Machine-Learning-Powered, Rapid, Accurate, and Multi-Target Mechanical Metamaterials Inverse Design.

Area of Science:

• Materials Science
• Mechanical Engineering
• Computational Design

Background:

• Multi-target inverse design is crucial for mechanical metamaterials (MMs) with varied application requirements.
• Existing methods like topology optimization are often slow, inaccurate, and limited to single objectives.
• Practical applications, such as footwear, necessitate designing components with distinct mechanical properties.

Purpose of the Study:

• To develop a rapid, accurate, and multi-target inverse design approach for mechanical metamaterials.
• To leverage machine learning and graded triply periodic minimal surface (TPMS) architectures for MM design.
• To address the limitations of current single-objective optimization techniques in MMs.

Main Methods:

• Utilized graded triply periodic minimal surface (TPMS) architectures.
• Developed a machine-learning-powered approach with a six-parallel pipeline network.
• Employed deep networks to map structural parameters to mechanical curves for inverse design.
• Selected optimal designs based on target curves and derived performance requirements.

Main Results:

• Achieved a normalized root-mean-square error (NRMSE) of 2.49% on the test dataset.
• Demonstrated rapid generation of design parameters within seconds.
• Successfully met multiple design targets simultaneously.
• Validated the approach through the design of adaptable footwear soles.

Conclusions:

• The proposed machine learning approach enables efficient and accurate multi-target inverse design of MMs.
• This method overcomes the speed and accuracy limitations of traditional optimization techniques.
• The approach has practical implications for designing customized MMs in various fields, including biomechanics and footwear.