scroll down for more


How can new technologies enable carbon fiber to be pushed to its threshold?

Today, carbon fiber remains an exotic material. Due to its material infancy, carbon fiber’s primary barrier to entry in the architecture / construction market is its cost. But carbon fiber (now mirroring steel’s growing pains before its mass-production during the 19th century) has the ability to be produced in a cheap and sustainable way. In actuality, labor and tooling account for the largest chunk of manufacturing costs after the precursor. This project aims to automate the process of braiding carbon fiber with mass-customization in mind.

braided carbon fiber prototypes


I first investigated the taxonomy of composites -- from the matrix (thermosets, thermoplastics, and elastomers) to the reinforcement (unidirectional, woven, knitted, braided, and stitched) -- in order to conduct a series of material studies using both traditional and non-conventional processes. Among the major processes currently being used in manufacturing, vacuum bagging and compression moulding are the most accessible for relatively low volume production. I replicated the processes of using both to bring a digital design into physical form.

The primary takeaway for vacuum bagging was the limitation of form to ruled surface geometries (doubly curved geometries such as minimal surfaces require clever unrolling and pattern cutting on a laser-cutter). While compression moulding geometries are somewhat more forgiving, the necessity of mould design (in additional to part design), mould protection, and de-moulding (which can be difficult even with proper mould protection) was a hassle. Ultimately, the tooling and de-moulding required more time and resources than the actual resination and fabrication of the part.

I produced several samples of the same mould using different types of cloth -- burlap, linen, glass fiber, and carbon fiber. Carbon fiber definitely exhibited the best strength to weight ratio; not only did the burlap sample break under force, it was also the most heavy.

Composite fabrication typically involves sourcing woven cloth (or unidirectional fabric stacked 0° and 90° to reap quasi-isotropic characteristics), but each end-use application is unique and parts are subject to specific sets of forces. Because carbon fiber is highly anisotropic, mechanical properties and the direction of energy dissipation can be dictated and designed when considering fiber arrangement.

Next, I explored the results from linear Finite Element Analysis of beams, cantilevers, and slabs with typical loading and support conditions for each. Principle stress curves travelling through the user defined design boundary show tension with positive values and compression in negative values (as well as Von Mises stress at points along these curves combining both principal stress values). As these force lines visualize the overall distribution of stress patterns across the entire geometry, I manually laid out the fibers in accordance to positive values for better tension resistance.

In real life applications, structural elements are actually three dimensional and not always planar with constant cross-section conditions. It is therefore possible for a beam or slab to have varying, sectional constraints.

Linear buckling analysis in FEA for a column with typical loading and support conditions showed rapid transition from axial loading response to a lateral response (the entire structure in compression); however, looking at real materials at their point of failures show less simplistic results of distortion (barreling) around the exterior of the element until a maximum limit load is reached. The structure may transition to a new moe shape that can carry further load (for example, the initial buckling of a drink can).

Tensional reinforcement around the exterior would help with failure. For example, the center of a tree trunk is “pre-stressed” as it dries to better resist compression, while the outer layers are in tension. Similarly, the prestressing of carbon fiber with high tensile strength in composites is a very efficient way of avoiding failure in compression.

Inspired by textile and traditional craft techniques, the final series of material studies explored knitting and braiding individual strands of carbon fiber into preforms. Different techniques of fiber arts (weaving, crochet, macrame, and etc) affect structural properties of the global geometry. As I manipulated individual fibers, I identified ways in which the process could be codified into actions performed performed by a machine.

What if a machine could perform the labor intensive tasks of composite making, while capitalizing on carbon fiber's highly anisotropic properties?


Conventional braiding machines used in the textile industry are both mechanically complex and difficult to design. In addition, they are limited to a set pattern and do not allow for varying cross sections. Braiding machines derived from the maypole tradition are comprised of a horizontal track plate provided with grooves in which two sets of bobbin carriers traversed in opposite directions are guided, usually travelling in a tortuous or sinuous trajectory and crossing each other at regular intervals. I also studied knitting mills, small mechanisms that make circular tubes; different numbers of needles influence the diameter of the finished tube. After examining these conventional machines, I built a new type of machine designed for flexibility and scalability. The amount of braided strands can be scaled up by expanding the amount of spools. The machine itself was inspired by several precedents: kumihimo marudai (a disk shaped frame for making Japanese braids), Ilan Moyer’s CNC friendship bracelet machine (MIT Machines That Make), and Evan Bowman’s space weaver (Future Cities Lab). As I made kumihimo friendship bracelets for my friends, I saw how my actions could be easily codified and explored with a machine.

The decision was made to use a CNC gantry system for simpler set up, and to allow more focus on software (toolpath generation and experimenting with pattern planning) rather than hardware. The end-effector picks and places a tool embedded with magnets holding spools of carbon fiber, while a build plate elongates the structure as it goes down. Ultimately, the fundamental mechanics are similar to the idea of a giant 3D printer. The 8 foot frame was designed to be easy to disassemble and reassemble, and collapsable to 2 feet when not fabricating super tall stuff; the whole thing rests on caster wheels for on-the-go composite fabrication.

I sourced the parts for my CNC gantry based off of the Shapeoko 2 framework, and cut my aluminum extrusion guide rails to fit a 24” x 24” work area. I then 3D printed 6 spools to hold the carbon fiber yarns while I built the CNC, whose end-effector engages and displaces the tools that magnetically attaches to the spools through the work area below (which was also 3D printed). The frame was mitered and welded with substantial help by GSD’s Digital Fabrication Lab TA. I later discovered that my ¾” plywood base for the gantry was too thick for my magnets to attach to each other, so I routed a hole for a ⅛” acrylic sheet, laser-etched with absolute starting positions for the spools. A custom machined translucent polycarbonate coupler (thanks to the help of the Harvard Science Center / Natural Sciences Demonstration Shop) connects to a garage door drum, moving a 4’ build plate vertically. This 4-axis motion system is controlled by TinyG and 5 stepper motors. In total, this machine is 24” x 24” x 8’ tall, but can be expanded for super tall or wide parts.


After the linear elastic analysis of 2D plate and 3D volumetric plate elements (such as beams, cantilevers, slabs, and columns), an intuitive set of rules can be established for tackling design problems. The design is explored by generative scripting, and the desired result is then abstracted as a toolpath and saved as G-code to be sent to the TinyG.

There was quite some difficulty in planning the toolpaths so as to not have the magnets on the horizontal plane interfere with each other. To shield and redirect the magnetic field, half of the magnets were wrapped in a hacksaw blade. Because it is ferrous, the magnetic field is redirected and elongated in the z direction. Only half of the magnets required shielding, because the spools work in two groups: one moving on an offset curve so as to not bump into the other group.

Craft in an age of technology provokes us to reconcile the handmade and the machine-made (just like when the sewing machine or photography was invented). I believe that code can be just as expressive and qualitative as traditional crafts. The process of creating with automation require similar amounts of thoughtfulness and expertise, transforming ordinary materials into something extraordinary in a new and redefined way.


There was quite a bit of debugging along the way, both hardware and coding-wise, but I have gotten to a point where there is a good amount of control and accuracy.

demo gcode c# braiding code

Future Work

Further mechanical testing of samples is necessary to verify if there are significant structural improvements from the FE analysis influenced braiding patterns. My next steps are to resinate the braids I have made, and evaluate their performance.

I would also like to incorporate the matrix into the machine’s fabrication, using either spray-up resin or thermoplastics for curing on-the-fly or in-situ consolidation in a single continuous operation, or simply towpreg with heat activation.

In the coming months I will post another update!