Tag: engineers

Rocket propulsion and other next-gen aerospace systems increasingly depend on GenAI models—a force for democratizing design.

By Greg Zacharias, Aerospace R&D Domain Lead and Executive Producer, AIAA SciTech Forum.

Originally published in the November issue of RAeS AEROSPACE.

From nuclear-thermal rockets to hypersonic aircraft, today’s aerospace systems are increasingly complex, relying on lighter-weight 3D-printed materials, as well as advanced structures, which can include a mix of different materials and thermal-management technologies. The control over form offered by 3D printing means that these components are exceptionally complex, requiring aerospace engineers to develop innovative design approaches. Not surprisingly, some of the most promising approaches tap into generative artificial intelligence, or GenAI, which will be featured at the upcoming 2025 AIAA SciTech Forum in January in Orlando, Florida.

“GenAI is more than just ChatGPT; it has applications in engineering design and it’s going to be used in critical engineering components in the not-so-distant future,” says Zachary Cordero, the Esther and Harold E. Edgerton Associate Professor in MIT’s Department of Aeronautics and Astronautics, who will present in two sessions at the forum. GenAI systems leverage vast datasets to autonomously generate novel solutions and designs, enhancing innovation and applications in diverse fields.

“GenAI is extremely powerful if you have a lot of data,” notes Faez Ahmed, Assistant Professor of Mechanical Engineering, who leads the MIT Design Computation & Digital Engineering (DeCoDE) Lab in the MIT Center for Computational Science and Engineering (CCSE), an interdisciplinary research and education center focused on innovative methods and applications of computation.

The lack of data for learning models – the oxygen that fuels GenAI training – is the biggest bottleneck, Ahmed adds. “Whenever someone says GenAI doesn’t work, a lot of times it’s not the model; it’s the lack of data.”

The DeCoDE Lab bridges this gap by creating design datasets, often by performing a lot of high-fidelity engineering simulations, including recent work for the automobile industry. The Lab created one of the largest and most comprehensive multimodal datasets for aerodynamic car design named DrivAerNet++, which comprises 8,000 diverse car designs modelled with high-fidelity computational fluid dynamics simulations.

Ahmed emphasises that his MIT team doesn’t always use data from good designs but also develops methods to leverage negative data, since bad designs “are cheap and much easier to get.”

Cordero’s Aerospace Materials and Structures Lab at MIT is pushing the boundaries of additive manufacturing for spaceflight through developing new processes and materials. Cordero is collaborating with Ahmed and MIT Research Scientist Cyril Picard on a US Department of Defense-funded research project on the design of next-generation reusable rocket engines.

According to Picard, the team is using GenAI to assess mechanical and thermal properties of materials to inform the design of 3D-printed multi-material parts, with the “long-term goal of making the engines more high-performing, efficient and lighter.”

Looking across the aerospace sector, GenAI offers many benefits, from optimising materials to reducing costly late-stage design changes when scaling production to enabling rapid validation and qualification, say the researchers.

To Ahmed, the biggest benefit of GenAI goes beyond making better products faster: it affords the time for people to explore new designs while also opening up design to innovators outside of traditional aerospace fields.

“I’m personally really excited about this idea of democratisation of design. Historically, design has been limited to the headquarters of major industries. But with tools, like GenAI, we can tap into the creative potential of people with good ideas, but who aren’t necessarily experts.”

Cross-Industry Collaboration Needed to Advance Autonomous Systems in Air & Space

By Greg Zacharias, Aerospace R&D Domain Lead and Executive Producer, AIAA SciTech Forum.

Originally published in the October issue of Aviation Week.

Agility matters when designing new capabilities like autonomous aircraft. So does thinking and partnering non-traditionally—it can lead to breakthroughs.

Who would have thought that the Secretary of the U.S. Air Force would fly on a X-62 VISTA fully controlled by a neural network? By working outside the box, the Air Force Research Laboratory (AFRL) and DARPA partnered together and pulled it off in only five years, not decades.

For the latest flight test in May, U.S. Air Force Secretary Frank Kendall flew on the aircraft, configured to behave like an F-16, where the modified fighter jet performed dogfighting maneuvers autonomously on par with an experienced F-16 pilot.

The feat left an impression on Dr. Kerianne “Yoda” Hobbs. “It changed how I view technical development. It’s non-traditional partnerships and integrated teams that are most effective,” says Hobbs, who serves as the Safe Autonomy Lead at AFRL.

Hobbs is part of AFRL’s Autonomy Capability Team (ACT 3), in which government researchers work directly with large and small businesses and university researchers to change the paradigm of how the Air Force innovates when it comes to autonomous air and space vehicles. They’re using a technique called reinforcement learning to train a neural network to control physical and digital systems.

Hobbs hopes to bring this same spirit of collaboration in aerospace autonomy to a new cross-industry task force.

New Roadmap for Aerospace Autonomy

The American Institute of Aeronautics and Astronautics (AIAA) Autonomy Task Force brings together all sectors of the industry—large and small commercial companies, government agencies, and academia—to drive faster and better collaboration in autonomy innovation across the air and space domains. Its initial focus includes three key functions of autonomous systems: sensing and perception; reasoning and acting, such as verifying that an autonomous entity performed within its delegated and bounded authority; and collaboration and interaction. In this last functional area, multiple autonomous agents such as a constellation of space vehicles may work together to navigate around each other.

The timing couldn’t have been better with the rise in advanced air mobility, a growing commercial presence in space, and rapid developments in defense systems.

Defining Autonomy in Aerospace

What is meant by aerospace autonomy? There isn’t an agreed-upon definition across the industry. The task force’s working definition is “a robotic air or space system set to achieve goals with delegated and bounded authority while operating independently or with limited external control.” Autonomous aerospace systems are categorized as either safety critical or mission critical, with the former applying when humans are involved and the latter more applicable to a robotic mission. Unlike traditional robotics that perform a single task, today’s autonomous vehicles need to be adaptive to a broken tread or a flash of sunlight on a sensor, but not so independent that the system would deviate and compromise the mission. The emphasis is on setting a boundary around what the autonomous system is allowed to do and how it’s allowed to operate. Technologies such as run time assurance are useful tools to enforce boundaries on autonomous system behavior. Trust plays a role as well, especially in human-autonomous interactions.

Speed and Lessons from Aerospace

A key research gap that the task force hopes to address is in the area of verification and validation (V&V) systems and processes that are cost-effective to implement. While the space domain has a long history of conducting extensive V&V of semiautonomous systems, the air domain is gaining ground in part because of the test opportunities, where getting a quadcopter or a small UAV or even an F-16 is significantly cheaper than procuring a space vehicle for an autonomy test. “Your opportunities to test are few and far between,” says Hobbs of the space environment. The current pace of autonomous system development remains a major concern for Hobbs. Even witnessing the AI-enabled F-16 test flight, which occurred on the AFRL’s VISTA test platform, pinpointed the limitations of current testing. “I realized everything I knew about traditional V&V wasn’t going to help us use this technology fast,” she recalls. Hobbs is challenging her team to ask themselves, “What is the path forward to do this quickly and competitively without compromising safety or mission? What is the right-size approach?”

Lessons from Computing

Aerospace autonomy builders also should embrace the computing industry’s market approaches that focus on the idea of a “minimal viable product,” says Hobbs. Instead of ensuring that all requirements are correct in the beginning of a program, teams can make a small investment as fast as possible to get from the requirements and development phases to ground and flight simulations more quickly. In this way, groups can learn quickly and iterate better autonomous system designs.

High Stakes for Getting Autonomy Right

Much is riding on getting aerospace autonomy right. “We need a strategy to fully harness these technologies. Without it, we risk other countries moving ahead,” warns Hobbs, noting that unequal access to autonomy breakthroughs within the commercial sector could also harm U.S. competitiveness. “The goal is for aerospace to continue to evolve. It’s going to take a tight-knit community across big industry, small industry, government, and academia working together to speed up the development process to catch up to other industries,” she concludes.