Field Test in Turbine Construction Can Artificial Intelligence Design an Aircraft Engine?

Related Vendors

FAULHABER Drive Systems

Taiwan Machine Tool & Accessory Builders' Association (TMBA)

SIMTOS (Seoul International Machine Tool Show)

Generative AI and large language models (LLMs) are transforming software development—but are they also suitable for complex hardware? An experiment at the Massachusetts Institute of Technology now shows that AI co-pilots can drastically shorten the design process. Ultimately, however, human engineering expertise remains the deciding factor.

In the mechanical and plant engineering sector, there is a great deal of discussion about the potential of artificial intelligence. To find out whether AI can actually shorten the traditional development, construction, and testing cycle for physical products, the Massachusetts Institute of Technology (MIT) Gas Turbine Laboratory launched the so-called JARVIS Challenge (Jet-engine AI Research and Validation Intensive Sprint).The task for the participating teams: Within just four weeks, they were to design, manufacture, assemble, and test a small gas turbine aircraft engine (single-shaft jet engine).
The specifications were strict: 50 to 100 pounds of thrust (approx. 222 to 444 newtons), operation with Jet-A kerosene, and the ability to withstand five 60-second test runs. The designers had complete freedom in terms of design, material selection, and manufacturing processes.

Design Tools Meet Unlimited AI Access

In addition to traditional engineering tools such as SolidWorks for CAD design, Abaqus for FEM analysis, and Concepts NREC for turbomachinery design, the teams had access to a newly developed AI interface called MIT Parley. This interface integrates state-of-the-art large language models (LLMs) such as ChatGPT and Claude. Thanks to industry sponsors—including Safran and Boom Technology—the teams had virtually unlimited access to these AI resources. The question: Can AI compensate for the lack of in-depth experience in thermodynamics and fluid mechanics?

Can AI compensate for a lack of in-depth experience in thermodynamics and fluid mechanics?

Concept Development Accelerated, CAD Design Remains a Bottleneck

In the first phase of the project, AI proved to be a massive accelerator. The teams used the language models for literature reviews, learning CAD and CAE software, creating Excel spreadsheets, and, above all, for concept studies (comparing variants). However, when it came to detailed CAD design and prototyping of the combustion chambers in the second week, artificial intelligence reached its limits. The teams found that AI-typical “hallucinations”—the blind confirmation of false assumptions and a lack of basic physical understanding of the models—slowed down the design process.“AI is a helpful tool. It’s great at finding and organizing information, but it can’t design,” sums up Elizabeth Tupaj, a member of the eventual winning team. “The moment the engineer doesn’t know exactly what’s going on and the AI takes the helm, the design becomes unreliable.”

AI is a helpful tool. It's great at finding and organizing information, but it can't create anything.

Elizabeth Tupaj, Member of the eventual winning team

Procurement and Manufacturing Remain the Bottleneck

Another hurdle that the algorithms couldn’t overcome was supply chain management. Although the AI suggested manufacturers during the supplier search, these companies were not interested in collaborating given the tight four-week timeline. The teams quickly realized that personal relationships and direct contact with contract manufacturers were far more effective than AI-generated lists. Physical manufacturing—rather than technical design—ultimately remained the project’s rate-limiting factor.

Domain Expertise Beats Blind Faith in AI

The outcome of the challenge was particularly revealing: Although the team that made the most extensive use of AI for the design made the fastest progress and achieved an initial ignition, it failed the final hot-fire test because the rotor rubbed against the housing and seized up.Instead, the victory went to the team that had more traditional CAD and turbomachinery expertise from the outset. This team was significantly more skeptical of AI and relied primarily on fundamental engineering knowledge (“first principles”) and established calculation methods. Their engine started smoothly, switched to kerosene operation, and generated the required net thrust.

AI Is Just a Multiplier

“The JARVIS Challenge has shown that AI can significantly accelerate the development of safety-critical hardware. However, technical judgment remains the decisive distinguishing factor,” explains Professor Zolti Spakovszky, director of the MIT Gas Turbine Laboratory.Accordingly, a modern designer is not primarily distinguished by the fact that they use AI, but rather by their ability to manage it competently. They must know when to trust the outputs, when to critically question them, and how to translate AI concepts into hardware that can actually be manufactured. The ideal balance, according to the MIT faculty, lies in possessing enough domain knowledge to maintain control over the CAD model, while at the same time being open-minded enough to use AI efficiently during the research and conceptual phases.

A modern design engineer must know when to trust the specifications, when to critically evaluate them, and how to translate AI concepts into hardware that can actually be manufactured.