Machine Learning-Evolutionary Algorithm Enabled Design for 4D-Printed Active Composite Structures

Active composites consisting of materials that respond differently to environmental stimuli can transform their shapes. Integrating active composites and 4D printing allows the printed structure to have a pre-designed complex material or property distribution on numerous small voxels, offering enormous design flexibility. However, this tremendous design space also poses a challenge in efficiently finding appropriate designs to achieve a target shape change. Here, a novel machine learning (ML) and evolutionary algorithm (EA) based approach is presented to guide the design process. Inspired by the beam deformation characteristics, a recurrent neural network (RNN) based ML model whose training dataset is acquired by finite element simulations is developed for the forward shape-change prediction. EA empowered with ML is then used to solve the inverse problem of finding the optimal design. For multiple target shapes with different complexities, the ML-EA approach demonstrates high efficiency. Combining the ML-EA with computer vision algorithms, a new paradigm is presented that streamlines design and 4D printing process where active straight beams can be designed based on hand-drawn lines and be 4D printed that transform into the drawn profiles under the stimulus. The approach thus provides a highly efficient tool for the design of 4D-printed active composites.

Automated summary

The study presents a novel approach using machine learning and evolutionary algorithms to guide the design process efficiently for achieving target shape changes. By utilizing a recurrent neural network-based ML model and 4D printing technology, the design process becomes more flexible, allowing the design of active straight beams based on hand-drawn lines.