A Machine Learning Approach to The Inverse Problem of Self-Morphing Composites

Composite materials are valued in architecture for their remarkable strength-to-weight ratio and ability to shape intricate structures. However, conventional methods relying on single-use molds raise environmental concerns. Recent advancements in moldless fabrication, particularly self-morphing techniques’ leverage geometric frustration— internal stresses gemrated by material architecture. Uniaxial shrinkage in composites, traditionally seen as distortions’ can be harnessed to create a self-shaping mechanism, enabling the achievement of complex geometries by varyingfiber orientations. This paper addresses the inverse problem of self-morphing composites, aiming at the generation of production plans from desired designs for morphing. We propose leveraging machine learning, notably Convolutional Neural Networks (CNNs), topredictfiber layouts using 2D data matrices. The paper outlines the use of simulations to construct a dataset for training CNN models to predict the fiber layouts required to achieve design geometry. The contribution of this work is to advance digital design and simulation methods and tools towards the implementation of self-morphing matter in architecturalfabrication.