UCB ARCH 259 Robotic Fabrication

Mark Ma, Yasaman Yavari

# Goals

The final object is to make something fantastic, while creating a workflow including techniques of 3D scanning, parametric modeling, robotic arm operations, 3D printing and bio-mechanical knowledge.

While breaking down the final object, Yasaman and I found our interest in the parametric modeling part. With instructions from Professor Simon, we began our quest to develop a robot-printing friendly curve generation algorithm. To be robot-printing friendly, the curve generated should have the following features:

- Artistic (subjective)
- Reasonably structured (objective)
- As continuous as possible so that saves the trouble of reconnecting during printing (objective)

# Modeling

## Base shape

With the conclusion that we’re to use elastic canvas for basic form-finding, the first step of our modeling would be simulating this process. Several opinions are available:

- Use finite element method to find the exact shape of the canvas under loading
- Use Kangaroo to simulate the physical process
- Use minimal surface

The minimal surface method is dropped due to the fact that we are to use form-finding in reality. The FEA method, despite exact and promising, does not go well with the rest of the project with its model format. Considering that the modeling precision would not be a big issue in this step, Yasaman and I decided to use Kangaroo for base shape simulation.

The drawback of Kangaroo is that we can hardly define an isotropic elastic material. The Quadrilateral Mesh, despite simple and straightforward, promises the same modulus only in the orthogonal direction of the mesh UV basis.

Through careful modeling, I add another restriction at 45° and 135° of the UV basis, making it closer to an isotropic material. Considering the exact model are to be found out using real form-finding and 3D-scanner, the remaining error is neglected.

## Structural reasonableness

Considering the problem to be print on the basic form, several ideas of curve generation have been proposed. One plan is to adopt optimized principle stress lines. The advantage is obvious, as such a structure would be very reasonable under loading. However, principal stress lines turned out hard to be adopted. The continuity of the principle stress line does not guarantee the well-distribution of these lines, which often leaves large holes in structure generated. Due to the high sensitiveness to shape and load, the principal stress lines are hard to manipulate too. Consequently, we decided to start from the curve generation itself.

## Curve generation

### Step 1. Curve generated according to force distribution

One way of curve generation is to divide and lengthen parts of a continuous curve, while keeping the curve away from intersecting with itself, which defines all the goals we need to implement this idea with Kangaroo:

- Lengthen curve sections
- Prevent curve from intersecting with itself

To adapt the generated curves responsive to structural behavior, it's natural to densify the area with large stress, which gives the third goal:

- Curve distribution responsive to stress distribution

With these goals the curves are successfully generated, meeting all our objective goals. Further development should focus on the subjective part, aka the artistic effects.

### Step 2. Curve generated freely

To gain more control of the curve generation, some restrictions have to be dropped. Now take a look back at the initial three goals, if we neglect the curve distribution requirement, we can achieve something looks very alike Zaha’s project.

Recall that Zaha’s project uses a manually defined base curve to approximate initial stress distribution, we now certainly have a better solution: to use the curve generated in 1^{st} step as the base shape for 2^{nd} step, which gives us the following result.

### Step 3. Combination

There is an infinite number of curves we can generate by tweaking with all the parameters, and an even larger infinite number of combinations with different layer position, tube thickness, color, etc. For a quick example, this is what it will look like if we combine Step 1+2 as a thicker base layer and Step 2 as a thinner decoration layer.

# Conclusion

The curve generation is a complex workflow not fully automated, as lots of attention needs to be paid to calibrate all the parameters in order to get the best shape. All generated curves have the potential to become the final project, with a combination of each other and further improvement based on printing practice.

Due to the virus situation nowadays, it’s becoming harder for our project to be actually constructed. I hope all these quests into curve generation can be preserved, developed, and applied in future practice.