$ significantly affects deformation, particularly when magnetic material overlaps with highly deforming regions. When magnetic material is localized away from primary bending regions (e.g., tip-magnetized beams), model dependence is negligible.
Model Selection: Based on experimental data, the Mooney relation for $G(\phi)$ demonstrated the best overall agreement with both extracted shear-modulus values and full stress–strain responses. The NH2 strain-energy function was selected for its numerical robustness.
Optimization Performance: The joint optimization framework successfully produced non-intuitive designs for three distinct cases:
1. Rotational Actuator: Optimizing $\phi$ and $\theta_{Br}$ in a fixed wheel geometry increased the tangential displacement by 59.7% compared to an intuitive uniform design.
2. Translational Actuator: Optimizing $\phi$ and $\theta_{Br}$ in a fixed scissor-like geometry achieved targeted horizontal motion while suppressing vertical parasitic displacement.
3. Restorative Beam: Simultaneously optimizing $\rho$ , $\phi$ , and $\theta_{Br}$ created a beam capable of restoring its shape under opposing mechanical loads using magnetic actuation. The design naturally confined magnetic material to structurally occupied regions.

Significance and Claims
The paper claims to provide a systematic, model-informed route for designing hMSMs. Its primary contributions are:

A unified effective shear-modulus framework that allows direct comparison of diverse structure–property relations.
Evidence that effective shear-modulus relations are critical for predictive accuracy when magnetic and mechanical fields overlap, whereas strain-energy forms are less sensitive for the tested cases.
A joint material–structural optimization formulation that treats magnetic particle volume fraction as a continuous design variable coupled with structural density and magnetization direction, resolving the physical coupling issue in low-density regions through energy interpolation.
Demonstration that optimizing material composition and structural topology simultaneously yields superior performance over fixed-geometry magnetic design.

The authors note that while the framework is demonstrated on hMSMs, the computational structure is applicable to other field-responsive elastomers (e.g., dielectric, magnetorheological, or shape-memory composites) by substituting the relevant energy terms. The implementation is made available in the open-source CEADpx/top_optim repository to support reproducibility and extension.

More like this