Abstract
Additive manufacturing (AM), commonly referred to as 3D printing, has undergone significant advancements, particularly in the realm of stimuli-responsive 3D printable and programmable materials. This progress has led to the emergence of 4D printing, a fabrication technique that integrates AM capabilities with intelligent materials, introducing dynamic functionality as the fourth dimension. Among the stimuli-responsive materials, shape memory polymers have gained prominence, notably for their crucial applications in stress-absorbing components. However, the exact 3D shape morphing of 4D printed products is affected by both the 3D printing conditions as well as the stimuli activation. Hence it has been hard to precisely control the 3D shape morphing accuracy. To model and optimize the dynamic 3D evolution of the 4D printed parts, we conducted both simulation studies and real-world experiments and introduced a novel machine-learning approach extending the concept of normalizing flows. This method not only enables the process optimization of the dynamic 3D profile evolution by optimizing the process conditions during 3D printing and stimuli activation but also provides interpretability for the intermediate shape morphing process. This research contributes to a deeper understanding of the nuanced interplay between process parameters and the dynamic 3D transformation process in 4D printing.