Adaptive 3D Printing for In Situ Adjustment of Mechanical Properties

Abstract

It is shown that naturally occurring under‐extrusion results in mechanically weak prints while over‐extrusion causes excess use of material with little strength gain. Herein, a deep‐learning‐based computer vision system to correct under‐ and over‐extrusion issues commonly found in 3D printing technology such as the fused deposition modeling (FDM) is developed. The adaptive correction system is created to acquire recurring images of print‐in‐progress, allowing pretrained convolutional neural network (CNN) models to classify the printing condition. Then the classification data allow the adaptive system to make subsequent changes of printing parameters in a simple feedback loop to correct printing extrusion in an average of four to eight printed layers. The result shows that the system can improve the strength consistency of the prints by reducing yield strength variance by a factor of six through in situ correction. This system strengthens weaker prints by up to 200% and can save up to 40% material amount in extreme over‐extruded cases. In the future, the deep‐learning approach demonstrated in this design can be expanded to correct different parameters and its corresponding defects in the other 3D printing technologies with the same methods.

Keywords

• 3D printing
• in situ extrusion control
• machine learning