Inverse Machine Learning Using Generative Adversarial Networks

This chapter delves into an inverse machine learning framework using generative adversarial networks (GANs) to revolutionize the design of mechanical metamaterials. The preceding chapters have demonstrated the efficacy of forward machine learning techniques in predicting structural properties and suggesting improved biomimetic rods and lattice unit cells. However, this approach is a forward design approach, i.e., structures → properties, unable to achieve “structures by design,” i.e., properties → structures, which needs an inverse design approach. Thus, this chapter focuses on inverse machine learning, which has not been explored in structural design optimization. The inverse machine learning framework leverages GANs to generate potential lattice unit cells and thin-walled cellular unit cells. Coupled with forward regression models and desired properties, this framework outputs desired unit cells tailored for high-performance sandwich structures. By bridging the gap between desired properties and optimal structures, this framework holds immense potential to revolutionize the field of material science and engineering, ushering in a new era of tailored structural solutions for diverse applications.