Researchers at Imperial College London have developed a computational framework for the inverse design of nonlinear mechanical metamaterials, using topology optimization to generate microscale unit cells from prescribed homogenized stress–strain targets.
Published in Advanced Engineering Materials, the study was authored by Charlie Aveline, Matthew Santer, and Robert Hewson from Imperial College London’s Department of Aeronautics. The framework incorporates internal contact, snap-through buckling, and bistability in a single workflow, allowing designers to synthesize unit cells with complex mechanical responses without starting from predefined unit cell geometries or machine learning datasets.
The authors state that the approach could support the development of mechanical metamaterials for morphing structures, soft robotics, and energy absorbing materials. Mechanical metamaterials derive their unusual properties from the geometry of their internal unit cells. Additive manufacturing has expanded the range of physically realizable metamaterial geometries, but the paper notes that their unintuitive and multiscale behavior still requires robust design tools.
Designing unit cells from target stress–strain behavior
This framework uses density-based topology optimization to tune microscale unit cells. Each element in the design domain is assigned a density value between 0 and 1, representing void and solid material. The optimizer iteratively updates these densities until the simulated homogenized stress–strain response matches the user-defined goal points.
The workflow uses open-source Python libraries including Firedrake, pyadjoint, and cyipopt. For each design iteration, macroscale strain is applied across the unit cell, the microscale equilibrium is solved using finite element analysis, and the resulting homogenized stresses are compared with target values. Sensitivities are then calculated and used to update the unit cell geometry.
A key element of the framework is its use of the third medium contact method. This allows void-like regions between solid members to stiffen when highly compressed, enabling the model to transmit contact forces without explicitly defining contact interfaces. This differentiable contact formulation makes it suitable for gradient-based topology optimization.
The authors also added constraints to improve the physical realism of the generated designs. These include a volume constraint, a penalty against intermediate “gray” densities, and a tensile stiffness constraint to avoid disconnected structures or unit cells that only become stiff once contact occurs.
Pseudo-ductile, monostable, and bistable responses
To test the method, the researchers generated unit cells for three nonlinear responses under compression. The first was a pseudo-ductile response, with an initially stiff behavior followed by a stiffness plateau. This type of response can limit peak force and is relevant to energy absorption.
The second was a monostable snap-through response, where the structure softens after a critical buckling point before restiffening. The third was a bistable response, where the negative stiffness region extends below the strain axis, allowing the structure to remain in a second stable compressed configuration without continued loading.
Starting from a ring-shaped density initialization, the optimizer produced distinct unit cells for each target response. The pseudo-ductile design formed protrusions that contacted the side walls during compression. This contact balanced the stiffness reduction caused by buckling and produced the plateau response. The monostable and bistable designs formed thinner compliant hinges, enabling sharper snap-through behavior.
The simulated responses showed close agreement with the prescribed goal points. The authors report that each optimization took approximately eight hours on a 2023 Apple MacBook Pro M3 with 8 GB of RAM, avoiding the dataset generation and specialist computing infrastructure associated with some machine learning-based approaches.
The study also tested Perlin noise initializations for the bistable target response. These random starting geometries produced different final unit cells that still tracked the target stress–strain points. According to the authors, this indicates the presence of multiple local minima in the nonlinear design space, meaning different geometries can satisfy similar mechanical goals.
Experimental validation using 3D printed molds
The researchers validated the computational framework using silicone specimens produced from 3D printed molds. The optimized unit cell geometries were extracted from the density fields, rescaled, and extruded to create 70 × 70 × 20 mm specimens. PLA molds were fabricated using commercial FDM 3D printers, coated with a release agent, and filled with AS40 addition cure silicone. The samples were left to cure for two days before testing.
Each unit cell was compressed at a rate of 5 mm/min using an Instron mechanical testing machine. The test setup was designed to approximate the periodic boundary conditions used in the simulations, with the unit cells fixed to aluminum blocks and supported laterally by low-friction linear bearings.
The experimental results captured the main nonlinear behaviors predicted by the simulations. The monostable sample showed strong agreement with the computational response for most of the test. The bistable sample demonstrated bistability, although it buckled earlier than predicted because its compliant hinges did not buckle synchronously.
The pseudo-ductile sample showed the largest divergence from the model. In the simulation, contact occurred earlier than in the physical sample, producing an earlier stiffness plateau. The authors attribute this partly to a known limitation of the third medium contact method, where void elements may transmit forces before physical contact is fully established.
Despite these differences, the authors conclude that the tested unit cells validate the computational approach. The framework was able to generate designs that demonstrated contact, snap-through buckling, and bistability in physical testing.
Designing nonlinear response before geometry
The study addresses a constraint in metamaterial design: how to generate unit cells with highly nonlinear behavior without limiting the design space to known geometries or relying on large machine learning datasets.
By combining topology optimization with multiscale modeling and third medium contact, the framework allows a designer to begin with a target mechanical response rather than a target shape. This could be useful where the desired behavior is known, but the unit cell geometry needed to achieve it is not.
The results remain limited to single-material unit cells tested under controlled compression. The authors also identify manufacturing imperfections and premature contact prediction as areas for further work. Improving numerical stability in the third medium contact method could help reduce this gap between simulation and experiment.
Still, the paper shows how inverse design tools could expand the range of nonlinear mechanical metamaterials available to engineers, particularly for soft robotics, adaptive structures, and energy absorbing systems.
Designing nonlinear response before geometry
The study fits into a wider effort to make additive manufacturing design tools work from functional requirements, rather than geometry alone. A recent review on 3D printed composites highlighted how computational design is being used to tailor structural, thermal, electrical, and responsive behavior through material distribution, fiber orientation, and topology optimization. For nonlinear mechanical metamaterials, designers need tools that can generate a unit cell once the target mechanical response is known.
Computational cost is another barrier. An Imperial College London-led study with ToffeeX recently proposed a multiscale framework for 3D heat sink optimization, reducing memory use by up to 90% and computation time by 70% compared with explicit single-scale simulations. The nonlinear metamaterials study addresses a related bottleneck by allowing target stress–strain behavior to drive unit cell design directly, while accounting for contact, snap-through buckling, and bistability without relying on predefined geometries or machine learning datasets.
3D Printing Industry is inviting speakers for its 2026 Additive Manufacturing Applications (AMA) series, covering Energy, Healthcare, Automotive and Mobility, Aerospace, Space and Defense, and Software. Each online event focuses on real production deployments, qualification, and supply chain integration. Practitioners interested in contributing can complete the call for speakers form here.
To stay up to date with the latest 3D printing news, don’t forget to subscribe to the 3D Printing Industry newsletter or follow us on LinkedIn.
Explore the full Future of 3D Printing and Executive Survey series from 3D Printing Industry, featuring perspectives from CEOs, engineers, and industry leaders on the industrialization of additive manufacturing, 3D printing industry trends 2026, qualification, supply chains, and additive manufacturing industry analysis.
Featured image shows a close-up of microstructure optimized for pseudoductile behavior with mesh distortions leading to contact behavior. Image via Aveline et al.
