Project Description
This project appropriates image-based neural network techniques to interrogate the full extent of the formal design space of bilayer composites through multi-material 4D printing. A Compositional Pattern Producing Network (CPPN) generates distributions of swellable and non-swellable material which define the complex behavior of the bilayer composite system. Thousands of distributions are simulated, and their resulting 3D forms are analyzed and clustered to search for new behavior types. The 3D forms are encoded into an RGB image so that the Pix2Pix machine learning algorithm can be trained to learn the relationship between the material distribution image and its resulting 3D form. With this trained model, a surrogate simulation system is created to predict the 3D form of any material distribution it receives, speeding up the process from one minute to less than one second and allowing designers to test material distributions more fluidly. In the final phase, the machine learning is run in reverse so that a designer may input a desired 3D form and the algorithm returns a prediction of the material distribution to achieve it.