Prototyping Artificial Muscles with Silicone and Iron Powder

During the June-July 2023 Oberlin College Research Fellowship, I worked with Prof. Patrick Simen to test whether cast silicone–iron composites could support a low-cost, muscle-inspired soft actuator. Across eight weeks, I moved from literature review and CAD to mold fabrication, material-ratio experiments, and magnetic-field measurements. The result was a repeatable prototyping method and a clearer experimental question—not yet a finished artificial muscle.

Research question

Soft robotics replaces rigid links with compliant structures that can deform around people and irregular objects. The literature includes pneumatic systems, dielectric elastomer actuators, magnetically controlled elastomers, and many hybrid approaches (Hines et al., 2017; Chung et al., 2021). My project focused on a narrower fabrication question:

The working hypothesis was that a reusable mold and tunable iron loading could provide a practical rapid-prototyping route. The quantitative work therefore measured how core material, coil current, sensor position, and iron-to-silicone mass ratio changed magnetic flux density. The iron ratio is defined as iron-powder mass divided by the combined mass of silicone parts A and B.

From CAD to measurement

The workflow combined Rhino and SolidWorks for CAD, fused-deposition modeling for the molds, and a benchtop magnetic-field test. This use of additive manufacturing followed the low-cost, fast-iteration approach described in soft-robotics manufacturing reviews (Gul et al., 2018; Schmitt et al., 2018).

The experiments used 5 Ω test coils, a fixed jig for sensor spacing, and both north- and south-side readings where available. Prototype records also tracked the two silicone masses, iron-powder mass, final part mass, visible bubbles, and demolding defects.

Prototype gallery

What the measurements show

Core material dominated the coil response

The first comparison held the geometry at a 1 mm sensor distance and swept current from 0 to 0.28 A. Ordinary least-squares fits show an iron-core sensitivity of 51.77 ± 0.81 mT/A (95% fit interval), compared with 22.47 ± 0.53 mT/A for silicone and 22.13 ± 0.43 mT/A for air. At the highest recorded current, the iron-core field reached 14.61 mT.

The silicone–iron mixture did not improve monotonically

For each mixture, I fitted the slope of magnetic flux density versus current. Among the complete 1 mm sweeps from ratios 0.00 to 0.86, the pure-silicone specimen produced the strongest south-side response and a near-strongest north-side response. Adding iron sometimes reduced the response, especially near the 0.50 ratio. A plausible explanation from the final reflection is that dispersed particles did not create a continuous high-permeability path; however, particle distribution was not imaged or otherwise characterized, so this remains a hypothesis.

Setbacks and how I addressed them

As part of the fellowship, I was responsible to write weekly reflections and it showed to me that much of the research value came from resolving practical failures rather than following a clean linear plan.

Paper review

As another delivarable of the fellow, I did a review focused on Cao, Gao, and Conn’s magnetically coupled dielectric elastomer actuator. Their device used two parallel conical membranes that could be driven separately by changing the phase difference between electrical inputs. The paper reported 14% compression and 8.3% expansion, making it directly relevant to my interest in magnetic repulsion and spring biasing even though my prototypes used a different actuation architecture.

What I would test next

1. Repeat each condition across independently cast specimens to separate regression fit uncertainty from manufacturing variability.
2. Complete the missing 1.43-ratio conditions and test additional high loadings before drawing a composition trend.
3. Control particle distribution with a documented mixing, degassing, and curing protocol, then inspect cross-sections to test the particle-continuity hypothesis.
4. Measure mechanical performance—stroke, blocking force, response time, hysteresis, temperature, and cycle life—rather than using magnetic field alone as a proxy for actuator quality.
5. Model the magnet-force and spring-force curves versus distance to identify stable regions where magnetic attraction can overcome elastic restoring force, an idea developed with Prof. Simen during week seven.

References

The project bibliography combined soft-actuator design, magnetically controlled elastomers, additive manufacturing, and general manufacturing texts. Citations used directly on this page are expanded below; the full research reading list is preserved for context.