PrintFixer Is Machine Learning Software That Improves the Accuracy of Your 3D Prints
Machine learning is ideal for finding the patterns that cause 3D printing inaccuracy, and PrintFixer takes advantage of it.
3D printing is an amazing technology that can be used to quickly produce intricate parts, but it is still far less accurate than just about every other manufacturing process. A decent CNC (Computer Numerical Control) milling machine can easily maintain a tolerance within a few thousands of an inch, but the vast majority of 3D printers can’t get anywhere close to that — especially FFF (Fused-Filament Fabrication) 3D printers. That’s due to multiple independent sources of imprecision stacking, which is very difficult to account for. Fortunately, that’s exactly the kind of challenge that machine learning is perfect for, and PrintFixer takes advantage of it.
PrintFixer was developed by a team of researchers from the University of South Carolina’s Daniel J. Epstein Department of Industrial and Systems Engineering. Its purpose is to improve the final dimensional accuracy of 3D-printed parts. As it stands today, those parts are often inaccurate due to factors like slop in the drive systems, variations in filament diameter, material expansion and contraction, and the settings and slicer algorithms that attempt to account for those. A single part can easily have a feature that is 0.1mm too big and another feature that is 0.2mm too small. Engineers have to use trial and error to experiment with the dimensions of the part in CAD in order to end up with a part that has the dimensions they actually want.
Machine learning is ideal for taking care of that for us. PrintFixer can look at the original CAD model (the input) and then the actual results (the output) in order to find the differences. By analyzing multiple examples, it can determine how to alter the CAD model in order to get the results that the engineer wants. PrintFixer can improve overall accuracy by about 50 percent, and the accuracy of similar parts to the ones it was trained on by as much as 90 percent. To avoid the need for every user to train PrintFixer, the team hopes that a database can be built and shared between users that contains the necessary alterations for common 3D printer and filament combinations. The team plans to make PrintFixer available to manufacturers and the general public in the near future.