Georgia Tech 2017, Robotic Fabricates, collaboration with Minsuk Chun, Priya Kandharkar, and Dajiah Suggs.

In the second edition of To Be Automated, I reflect on projects which attempted to translate manufacturing concepts from automotive and aerospace industries to the field of architecture.

How did we get here so fast?

History classes in architecture school have a tendency to reduce the effects of the industrial revolution to the steam engine and the introduction of iron and glass in structural design. We speak conceptually about “machine,” but we rarely get into the specifics. When we think of manufacturing today, we picture factories filled with assembly lines and all sorts of specialized machinery, but the process to get here was full of incremental change and refinement.

Precursor: A Brief History of Manufacturing

From a manufacturer’s perspective, the crucial advancement that lead to the proliferation of the steam engine was not the invention of the engine itself, but the critical moment in 1765 when James Watt realized a separate condenser could triple the mechanical efficiency of the engine. This modification initiated a series of improvements which saw the engine becoming commercially viable by 1776. However, at this time the machine parts were still being produced by hand rather than by machine. Every piece would be slightly different, requiring modification to work inside mechanical assemblies. In 1798, Eli Whitney, inventor of the cotton gin, won a contract with the U.S. government to manufacture 10,000 muskets. The musket-making process consisted of 29 unique parts and 195 steps, and Whitney had no experience making guns. Instead, he had an idea to introduce uniformity and exactness in the production of the parts. In a demonstration for the U.S. government, he pulled parts from a pile seemingly at random, successfully assembling a musket without modifying a single part by hand. This demonstration secured future funding to develop the idea, and although Whitney was several years late on the delivery of the initial musket order, the development of interchangeable-machine made parts was a significant step toward mass production. (Bonvillian 15-22)

Henry Ford’s production of automobiles in the early twentieth century introduced many of the staple manufacturing ideas that we think of today. Ford worked with engineers to set up factory layouts to allow not only assembly line style production, but also quality control and delivery of parts to work stations. (Bonvillian 23) Beyond automotive production, the U.S. military would invest heavily in aerospace manufacturing, ramping up during the world wars and continuing well into the latter half of the century with the creation of NASA. Post-war America established many of the standards of manufacturing that enabled mass production of consumer good, but it’s worth noting that the U.S. is not necessarily the global leader in manufacturing. Japan pioneered high quality manufacturing specifically in the semiconductor industry, enabling the trend of increasing technological capacity in consumer goods at a decreasing price. Germany stands out in the manufacturing world for their resilient system of distributing training and expertise amongst the workforce. In American manufacturing, each worker is trained to do their job and their job alone, specializing in one specific task. This practice leads to bottlenecks in production when the specialists are unavailable and no one is able to fill in. In Germany, when new stations are introduced to the production floor, everyone gets training so that laborers can substitute for each other, allowing production to continue in spite of changing workforces over time. Especially in the late twentieth century, the IT revolution and the proliferation of computers caused the U.S. to shift from heavy focus in manufacturing toward software development. Recently, this shift has led countries with lower labor costs to step in, with China largely taking over the gap left by the U.S. decline in manufacturing.

Bonvillian, William B. Advanced Manufacturing. The New American Innovation Policies. Cambridge, MA: The MIT Press, 2018.

Built on Small Improvements

The multitude of small improvements leading to modern manufacturing have established a robust set of principles and practices that allow for contemporary mass production. British firearms manufacturer Geoffrey Boothroyd produced multiple books detailing procedures for designing for assembly. His 1989 book Product Design for Manufacture and Assembly stands out as a resource for best practices. Among the various considerations for design that architecture school does not typically cover, he details best practices for designing objects to be moved through feeders, to be oriented for gripping with a robot, for sequencing assembly, and for minimizing number of parts. This book and others like it reach a level of technicality beyond what the architectural discipline typically requires, but the principles are useful. Most of all, books on manufacturing emphasize the value of every incremental improvement. Some of the most radical shifts in efficiency which enabled leaps forwards came from very small adjustments.

Boothroyd, G., and Peter Dewhurst. Product Design for Assembly. Wakefield, RI: Boothroyd Dewhurst, 1989.

The Difference between Architectural, Automotive, and Aviation

When discussing digital fabrication in architecture academia, we inevitably look toward the industries which are pioneering the manufacture of products. Automobile production gave us the assembly line and introduced mass mechanical production for a consumer market. Aircraft production necessitated high precision modeling and manufacturing: Famously, CATIA started as an in-house modeling software developed for French aircraft manufacturers before software developer Dassault Systèmes began distributing the software. This software would allow Gehry Techologies to realize designs previously unachievable, and eventually led to the development of Solidworks and other 3D modeling software from competitors such as Rhino. The long list of innovations coming from these two industries were enabled by high demand either from consumers or militaries, as well as substantial investment into research and development to accelerate new technological developments. When architects suggest that architecture could learn from these industries, we need to reconcile with the specific obstacles architecture is facing. For consumer products like automobiles, buyers are looking for predictable quality and function. It’s acceptable that every single car of a certain model looks and functions exactly the same, which is not necessarily true of architecture. When housing is repetitive, there is a distinct reduction in desirability and quality in the eyes of the occupant. People want residences that reflect their individual site, context, family needs, and aesthetic preferences. So far, no mass produced house has captured the market because people do not want to live in endless suburbs where all of the neighbors live in identical houses. So, in terms of mass production, architecture does not have the same type of demand as the automotive industry. Architecture also does not need the same level of precision as the aerospace industry. Airplanes and space shuttles need much more precise parts to perform, whereas building tolerances in architecture are quite forgiving. Not to mention, the aerospace industry has benefited from enormous military funding to overcome production obstacles especially in times of war. Architecture has not seen the same surge in demand, meaning there is no need to mass produce architecture in the same manner.

2018 Drawing of assembly line applied to masonry pre-fabrication

Mass Customization versus Mass Production

We might never see architecture adopt the same mass production approach that other industries have undergone, but mass customization might emerge as a suitable alternative model. Mass customization could be considered a type of mass production, but the key difference is that mass production is a fixed process with a fixed outcome. Mass customization is a mass production framework with affordances built in to allow variation. The simplest version of this might be some binary variation like putting a different color of paint on a product at the last step of the assembly process. A more exciting version for the purposes of architecture would be parametrization of steps along the process, like the ability to vary the heights or types of windows into a pre-fabricated stud wall assembly. These tasks are simple in isolation, but to account for potential variation within an efficient mass production of parts becomes exponentially more complex. Suddenly, the system not only needs to process a variety of parts, but move those parts onto the next step of the process. For this reason, mass customization is most accessible in surface level operations that don’t change the geometry of parts. Such services allow customers to order custom images in panelized assemblies or specific patterns in window frit without negatively affecting the production time of units.

This mass customization is one of the key opportunities where robots are more suited to the task than humans. Since the robot is just executing a script, changes in parameters are negligible. A human worker folding sheet metal would need to take time and mentally process information to create variations in angle, whereas a robot could vary the angle without time to think.

Case Study: Robotic Metalworking

My first chance to approach robotic fabrication from a manufacturing perspective came as I was TAing a course at Georgia Tech called Robotic Fabricates: the class was intended to explore the advantages industrial manufacturing capability could bring to the production of architecture. Sheet metal was the material of choice, and the class engaged various experiments finding out how to interface with the robot while learning about the material. The robot’s ninety kilogram payload was suitable for deforming thin steel sheets, but the first obstacle was that we needed to engage the design of the robot environment to take advantage of this strength.

Two things needed to happen to prepare the physical environment for fabrication with the robot. We were cutting parts on the waterjet, but the robot needed a way to interface with the parts. The tool that goes on the end of a robotic arm is referred to as an end effector. In our case, we used a pneumatic gripper, sending signals to the robot when we needed to open or close the claw. We placed a table in the enclosure and marked locations where our parts would be located for the robot to grab. These simple steps are fundamental to any sort of manufacturing process: it’s always crucial to consider how the parts are being fed through the steps of the process, and each step needs to be designed to interface correctly with the parts. In this case where we were opting to use mechanical deformation to alter the shape of the parts, we needed a second grip to hold the parts while the robotic arm did the manipulation, a pneumatically activated vice clamp. This is a very rudimentary way to allow to machines to interact. The robot grabs the part, lowers it into the vice, the vice closes, and then the robot is able to fold or twist the part in the designated manner.

Georgia Tech 2017, Robotic Fabricates, Vice used as stationary grip for sheet metal parts.

Though the operations of folding or twisting here are very simple, these operations begin to reveal the numerous considerations that need to be accounted for when producing parts at volume. With the stainless steel we were using, we needed to account for spring-back in the folding. When the steel is folded, say to a 45 degree angle, the steel will spring back upon release, meaning that the resulting part is less than the desired angle. These are the same principles that must be accounted for when designing products to be mass produced with tools like punches or press brakes. Minor adjustments need to be made to compensate for the ways that the materials tend to respond to forces. If these factors are not accounted for, the tolerances will accumulate, leading to major issues or collisions of parts in larger assemblies. Furthermore, relief cuts and perforations had to be incorporated into the 2D geometry of the parts to allow them to successfully be formed into the desired 3D forms. Even with the measures that we took to ensure that the process was as consistent as possible, our set-up did not achieve true interchangeability of parts. Manual adjustments had to be made to correct for miscalculations in spring-back or minor misalignments that lead to unintended deformations.

In spite of the iterative troubleshooting nature of the process, the addition of an industrial robot to the process certainly expanded the scope of what a group of students could achieve. It’s certainly worth remembering that the initial research investment is always the highest cost in establishing a manufacturing process. By the end of this semester, the scripts and digital-physical calibration were consistent enough that our assemblies were not constrained by complexity or how many parts they had, but rather how much material we had available.

Case Study: Incremental Forming

In a separate research project following sheet metal folding, I explored one particular technique that the automotive industry makes wide use of that is almost entirely inaccessible to architecture. Incremental forming is the process by which sheet metal is mechanically formed into a precise geometry, usually by being pressed against a mold. Pressing thin metal against the mold deforms the metal into the shape of the mold, such as a car door or hood. This process makes it possible to stamp out enormous quantities of the same complex part over and over, perfect for automobile manufacturing. In the case of architecture, production volume rarely reaches a level where such a process is viable. One example would be Neil Denari’s Highline 23 project which uses incrementally formed metal panels on the façade, but it’s a luxury apartment building where the high cost could be justified. Similarly, on jobs like production of panels for stadiums or skyscrapers where there can be tens to hundreds of thousands of pieces, the cost per part is reduced and the process becomes more viable.

Here, I wanted to explore what it might require for this process to be modified for mass customization. For architectural applications, an incremental forming process would be more viable if it did not rely on a specific mold. Possible improvements would be reconfigurable molds or using punches to create perforations along the sheet for deformation to occur more easily. Either of these scenarios would allow architecture to have some of the benefit of incremental forming in creating complex geometries, with the added ability to create variation between different parts.

The alternative process I explored used auxetic cut patterns to increase the pliability of the sheet metal, reducing the force required for surface deformation. Unfortunately I did not have access to a punch to quickly perforate the sheet, so I used a water-jet instead for a proof of concept. Cutting very light gauge aluminum, the job took several hours for a single panel, so certainly not even close to viable. However, once the perforation pattern was cut into the sheet, the process proved to be quick. For the end effector, I drilled a hole into a lacrosse ball, stuck it onto a pipe, and had the robot hold the pipe. Then a simple steel frame was used to hold the sheet vertically. Once the robot started punching the aluminum sheet, the process was extremely quick. Though the process was extremely far from manufacturing feasibility, I was able to form the target geometry with no mold. Such a process, if deployed at scale, could allow for great variation in geometric complexity of large assemblies. The lesson from this experiment was probably that each process has a cost. By side-stepping the high barrier to entry of developing a mold, I instead incurred the cost that my parts were heavily perforated and relatively low resolution. These types of trade-offs might be necessary for such mass-custom procedures to be deployed in architecture within the near future.

Is it Worthwhile to Emulate Manufacturing?

Architecture is a slow discipline to adopt new technologies. Personally I suspect that architecture has some catching up to do in the changing landscape of manufacturing. However we can also consider that architecture has the opportunity to reflect on other disciplines’ explorations and selectively adopt the most successful parts, like aviation modeling software. I find it unlikely that architecture is going to have the opportunity to engage in mass production. At the same time, model-making techniques in schools have been evolving to incorporate 3D printing and laser-cutters, small steps toward normalizing mechanized production capacity in architecture. Large scale 3D printing services, such as those offered by Branch Technologies in Chattanooga, Tennessee are already available, allowing designers to simply print large portions of complex geometry. At the same time, these operations have not had the time or investment to develop the resiliency of established manufacturing sectors. If these techniques become affordable and especially if they can become environmentally friendly, then we very well could see widespread adoption.

For designers in school, I think it’s worth keeping manufacturing in mind. Some of the considerations that go into making parts manufacture-able can just as well be applied to architectural design. Moreover, numerous techniques and approaches to manufacturing exist that have not been deployed in architecture yet. Though they won’t all translate due to the different conditions faced by the architectural discipline, those that do could greatly expand the potential tool-set of architectural designers.

To Be Automated: Manufacturing