4D printing of generative AI-assisted designs

This paper presents, for the first time, the concept of four-dimensional (4D) printing of actuators that are designed using a generative artificial intelligence (AI) tool. The generative AI tool used in this work is Ideogram, which is a text-to-image tool that allows creating images based on the user’s prompts. These prompts are carefully formulated to assess the ability of Ideogram to generate designs that have uniform shapes with uniform infill patterns, uniform shapes with nonuniform infill patterns, and nonuniform shapes with nonuniform infill patterns. Two prompts are used to address each case in a different way, adding up to a total of six prompts. Each prompt is used to generate four different images that are used as designs. This process is followed by extracting each design’s path using Inkscape and exporting it to SolidWorks. Each design is then sliced using two different approaches via Simplify3D, which leads to the development of 48 designs. The first method prints the passive parts of the actuator with a 100% infill and the active parts with a 0% infill (except for a single outline perimeter), while the second uses a 100% infill for both parts. The designs are then 4D printed from polylactic acid (PLA) using fused deposition modeling and then activated in hot water. The bending performance of the activated PLA actuators is then characterized using Kinovea. The different approaches of printing cause the bending performance to be significantly altered. Therefore, the bending angles of these designs range from 9.3° to 72.8°, while the directions of bending range from 0° to 170°. The developed designs in this work demonstrate complex bending performance due to the randomness of the directions of the induced strain during the printing process. The results illustrate that the proposed concept in this work is promising for various types of 4D printing applications that require developing complex and creative designs, offering new possibilities for 4D printing where standard design methods may fall short.