Research · Compiling intent to fabrication

Toolpath planning for experimental printers at ORNL

Slicing and toolpath-planning methods that let Oak Ridge National Laboratory's experimental printers, from large-scale extrusion to multi-laser metal powder bed fusion, build parts that conventional slicers could not.

The key idea. New printing hardware often outpaces the software that drives it. Most conventional slicers were built for desktop-scale printers, so they struggle with large-scale and experimental systems, such as printers with build volumes measured in meters, rotating build plates, or several lasers working at once. Research printers need slicing and toolpath planning designed around them.

A 3D-printed metal lunar rover wheel with a wavy tread and wire spokes, standing on gravel
Prototype lunar rover wheel modeled on the wheels of NASA's VIPER rover, printed at ORNL's Manufacturing Demonstration Facility on a rotary, multi-laser powder bed fusion system. Photo: Carlos Jones/ORNL, U.S. Dept. of Energy.

As a staff research scientist at Oak Ridge National Laboratory’s Manufacturing Demonstration Facility (MDF), I worked on slicing and toolpath planning for the lab’s experimental printers.

Printing a lunar rover wheel for NASA

ORNL and NASA printed a prototype wheel modeled on the wheels of VIPER, NASA’s lunar rover, on the MDF’s rotary laser powder bed fusion system, where coordinated lasers melt powder on a rotating build plate, a natural fit for a wheel. The build needed tens of millions of planned and optimized motion commands. Most slicers plan an entire build in memory at once, which for jobs like this can require more than 500 GB of RAM. I built a method called streamed slicing, in which the slicer works alongside the printer during the multi-day build: the printer requests the next layers, and the slicer plans and sends them. Spreading the work across the build kept peak memory to about 200 MB and made adaptive re-slicing from a laser profilometer possible. I also helped develop the method for balancing work between the system’s lasers.

Close-up of the printed metal wheel tread, a series of wavy ridges
Printed detail. The wheel’s wavy tread, printed in metal. Photo: Carlos Jones/ORNL, U.S. Dept. of Energy.
A printed metal ring of triangular lattice cells on a round build plate
The rotary system. A lattice ring printed on the rotating build plate, from the load-balancing study.

ORNL Slicer

I was a core contributor to ORNL Slicer (formerly ORNL Slicer 2), the lab’s open-source research slicer. It began as the slicer for BAAM and grew to support wire-arc and other directed energy deposition systems, laser powder bed fusion, and highly experimental machines.

What sets it apart from desktop slicers:

  • Built for new machines. Its architecture is designed so that new printer types and G-code dialects can be added quickly, which let the lab support unique research systems as they were built.
  • Fine-grained control. More than 450 settings, which can vary by layer, by part, or by region of a mesh.
  • Part of a larger system. A network protocol lets the slicer work alongside simulations and printer controllers and slice on demand, the same idea behind streamed slicing.
  • Closing the loop. Sensor data, such as thermal images and laser height maps, feed back into the slicer so it can re-plan the next layers.

The methods I worked on range from sensor-driven re-planning to five-axis support-free printing, arc fitting for smoother motion, travel reduction to shorten prints, and automatic print orientation. ORNL Slicer is used by more than 200 people at more than 40 organizations across industry, national laboratories, and universities, and I supported these collaborations.

A large gantry 3D printer in a factory bay, with a printed outline on the bed
Where it started. ORNL Slicer began as the slicer for BAAM, ORNL’s Big Area Additive Manufacturing system.
A cycle diagram with steps single-layer slice, print, sense, analyze, and defect mitigation around a cloud labeled enables true real-time path planning
Real-time path planning. Each layer is sliced, printed, sensed, and analyzed, and defects are corrected in the next layer.
Sliced outlines of several parts, and a circle printed with and without arc fitting, where the fitted version is a single arc
Arc and spline fitting. Runs of short line segments are replaced with arcs and splines for smoother motion.
A faceted test part and a bar chart of motion command counts for hybrid fitting, arcs only, splines only, and no fitting
Fewer motion commands. For this part, combining arcs and splines cut the number of motion commands from 106,503 to 15,662.
A robotic welding head above a turn-and-tilt table, a toolpath preview of a flared part, and several metal flared and fluted parts printed on plates
Multi-axis printing. A wire-arc system with a turn-and-tilt table, and metal parts printed with it.
Screenshot of ORNL Slicer with a part on the build plate and a plot of candidate orientations
Automatic orientation in the slicer. The orientation tool in ORNL Slicer, with the best candidate highlighted in green.
A tower model inside its pink convex hull, a table shown in three orientations with dashed convex hull outlines on a blue build surface, and a perforated spiral model
Choosing print orientation. Candidate orientations come from the faces of a part’s convex hull (pink), and each is scored by bed adhesion and the support it needs, fast enough to run interactively on the GPU.
Slide titled Our Userbase listing more than 40 organizations and 200 users, with logos grouped under DOE and DOD, academia, and industry
Who uses it. ORNL Slicer users across national laboratories and the Department of Defense, universities, and industry.

Publications

  1. ORNL Slicer (formerly ORNL Slicer 2)
    ORNL Slicer (formerly ORNL Slicer 2)
    Alex Roschli, Michael Borish, Abigail Barnes, Charles Wade, Breanne Crockett, Liam White, and Cameron Adkins
    Core contributor. Open-source research slicer for experimental additive manufacturing systems, Oak Ridge National Laboratory.
  2. Load Balancing for Multi-Beam Additive Manufacturing Systems
    Load Balancing for Multi-Beam Additive Manufacturing Systems
    Benjamin C. Stump, Brian T. Gibson, Jay T. Reynolds, Charles Wade, Michael C. Borish, and Peter L. Wang
    Additive Manufacturing 74, 103708, 2023
  3. Determining Optimal Print Orientation Using GPU-Accelerated Convex Hull Analysis
    Determining Optimal Print Orientation Using GPU-Accelerated Convex Hull Analysis
    Charles Wade, Breanne Crockett, Michael Borish, and Robert MacCurdy
    SCFProceedings of the 8th ACM Symposium on Computational Fabrication (SCF ’23), 2023
  4. Single Path Generation for Closed Contours via Graph Theory and Topological Hierarchy
    Michael Borish, Alex Roschli, Charles Wade, Brian Post, Liam White, and Cameron Adkins
    Proceedings of the 34th Annual International Solid Freeform Fabrication Symposium, 2023
  5. Hybrid Curve Fitting for Reducing Motion Commands in Object Construction
    Hybrid Curve Fitting for Reducing Motion Commands in Object Construction
    Charles Wade and Michael Borish
    SFFProceedings of the 33rd Annual International Solid Freeform Fabrication Symposium, 2022
  6. A GPU-Based Approach for Path Planning Optimization via Travel Length Reduction
    A GPU-Based Approach for Path Planning Optimization via Travel Length Reduction
    Michael Borish and Charles Wade
    NAMRCProceedings of the 49th SME North American Manufacturing Research Conference (NAMRC 49), Procedia Manufacturing 53, pp. 310–317, 2021