DeM Banter: something will have to give…
Aviation Week & Space Technology
November 5, 2012
By Graham Warwick
No one could accuse the U.S. Defense Advanced Research Projects Agency of lacking imagination. But to rethink the entire design, development and manufacturing process used by aerospace for decades is ambitious, even for the agency that helped bring us GPS, the Internet and stealth.
Under the Adaptive Vehicle Make (AVM) program, Darpa is working to combine model-based design, virtual collaborative engineering and foundry-style manufacturing into an end-to-end process that cuts development timescales by a factor of five, eliminating the design-build-test-redesign cycle that is driving costs and delays.
“It’s safe to say that the direction we have been going in military acquisitions is not a sustainable path,” says Lt. Col. Nathan Wiedenman, Darpa AVM program manager. “We simply can’t continue to spend more and get less for the money we spend.”
He points to a 2010 review that concluded the U.S. Army had spent $22 billion over the previous 15 years on programs that were eventually canceled — and to “Augustine’s Laws.” In 1984, former Lockheed Martin Chairman Norman Augustine looked at cost trends and predicted that, within 50 years, the entire U.S. defense budget would buy one tactical aircraft. “He wrote that book 20 years ago and we have hewed to those trends ever since.”
Darpa blames the limitations of today’s design methods. These break a system down along engineering disciplines (structural, thermal, etc.), optimize the parts and subsystems for their individual tasks, then put them together. “After we build it, we test it, to see if it works the way we expected. Invariably it doesn’t, because all those components and subsystems interact in ways we couldn’t anticipate.”
Nothing is just a power or thermal system. “All have behaviors that affect the components and subsystems around them. Things overheat or vibrate loose because it would take too much computation to predict and evaluate all those interactions. That means we have to go back and redesign, rebuild and retest,” he says.
The problem is complexity, Darpa believes. “Even though the systems we are building are much more complex, the way we engineer those systems has not fundamentally changed for about 50 years,” Wiedenman says. Industry is embracing model-based design, but AVM takes another step, using component models for design, context models for virtual testing, and manufacturing models to provide automated feedback on cost and schedule.
“There is no way to know how a complex system works until you build it. We need new approaches to design and modeling that enable us to predict those interactions,” he says. AVM aims to capture the interactions and understand how a system works before it is built. “We are building tools to enable a designer to create a ‘correct-by-construction’ system, meaning that when we build it, it works the way the design predicts, first time.”
Darpa is following the lead of the semiconductor industry, which broke through the complexity barrier decades ago. “The integrated-circuit industry went through this same transformation 30 years ago by building a set of tools called electronic design automation,” Wiedenman says. Using automated tools, “integrated-circuit designers are able to create correct-by-construction designs, to maintain the 24-36-month product cycle that has helped them sustain massive growth.”
AVM begins with the Meta automated tools, which raise the level of abstraction in the design process, in the same way high-level programming languages have for computers. As a software engineer works with applications and not electronic circuits, so the aerospace designer can work with systems and not individual parts.
Meta is “a design flow for cyber-physical systems where hardware and software components are interlinked,” says Janos Sztipanovits, Meta principal investigator at Vanderbilt University. “Physical and computational systems can be co-designed, with trade-offs between hardware and software.”
Using Meta tools, a designer takes component models from a library and assembles them into a system. “A model is a full representation of that component with its surrounding environment: other components, inputs/outputs, heat, vibration, behaviors, etc. It’s a meta-model, not just a drawing.”
Capturing every part down to the screws and modeling how they interact would take decades of computation. “We are looking at ways we can intelligently use levels of abstraction so I’m not dealing with every bolt, I’m just looking at how this engine couples to this transmission, couples to this final drive, to this suspension,” he says.
Meta allows designers to evaluate rapidly hundreds of thousands of concepts to meet a requirement. “Traditionally we start out with what we think a vehicle should look like, then iterate around that one particular design, so we’re only exploring a very small region of the design space,” says Wiedenman. “We’re providing tools that allow you instead to explore the entire design space, without taking decades of computation.”
Models can be parametric, “so we can vary things to create a design,” and progress through different refinement layers as the design becomes more detailed. “I’ve defined at a high level how these things interact, so I populate a design trade space based on the component options I have,” he says. “Then I use fairly low computationally intensive analysis tools to cull that trade space down to a few designs that need highly complex analysis.”
Using context models to simulate the operating environment, an entire family of concepts can be tested against the requirements. “This allows them to refine their design and further explore the trade space to find where an unusual design might better satisfy the requirements,” Wiedenman says.
Manufacturability feedback during design is a function of the Instant Foundry Adaptive Through Bits (iFAB), which will “receive the design, determine if it can be made, how long it will take and what it will cost,” says Mike Yukish, iFAB principal investigator at Penn State University. “Even for a simple widget there are lots of ways to make it, each with a different cost versus schedule.”
IFAB is an information architecture. “Rather than a single factory built around something, it is a distributed reprogrammable manufacturing capability,” says Wiedenman. IFAB will take the design representation in Meta, select manufacturing processes and equipment, sequence product flow and production steps, and generate computer numerical codes for machines and instructions for human workers.
IFAB will receive the design and flow it down the supply chain to get realistic cost-versus-schedule trades back to the designers. “The analog is Google Documents. It spell-checks as you go, but the spell-checker is not on your computer,” Yukish says. “On the manufacturing side, iFAB is constantly responding during design — push a button and you get a cost and schedule assessment.”
Here AVM bumps up against reality. “You cannot take an arbitrary design file and tell how much the part will cost. You only know when you have a process to make it,” he says. “So we parameterize and pre-negotiate between the designer and manufacturer. It’s what they do in the chip industry, get together ahead of time and agree on materials, general topology, etc. By restricting the design space they can do incredibly fast design and manufacture without iterations,” Yukish says.
Restricting designers to parameters that allow automatic manufacturability assessments “determines the type of vehicle you can build,” says Yukish. “It’s about living within our means, instead of pushing technology and getting cost growth.” This could limit the ability to optimize the design, but Darpa plans to tap the creativity of a wider pool of designers using a web portal to enable collaborative development .
“We’ve learned from the semiconductor and software industries that when you do this, you open the aperture for innovation by increasing the number, diversity and speed of those who can contribute,” Wiedenman says. In aerospace , limitations imposed by current tools ensure “the only entities able to contribute to the design are corporations with the wherewithal to build multimillon-dollar prototypes. That limits us to a handful of companies and maybe a few hundred brains.”
“There are a lot of people out there with those skills who don’t work for that handful of large companies. Creating tools that allow an engineer to understand how a system works before building it opens the aperture to a much broader collection of brains that we can bring to bear in designing complex systems,” Wiedenman says.
“If we only use architectures that can be automatically assessed for manufacturability, the crowd can help us overcome the limits of living within that constraint through the innovative assembly of easy-to-assess components,” says Yukish. “Manufacturing constraints force us to be simpler and the crowd means we have enough people with expertise staring at the design.”
“What we are trying to achieve is the ability to reach an optimal design much faster, before there is opportunity for requirements to creep and the threat environment to modify,” says Wiedenman. “It’s not about high performance, it’s how quickly it can come out,” says Yukish. “What’s Darpa-hard is exercising discipline, staying within what we can do. The program ‘s absolute insistence on schedule is unique.”