At long last, the final project. A seemingly simple assignment, yet one that would take nearly as long as the Cretaceous Period to complete.

Our task was to CNC, mold, and cast two versions of a high-quality chess piece into existence—one white and one black. Pedro and Noah (this is a joint write-up, so this blog post is going to be oddly third-personal at times) teamed up and decided to create a knight, mostly because it’s the most visually interesting chess piece. What we didn’t realize at the time was that it’s also one of the only chess pieces that is symmetric across merely one plane, meaning that separate CNC cuts are needed for its two reflected halves. More on this later.

We began by browsing Thingiverse for chess knights, and it was surprisingly difficult to find anything we could use. One of the central constraints of CNC milling is that there can be no undercuts. This essentially means that all components must extend continuously from the plane that will ultimately divide the two molds used to create the part; if a straight line cannot be drawn orthogonal to the dividing plane to every point within the part without leaving the part’s interior, then those inaccessible points cannot be cut. Sciencey language aside, this meant that horses with ears extending from their heads were out of the question. In fact, almost anything with 3D ears, horns, arms, legs, etc. was off the table. With our constraints as they were, we collected a few options: a classic horse-based knight without 3D ears, a rocket-ship that looked like it had been hastily created in Tinkercad, an Easter Island moai, and a standard knight wearing a parasaurolophus skull.

Ultimately, we decided to proceed with the parasaurolophus knight. The parasaurolophus, or parasaur for short, was a genus of dinosaurs with singular long, distinctive horns on their heads, as seen on the right. The Thingiverse design, shown below, featured a 3D scan of a real parasaur skull, which we found very interesting. It wasn’t until we manipulated the CAD file for ourselves that we found that the skull really did have a regular chess knight underneath, upon which the skull was superimposed. This idea of the knight wearing the skull appealed to us, as well as the general aesthetic of the model, so we decided to move forward with it as our final project design.

We began by importing the design into Meshmixer to cut it in half and lower the number of triangles that comprised its 3D model. Upon executing a planar cut on the design, we found that while the knight portion of the model was solid, the skull scan was a hollow shell. Meshmixer (and later Fusion 360) was not particularly fond of this. Nevertheless, we managed to reduce the design to less than 10,000 triangles, export it as an STL, and import it into Fusion 360. Following the steps we learned in class, we prepared the design to be cut via CNC. We scaled our file to match the standard Staunton knight height, and although Fusion 360 indicated that the cut would consume almost the entire depth of the wood, we stubbornly refused to adjust the scale. This meant that the design had to be rotated around the z-axis in order to even fit within the boundaries of the stock material. After this rotation, we applied an adaptive clearing cut and a parallel cut, both generated using the standard settings (except with the rough stepover lowered to 0.25 in, per Dr. Wettergreen’s request). The adaptive clearing cut was to be performed with a 1/8-in. flat drill bit, and the parallel cut was to be performed with a 1/8-in. ball drill bit. Exporting these two g-codes, we decided to test the cuts on a block of pine and proceeded to the Carvey.

Our rough cut (i.e., adaptive clearing) was successful, despite the fact that we approached disaster from a number of angles. First, we must’ve measured the width of the block to be slightly greater than it was, because as the Carvey reached the edge of our bounding box, it cut right through the edge of the pine stock. This was not a huge concern, but if we had used this cut to create a mold, an additional wall would’ve needed to be clamped to the CNC cut in order to prevent molding material from spilling out. More concerning was the fact that the bottom of our CNC cut was wafer-thin, meaning that registration keys could not have been cut into the material without going all the way through. As close as we were to annihilating the base of our CNC cut, we were perhaps even closer to damaging one of the clamps that hold the stock in place. Despite our best efforts to eyeball the size of the cut, our placement of the clamp was too risky and almost resulted in irreversible damage. Thus, the lesson we learned was to be much more gracious with tolerances, whether they apply to width, depth, or distance from sharp, spinning objects.

 

 

 

 

 

 

 

 

 

Our parallel cut was also a learning experience. As quick as it was, we learned that the default settings are not sufficient, as the 0.0625-in. pass width is not short enough to provide a satisfying level of definition. Sizeable ridges were left up and down our cut in between the drill bit’s passes. However, given the number of close calls we experienced with the rough cut, we decided to move on rather than spend time perfecting this iteration’s parallel cut.

It was clear that we needed to decrease the depth of our cut, which meant ceding the official Staunton height of our part. We rescaled the file to 90%, but this also appeared as though it wouldn’t leave enough depth under our part, so we downscaled the file to 90% again. Thus, at 81% of the Staunton size, we once again exported our g-codes, this time with the parallel pass width decreased to 0.025 in. Upon running our new rough cut, we found that it exhibited a few strange features, such as the face shape being cut more “squished together” than before, the base of the piece being much less round than before, and the boundary of the cutting region being much less uniform and—in one place—cut entirely through. Moving on with the parallel cut anyway, we found that the drill bit was tracing out the design’s contours at an offset. Essentially, the Carvey thought that the carved design was closer to us (i.e., lower on the y-axis) than it really was. We terminated the cut and concluded that the Carvey had lost its calibration during the rough cut. Grabbing a new piece of wood, we prepared to try again.

Our third attempt went off without a hitch, with a perfect rough cut and a successful parallel cut. However, the 0.025-in. parallel pass width was still not low enough to eliminate all of the ridges on the part. Along contours parallel to the direction of the passes, like the front of the parasaur’s face, we noticed that the ridges were particularly egregious. This led us to a solution—why not run a second parallel cut in the direction perpendicular to the first one? We quickly created such a cut in Fusion 360, this time with a pass width of 0.01-in. to make things even finer. Running this g-code yielded a very smooth cut, such that little to no sanding was required. The results of each of these cuts are shown below.

Next came a crucial decision for how to move forward with our project. Now that we had a good CNC cut, should we proceed to use it for our molding and casting, or should we spend our time adding registration keys, a fill line, and an air line to our Fusion 360 design? We noticed that almost everyone was improvisationally adding physical registration keys and whatnot to their CNC cuts, rather than figuring out how to construct them in CAD. However, the whole point of registration keys is to make the two mold halves fit together in the exact right orientation, so it seemed useless to estimate their positions. Thus, with little to no training in this specific area, we set out to create registration keys, a fill line, and an air line in Fusion 360.

With a bit of high school Autodesk Inventor experience in tow, we opened Fusion 360 ready to extrude some new design features. Creating a sketch on the bottom plane of the cut, we drew three circles of  0.25-in. diameter to serve as registration keys, and we extruded them upward 0.1 inches. We decided that the fill line, since it would ultimately fill with casting material and leave behind a relatively thick cylinder, should feed into the base of the chess piece, where it will be easy to belt-sand off. We also wanted the casting material to flow towards the parasaur’s relatively thin horn, as we anticipated difficulties in getting this region to cast properly. Thus, in the same plane as before, we sketched a rectangle and a triangle such that, when the sketch was revolved around the edge of the rectangle, a cylindrical fill line extending from the chess piece’s base was formed, which tapered out into a 0.5-inch diameter opening. Finally, we sketched a thin 2 mm-wide air line from the tip of the parasaur horn back upward to the face with the fill line. Extruding this sketch upward, we finally had a design with all of the necessary facets.

After going to the trouble to design registration keys and the like, we thought we were in the clear. Nevertheless, this is where the classic tragic fall comes in. Exporting new g-codes that took the new features into account, we proceeded to run our cuts on a new piece of wood. However, before even completing the first layer of adaptive clearing, the Carvey began making grinding noises and cutting in obviously incorrect patterns. Flipping the wood over, we ran the cut again, and this time we directly observed the carriage vibrating along the y-axis as the drill bit ground against the walls on the upper and lower sides of the cut, unable to move any further. This ruined the Carvey’s calibration and rendered the cut useless. To help alleviate the problem, we decreased the Carvey’s speed, as well as the depth of each cutting layer, so that it hopefully wouldn’t get caught along its y-axis. Despite our best efforts, the problem persisted until the Carvey’s speed was 11 inches per minute (less than a third of its default speed) and we were only carving 0.08 inches deep per layer. This led to extremely long, yet still unsuccessful, cuts.

Given the large decrease in the capabilities of the Carvey within a matter of days, there was clearly an issue with the machine, and we were unsure how to proceed. A quick internet search found that slippage on the Carvey’s y-axis is relatively common. Following a troubleshooting guide, we opened the central panel under the Carvey’s carriage and located the y-axis belt. Loosening a screw at the end of this belt allows the user to pull the belt tighter, and then screw it back into place. However, accessing this screw requires a Phillips head screwdriver shorter than any we could find in the OEDK. Other troubleshooting advice recommended increasing the power to the y-axis belt via a potentiometer, but we were unable to ascertain this potentiometer’s location. Thus, the best we were able to do was vacuum out the region under the Carvey’s central panel and hope that this would help the device run more smoothly.

At this point, we were running out of options, and we were preparing to figure out how to use the X-Carve. Nevertheless, in our darkest hour, our deus ex machina descended into our Greek tragedy: several blocks of wax, including three thick bricks capable of containing our CNC cut. Figuring that wax might be soft enough that the Carvey could cut it without its y-axis belt slipping, we loaded a block of wax into the machine. However, with just three possible attempts at cutting (and two separate parts to cut), there was little room for error. Thus, even though the wax probably could’ve withstood a faster speed and greater plunge depth, we maintained our 11 inches per minute speed and 0.08-inch-deep layers. Furthermore, we wanted the design to have a very high level of definition, so we used parallel passes with a width of 0.0075 in. for both parallel cuts.

This ultimately resulted in three g-codes that, in total, took approximately six hours to run. While watching the Carvey perform these Herculean cuts on the first block of wax, we finally got to work designing the mirrored version of our design with registration keys in the opposite direction. Although a Mirror function exists within Fusion 360, the program simply refused to mirror our design, probably because of the parasaur skull’s weird hollow-shell-like nature. The workaround we finally devised was to export the entire design as an STL, import it into Meshmixer, reflect it in Meshmixer, verify that it consists of less than 10,000 triangles, export a new STL, and import it into Fusion 360. After employing this circuitous method, we arrived at the next problem: these registration keys would need to be cut into the wax, but there’s no clear way to model a hole in empty space. If we added a design feature below the bottom layer of the piece, the g-code would simply cut all the way down to the base of the new feature. Thus, we decided to extrude a wide rectangular slab, on top of which the CNC design would sit. With the top surface of this slab blocking the g-code from cutting where it shouldn’t, we could extrude holes into the surface to show exactly where we want to cut.

With a slab modeled beneath our mirrored design, we reconstructed the circular sketches from the new, weirdly polygonal registration keys using the three-point circle function. Using an extrude-cut, we removed these old registration keys. Then, as the inward-cut cylinders will actually form the outward-pointing pieces of the registration keys, we reduced the diameters of the circular sketches to 0.24 in. and extruded cuts that were 0.99 in. deep so that, hopefully, the registration keys would have suitable tolerances. Exporting three new g-codes, we waited for the first wax block to complete its CNC cuts.

After nearly turning into fossils while waiting, we finally found ourselves with a finished part. The two parallel passes yielded a very high level of definition that didn’t require sanding. We trimmed some wax flakes from the edges of the design with X-Acto knife and sprayed the interior of the CNC cut with mold release. We then mixed together the two components of OOMOO and poured the mixture into our CNC part. During the 75-minute curing period, we began our CNC cuts on our second brick of wax.

When our mold was finished curing, we removed it from the wax and found that had successfully captured the CNC cut’s high level of definition. It was also a very clean mold with no noticeable air bubbles. We were very happy with the mold’s quality and hoped we would achieve similar quality with the mirrored version of the mold.

The mirrored CNC cut was also a resounding success. The holes for the registration keys were properly cut into the wax, and the parallel passes had created such a level of definition that there were almost no wax flakes. We mixed together more OOMOO and poured it into the CNC cut. After curing, we removed the mold and found that it was of a high quality, and its outwardly-extending registration keys were flawless (although one of them became partially squished later, but was still functional).

It was finally time to cast our chess pieces. Since we were the last group to reach this step, there was only so much Smooth-Cast 300/300Q remaining. Thus, for our first cast, we tried using Smooth-Cast 325, which was in lower demand. This casting material produces semi-clear plastic parts, which was not exactly preferable for our black and white chess pieces. However, in this iteration, we learned a valuable lesson: although we had oriented our fill and air lines so that the parasaur’s horn could be accessed by the casting material as easily as possible, we had left the parasaur’s nose relatively difficult to reach. Liquid would not naturally flow into the front of the parasaur’s face, as this would require it to flow upward as we filled the mold. As such, we would need to tilt the mold in a variety of directions as we filled it in order for the casting material to fill that crevice. Since we didn’t perform this tilting with our initial cast, a large chunk of its snout was missing, as seen below.

After removing this cast, we found that our air line was thin enough that we could simply pluck off the extra strand of material along the back of the piece. Additionally, perhaps because the tolerances of the registration keys were too generous (they were only supposed to be a hundredth of an inch, but the way in which the g-codes were generated probably increased them), the two halves of the piece did not line up as nicely as we would’ve liked. They were pretty close, but it wasn’t the perfection we had hoped for. Thus, for our subsequent casts, we had to double-check our registration keys carefully to make sure that they were approximately centered within their respective holes.

Our next cast was created using the Smooth-Cast 300/300Q. This material produces a nice, white plastic suitable for the white version of our final product. However, despite our best efforts to tilt and shake our mold as we filled it, our parasaur was still somewhat nose-less. We did have more success with regards to lining up the two halves of the chess piece, though. Pocketing this iteration as a potentially usable piece, we proceeded to our third cast.

In our third casting, we shook and tilted the mold even more vigorously than before, essentially dancing with our mold in hand. However, we did not hold the mold halves together as tightly as before, which resulted in some leakage. This led to a loss of material in the base of the piece, leaving the base flaky and hollow. Nevertheless, this was our first cast in which the snout of the parasaur was fully captured. Not only that, but the two halves of the piece were perfectly lined up with one another. Frustrated that the hollow base rendered this beautiful parasaur unusable, we strove to replicate our process with the next cast.

Our fourth cast also had a fully-formed face, and its edges were very clean. However, while the front of the part looked perfectly symmetric, the two halves of the part were oddly offset around the back. As shown below, this offset was not horrendous, so we held onto this piece for possible use as our final product.

At this point, we grew upset enough at how perfect our third cast was—except for its hollow base—that we decided to attempt some unorthodox methods. Using the belt sander, we removed the base of our third cast, and we placed it back in our mold. Mixing together a smaller volume of casting material, we re-filled the base of the mold. We hardly expected this new material to bond to the pre-existing cast, but upon removing the cast, we found that our bizarre alchemy had somehow worked. What’s more, the two batches of casting material were almost indistinguishable, such that no one without knowledge of our blasphemous doings would ever notice a difference. After Frankenstein-ing together the perfect white parasaur knight, we proceeded to create our black piece.

Not only was there a frighteningly low volume of Smooth-Cast 300/300Q remaining, but there was also a very small amount of black dye. As such, in the process of creating our fifth cast, we used a straw to scoop two tiny volumes of black dye into one of the two components of the casting material. Mixing the components together and repeating the rest of our previous process, we found ourselves with a knight that was decidedly gray. Moreover, its nose not quite fully formed, and it exhibited the same weird half-alignment as our fourth cast. With only enough Smooth-Cast 300/300Q to produce one more cast, we decided to give our black knight one final try at perfection.

In our final attempt, we doubled our black dye usage, transferring the previous amount of dye to each of the components of the casting material. We were ridiculously careful to align our registration keys as closely as possible, and we shook and tilted our mold like never before. At last, we removed our cast to find the essentially black, perfectly aligned, and fully-nosed parasaur seen below.

Thus, with our favorite black and white casts in hand, we went downstairs and removed their fill lines using the belt sander. We also used fine-grit sandpaper to smooth down the bottom faces of our pieces, although we did so sparingly because it created slight discoloration. Finally, we used an X-Acto knife to remove the ridge around the center of each cast the best of our ability. While we could’ve removed the rest of the ridge with sandpaper, we didn’t want to lose our pieces’ definition or create discoloration on areas of the pieces that are visible during gameplay. In this ridge-trimming process, however, we came across an issue that had previously been hidden by the flaky ridge: our black piece had a hole in its lip created by an air pocket, as seen to the upper-right. We toyed with the idea of spray-painting a different piece black and using it instead, but we decided that this piece was still our next-best cast with all things considered. After all, since the hole is on the underside of the parasaur’s snout, it would not be visible when looking down at the pieces during a game of chess.

At last, we were left with the two complete chess pieces cinematically depicted below. They are very satisfying to manipulate, and we feel that they would be genuinely fun to use in a game of chess. While we were always a bit nervous that the pieces’ horns would cause them to fall backwards, they are nicely balanced. Furthermore, we may have watched the Carvey cut our wax for twelve hours, but the level of definition we achieved with our final pieces was well worth it. All in all, we are very proud of the product we have produced, and we feel that it was a great way to close out the semester.

Finally, the image below shows one last look at every physical iteration we created over the course of completing this project.