mimetikxs / 2d-diffusion-limited-aggregation-experiments Goto Github PK

Read my Medium article to learn more about diffusion-limited aggregation and this project.

Additional media is available on my portfolio

This repo contains a series of visual experiments built with JavaScript that explore the topic of diffusion-limited aggregation (DLA) as a method for generating interesting 2D forms.

I am particularly interested in the application of such techniques in the context of digital fabrication, so these experiments will be more focused on schematic representations (colorless, vector-based, SVG/STL exports) over purely visual effects.

Diffusion-limited aggregation (DLA) is a process in which randomly-moving particles diffuse through a medium and clump together (aggregate) over time to form long, fractal, branch-like chains (sometimes called Brownian trees). It closely models various interesting phenomena seen in nature at different scales and in different mediums.

A classic example is that of the formation of copper sulfate crystals in the presence of an electrodeposition cell. When electricity is applied, individual copper atoms are stripped from the system's anode and randomly float (diffuse) through the liquid medium until they come in contact with other copper atoms that have accumulated on the system's cathode where they form a strong molecular bond and aggregate over time.

Another example can be seen in the rather more violent phenomena of Lichtenberg figures, wherein an electrical discharge of very high voltage travels through an insulator like wood, burning a curious fractal branching structure in it's wake. In this example, it would seem that the electrical discharge itself diffuses through the wood, limited by the insulating nature of the wood, forming an "aggregate" of burnt wood as it progresses.

In classical implementations this algorithm acts upon a regular 2D grid of pixels wherein each "particle" can have up to 8 neighbors. Though simplistic, this so-called "on-lattice" approach can run at blistering speeds because no expensive distance calculations, spatial indexing, or collision detection is required - just array lookups.

However, this approach results in an inherently low fidelity raster image that has a pretty characteristic aesthetic style and limited usefulness in modern digital fabrication workflows. In the world of digital fabrication vector-based graphics are preferred because they can be easily transformed into machine toolpaths and manipulated in interesting ways in CAD software.

To achieve vector-based results from the DLA process one must move away from pixels and towards particles, which also affords one the ability for more parameterization that can be fun to explore creatively. For example, one could vary the size, shape, and movement behaviors of these particles to achieve interesting effects.

• p5.js for canvas drawing and miscellaneous helper functions (like lerp and map).
• Webpack for modern JS (ES6) syntax, code modularization, bundling, and serving locally.
• collisions for robust, lightweight collision detection without the use of a full physics package.
• svg-pathdata for parsing path information from SVG files in order to create custom shapes.
• svg-points for generating the d attribute of SVG <path> elements for export feature.
• file-saver for initiating a download of exported SVG fles to the user's machine.

1. Run npm install in both the root (/) and ./core folders.
2. Run npm run serve to start a local development server and launch it in a browser.

To statically build the code in this repo, run npm run build in the root folder.

The code in this repo can be improved or optimized in a few ways to achieve faster performance and larger scales. Here are some ideas that come to mind:

1. Rather than using brute-force movement and collision detection (which limit overall scale + speed), implement an algorithm like Michael Fogleman's dlaf. This may mean that one would have to do away with variable particle shapes and sizes, but I imagine variable sizes could be accommodated with some tweaking.
2. Port code over to a more performant language / library like openFrameworks or Cinder.

• DLA - Diffusion Limited Aggregation by Paul Bourke
• Diffusion-limited aggregation on Wikipedia
• Coding Challenge #34: Diffusion-Limited Aggregation by Dan Shiffman (Github repo)
• Coding Challenge #127: Brownian Tree Snowflake by Dan Shiffman (Github repo)
• simutils-0001: Diffusion limited aggregation by Karsten Schmidt (toxiclibs)
• Diffusion Limited Aggregation (PDF) by André Offringa
• Simulate: Diffusion-Limited Aggregation from FORM+CODE book examples
• Diffusion-Limited Aggregation by Softology
• Diffusion-limited aggregation: A kinetic critical phenomenon? (PDF) by Leonard M. Sander
• Dendron Processing sketch by Golan Levin