To Go Where No Army Has Gone Before
Among the many issues that surround the future of North American manufacturing, the Defense Advanced Research Projects Agency (DARPA)—the American military's research and development arm—has identified a new one. DARPA says that the design and development techniques now in use take too long, cost too much and are often incapable of building highly complex systems.
So DARPA, which among many other things invented the Internet, relational databases, speech recognition software, computer workstations and very large scale integration (VLSI) design, has accepted a new mission: reinvent manufacturing. To do this, it will use a technique familiar to geeks, nerds and techno-wizards called “crowdsourcing.” Along the way, DARPA stands to develop techniques with substantial implications for North American manufacturing as a whole.
Crowdsourcing aggregates brainpower by inviting perhaps thousands of very smart people to contribute to the solution of a problem using websites, such as SourceForge.net, as the intermediary. So far, crowdsourcing, which depends heavily upon the skills and goodwill of volunteers, has been used to develop open-source software, such as Mozilla Firefox, the second most commonly used web browser.
“In crowdsourcing, people volunteer their services and work in a geographically distributed manner,” says Scott Collison, president and CEO of Geeknet Media, which operates SourceForge.net, Slashdot, the “nerd news” site, and freshmeat.net, the locus for Linux-based software. “This works very, very well for software operating systems like Linux or a design application such as Gimp.”
An Infantry Vehicle
But DARPA doesn't plan to make only new software. It plans to demonstrate the power of its revolutionary approach to manufacturing by designing and building a next-generation infantry fighting vehicle. The armored vehicle will carry a squad of soldiers and their gear, be able to withstand explosives and a wide range of other weapons, and be fully networked into the modern Army's battlefield technologies.
Crowdsourcing will play a role, as will a series of linked DARPA innovations. They include an entirely new engineering and design metalanguage (a language used to talk about other design languages); the website vehicleforge.mil; a program to seed digital manufacturing skills among high school techno-whizzes; and a plan to construct the combat vehicle in a novel, all-digital factory setting that can “mass produce” the fighting machines with nearly infinite customization, with all changes made on the fly.
Dubbed the Adaptive Vehicle Make (AVM) program, the project is expected to take about four years. Coincidentally, the Army has launched a separate, conventional procurement program—the Ground Combat Vehicle (GCV) program—to make exactly the sort of next-generation infantry combat vehicle planned by DARPA. DARPA, which is part of the Defense Department and not any individual service, says its work has nothing to do with the Army's formal program to replace the Bradley Fighting Vehicle. The Army says its program has nothing to do with DARPA's work, except Army officials plan to watch closely DARPA's progress. The Army says it will require seven years to finish its GCV.
“The complexity of warfighting systems has increased, and as that has happened, the development times for systems have increased superlinearly,” Paul Eremenko, the DARPA program manager, said at a briefing for potential contractors late last year. “Development times have lengthened and gotten to the point in many domains, in ground, air and space, where next-generation systems are just being cancelled across the board. The timelines are unpalatable from a warfighting perspective.”
Engineers don't think about cost as a design factor, said Eremenko. Instead, they make tradeoffs among a machine's size, weight and power, which slow development. Further, existing manufacturing techniques invariably anticipate production of multiple prototypes, in which components developed largely in isolation from one another are assembled to see what happens. And what happens, says DARPA, is that “we discover emergent behaviors and unanticipated issues” as the subsystems interact.
“We pay attention in development to the thing itself, but not to the seams among components,” says Eremenko. “In aggregation, those seams become challenging. The more the seams, the worse the cost and time overrun.”
Take the F-35 Lightning, the fifth-generation tactical fighter aircraft program managed by Lockheed Martin with participation in system design and development by nine nations. A 2009 study by the United Technologies Research Center found that existing and widely used fourth-generation fighters, such as the F-16, have 15 subsystems, 103 “interfaces” and fewer than 40% of their functions managed by software. The F-35, which comes in three variations, has 130 subsystems and 105 interfaces, with 90% of its functions managed by software. So far, the F-35 program has required 10 years; testing is still underway, although nearly 100 of the lethal airplanes have been ordered and various nations have announced plans to order more than 3,000.
Despite the “exponential growth in complexity of modern aerospace platforms,” the United Technologies engineers said, “in many aerospace subsystems the design architecture and technologies have not changed for decades. The arrival of new architectures and technologies has not been accompanied by a commensurate advancement and adoption of the design processes and tools needed to develop these very complex systems.”
A Plan with Three Goals
The Adaptive Vehicle Make (AVM) program has three specific goals.
The first is to shorten substantially the development time for complex defense systems. If a modern weapons system, such as the Army's ground combat vehicle, now takes seven to 10 years to develop and prove, DARPA wants to reduce that process to one to two years, says Army Lt. Col. Nathan Wiedenman, the AVM deputy program manager.
Second, DARPA wants to shift high-value design activities away from managing unanticipated problems when subsystems interact and towards greater innovation throughout the process and product.
Finally, DARPA wants to “democratize the design process,” says Eremenko.
“If you look at the way the design of next-generation defense systems is done today, most industry niches and domains are oligopolies,” he says. “There's a handful, three, four or five if you're lucky, of big players that have the critical mass to produce systems of this complexity. Each of them has an 'A-team' dedicated to advanced concepts. But if you look at it in the aggregate, our total idea pool consists of maybe 200 brains. It's kind of sad and sobering. We want to increase the number of brains that contribute by many orders of magnitude.
“So here's what we want to know. Can we crowdsource defense vehicles, complex cyber-electro-mechanical systems? And can we democratize demand by grooming a new generation and a future cadre of designers and manufacturing innovators?”
Put another way, says Wiedenman, can DARPA change how major military systems are designed, produced and acquired?
The Tools for Change
The AVM program is divided into a number of different parts that, together, produce what Wiedenman calls “an integrated process that feeds logically from design through manufactured products,” which, in turn, will be “correct by construction.” That means, he says, that the system “fully captures cross-domain interactions before metal is bent.” Or, more simply, what DARPA expects is that the first vehicle won't be a prototype. It will be a working war vehicle, ready for use in all respects, with no bugs, unexpected glitches or other problems left to resolve.
The first part of the program, called META, requires creation of a kind of design language that includes metrics, a representation metalanguage, design tools and verification techniques that will permit construction of a vehicle. Eremenko says the language will help designers manage complexity. It will also be the language used by fabrication machines employed to actually construct the vehicle. The idea, in part, is to use the new language to move programming to “higher levels of abstraction,” or farther away from basic machine code so that engineers can focus on innovation, not coding drudgery.
With the new language, DARPA then plans to build a “completely programmable factory,” dubbed the Instant Foundry, Adaptive through Bits, or IFAB. Machinery used to build the combat vehicle will be controlled by the new engineering language developed by the META part of the AVM program. Every part of the production process will be flexible. An example, and far from the only one, might be a process similar to three-dimensional printing, in its infancy today, in which objects are made by a process similar to ink-jet printing, only in all three dimensions, with layer upon layer of material printed in precise configurations to build the object.
To bring together the many ideas about the vehicle, and to provide working teams a common area in which to exchange information, DARPA will build the vehicleforge.mil website. It will be the first open-source collaboration environment for complex cyber-mechanical systems.
“Opening our component model library and design tools to the world is intended to enable access to a community of sympathetic contributors that will outweigh the drawbacks of exposing this information to potentially harmful individuals,” says Wiedenman. “By rapidly turning a design into physical reality while minimizing post-manufacturing validation testing, AVM enables strategic surprise. People may see us coming, but they won't have the time to do anything about it.”
Finally, to seed enthusiasm for this entirely new type of manufacturing among highly intelligent young people, DARPA has devised Manufacturing Experimentation and Outreach, or MENTOR. Under MENTOR, the agency will send programmable manufacturing equipment, such as those 3D printers, to high schools. A series of design challenges, each with an associated prize for victory, will be issued to encourage both competition and collaboration among high school teams as they build cyber-electro-mechanical systems of “moderate complexity.” Eremenko says this might include go-carts, robots and “small, unmanned aircraft.” Eventually, DARPA hopes to have design teams active in as many as 1,000 high schools.
So, can it work? Geeknet's Collison says he doubts it. “I've never heard of anyone succeeding with anything nearly that complex,” he says. “Even in open-source software, for every 100,000 projects there are a single-digit number of projects that are successful.”
Some existing open-source vehicle projects confirm the difficulties. OScar, a European effort to develop an open-sourced car, is now 12 years old and still hasn't decided whether its vehicle should have three wheels or four. Organizers, in a news snippet entitled “Back from Hell,” confess that they “underestimated the expenditure of time and also the technical and legal conditions required for success.” Riversimple, a 10-year-old U.K. open-source effort to build a two-seat vehicle powered by hydrogen fuel cells, is looking for new funding. “It takes longer and goes slower than we thought it would,” says Patrick Andrews, the Riversimple corporate secretary and leader of its 40 Fires open-source website.
So, can DARPA do what others cannot? It has a track record of doing so, including in areas related to the AVM project. In 2009, for example, DARPA tested one aspect of crowdsourcing. It launched and then moored 10 red balloons in unannounced locations across the nation, from Waterfront Park in Portland, Oregon, to Collins Avenue in Miami, Florida. A $40,000 prize was offered to the first person or group to report the correct locations of all 10 balloons. The winner, the Red Balloon Challenge Team from the Massachusetts Institute of Technology, correctly identified the location of all 10 balloons in eight hours and 52 minutes using a crowd — 5,400 people recruited and monitored using social media.
Much closer to the task at hand, a team from the University of California, Santa Barbara, won DARPA's $50,000 digital manufacturing challenge. The goal was to develop models to predict the maximum compressive load that could be supported by digitally manufactured titanium spheres and polymer cubes.
“I'm excited about this,” says Wiedenman of the AVM project's potential. “Look at SourceForge and the Linux operating system. People migrated all over the place, shifting from one aspect to another. It's chaotic. But from an innovation standpoint, it's very good.”