Using Iterative Learning Control to Improve the Accuracy of Desktop Fused Deposition Modeling Printers: An Experimental Case Study

Abstract

Additive manufacturing (AM) sits poised to make a large impact on the manufacturing sector. Fused deposition modeling (FDM), a type of AM, while versatile, and increasingly appearing in full production systems, has performance limitations in certain geometries, such as arcs and holes. This is especially true for the desktop setups common in College Maker Spaces and other prototyping environments. For these use cases, it is critical to obtain accurate parts quickly, yet often difficult, diminishing the value of using FDM, whether it be to prototype new designs, make final parts, or anything in between. Iterative Learning Control (ILC) has been applied to robot control, plastic extrusion, and other similar processes where disturbances to a system are present and relatively constant, but difficult to model and correct. Since desktop printers perform repetitive tasks subject to nearly constant disturbances that induce inaccuracies, a natural research question arises: can ILC be used to allow desktop printers to learn these inaccuracies and account for them, allowing such printers to create more accurate and useful parts for the average prototyping user? Details on the printer, a LulzBot Taz 6, and the scanner, an Einscan 3D Scanner, being used to answer this question are first presented with some baseline data to establish the scanner’s nominal accuracy. Subsequently, a simple bounding box approach was developed and tested where only the part’s length, width, and height were monitored and adjusted. This approach determined an error metric for a scalene triangular prism by determining the length, width, and height of a box that bounds the shape. The ILC algorithm used this error metric to generate a new file to print for the next iteration, thus creating parts that became more and more accurate. While this approach exhibited some success, it cannot account for larger, more common issues such as warping (where shrinking occurs as the plastic cools over time causing bending or bowing in the part), or a hole being geometrically inaccurate compared to the desired diameter. To address these concerns, a grid approach was developed where the cardinal dimensions had a grid overlaid so that points along each dimension could be checked and adjusted in subsequent prints to account for such issues. This approach was applied to rectangular bars with relative success. The overall dimensional accuracy (e.g. length, width, height) was not significantly improved, however, warping along the length of the bar was significantly reduced. A similar approach for more complex geometries, (i.e. holes and arcs) is currently under development. Initial thoughts and plans are presented as concluding remarks. Using ILC to account for common issues with desktop FDM printers could enable higher quality parts to be made, without a substantial investment in higher-grade equipment.