G-code Net: Learning-based rational design and optimization for additively manufactured structures

In this work, we propose an integrated approach for the rational design of 3D-printed parts using machine learning. A Long Short-Term Memory (LSTM) neural network (G-code Net) was used to construct the functional relationship between the G-code and the mechanical properties of the 3D-printed part. Results show that a well-trained G-code Net can make accurate predictions of the mechanical responses and achieve a speed-up of three orders of magnitude compared to finite element. By further combining with genetic algorithm, one can efficiently perform inverse designs for 3D-printed parts with target mechanical response under other design constraints.