At the same time, the mechanical properties of other materials have similar characteristics. The toughness and tensile strength of a polymer are affected by its molecular structure and degree of crosslinking. Ceramics have high hardness and compressive strength but are brittle and easy to break. Depending on the application scenario, we can choose different types of materials to meet the needs. For example, in high-temperature applications, ceramic may be a better choice because it has better heat and corrosion resistance. In contrast, polymers may be more suitable for low-temperature and high-resistance applications because of their high insulating and mechanical flexibility.
Through three-dimensional metamaterial microarchitecture design, additive manufacturing enables customization of the mechanical properties of materials that exhibit unusual properties such as negative Poisson’s ratio, negative compressibility, ultra-light and ultra-stiffness, shape recoverability, and multiple steady states. These architecture materials implement previously unattainable properties in material performance selection diagrams such as Ashby charts, which are often designed through forward design methods, topology optimization, and machine learning. The forward approach iteratively adjusts material design parameters (e.g., element size, wall thickness, etc.) until the measured or simulated material properties meet specified design standards, which often require extensive knowledge from experienced designers. While topology optimization and machine learning have shown the potential to produce designs that provide the desired properties, they have not yet accurately captured all the required mechanical behavior in practice because design mappings and responses are not unique, along with a large number of variables. These design methods are also further complicated by the presence of manufacturing defects, process variability, and uncertainty, requiring extensive calibration to account for defects in additively manufactured samples with hundreds of millions of spatial components. Due to these challenges, the actual mechanical properties of manufactured samples often deviate significantly from the design properties.
Researchers have developed a new method of using artificial intelligence and 3D printing to manufacture materials with precisely defined properties, an innovative technology that will help solve bottlenecks in the development of traditional materials. This technology allows users to precisely define the mechanical properties of materials as needed, avoiding material manufacturing errors that can negatively affect performance. The introduction of this innovative technology will help drive development in areas such as construction, aviation, automotive, and healthcare. We believe that this new approach based on artificial intelligence and 3D printing will have a positive impact on the future of material development.
As far as the design approach is concerned, several steps are followed. Using the generation of inverse and proxy forward neural network models, the input includes a user-defined uniaxial compressive stress-strain curve (in the form of curved features and manufacturing parameters, given the maximum build volume size and minimum printable feature size of the 3D printer), outputs a set of optimal design parameters to describe the digital lattice design, and then 3D printed and tested. To achieve this, the researchers developed a series of lattice units capable of capturing different curve shapes under monotonic and cyclic compressive loads that cover a wide range of mechanical behavior of lattice/lattice/honeycomb structures. These units act as building blocks for creating training datasets distinguished by two different (brittle and flexible) polymer substrates, from which the machine learning pipeline learns relationships between various mechanical behaviors, topologies, and process-related manufacturing errors, and generates printable lattices that replicate the target stress-strain curve. This approach allows for the rapid creation of materials with fully customizable mechanical behavior, taking into account manufacturing process errors and nonlinear behavior.
To demonstrate the customizability of the design process, the researchers reverse-engineered shoe midsoles to enhance running performance by graphically customizing stress-strain curves measured from commercial midsoles. During heel-toe running, the midsole is divided into four sections at different load levels, and the target stress-strain curve for each section is created by customizing the baseline curve to maximize the thrust and cushioning forces of the run. The tailor-made midsole consists of a hard but comfortable toe section, a softer and higher propulsion forefoot section, and a stiffer but energy-intensive heel section. In addition, the target curve is scaled according to the proportional relationship between the strain rate and the mechanical properties of the base material (TMPTA), allowing the dynamic response in the running scene to be reverse-engineered using quasi-static training data. The results show excellent agreement between the experimental curve and the target curve for each custom section (average prediction accuracy >90%), indicating that the machine-learning pipeline is capable of producing materials that meet multiple custom mechanical responses under different loading conditions.
The researchers also reverse-engineered three stress-strain curves to experimentally verify the effectiveness of the method, and the results show that these advanced reverse-engineered stress-strain curves have continuous peak stress and coordinated collapse mechanisms as well as customized softening effects, making the reverse-engineered composite lattice an excellent candidate for custom energy-absorbing filler materials designed by machine learning.
Scientists are confident in the combination of artificial intelligence and 3D printing, believing that this technology will bring unprecedented adaptability and innovation to material development. With this technology, the manufacture of materials with specified characteristic values will become more precise and efficient.
