New AI Simulates Jet Engine Airflow 10,000 Times Cheaper
The same AI trick behind ChatGPT could make engines cheaper to design.
Jet engines and gas turbines are built around rows of carefully shaped blades that spin in extreme heat. Designing each blade means testing how air moves around it, and that normally requires a physics simulation called CFD (basically, a computer wind tunnel). These simulations are accurate but painfully slow — engineers run them thousands of times, tweaking a curve here and a thickness there. A team publishing in the Chinese Journal of Aeronautics has now shown an AI can stand in for much of that work.
Their trick is a two-step approach. Most AI models try to guess the final answer directly — efficiency, pressure ratio — and get it roughly right. This one instead predicts the basic physics quantities underneath: temperature, pressure, and density throughout the airflow. From those, any number of performance measures can be calculated afterward. That makes it behave less like a one-answer calculator and more like an actual simulator a designer can query repeatedly. The method uses a transformer (the same underlying technology as ChatGPT) combined with a neural operator (AI trained to mimic physics equations).
On a well-known test case, a transonic compressor blade called Rotor 37, the model predicted efficiency, mass flow, and pressure-ratio patterns more accurately than two leading rivals, FNO and DeepONet. The bigger headline is speed: the paper reports downstream tasks like sensitivity analysis and design optimization running about 10,000 times cheaper than conventional simulation.
The catch is real. This AI was trained on one blade geometry in a simplified 2D slice of the engine. Give it a radically different design and it must be retrained, and it can't yet be trusted to replace final safety testing. Think of it as a fast sketch artist, not the certifying engineer — it narrows thousands of options down to a handful worth simulating properly.
- The AI mimics a 'computer wind tunnel' for turbine blades, predicting airflow physics rather than just guessing final numbers.
- It cut the computing cost of design tasks by roughly 10,000 times — the equivalent of a job that took weeks now finishing in minutes.
- It only works on blade shapes it has already studied, so real-world testing and certification aren't going anywhere yet.
Why It Matters
Faster, cheaper engine design could mean more efficient jet travel and power plants — and lower fuel bills over time.