The SpaceX Starship booster landing is easily one of the top three aerospace moments of the decade. Watching a 232-foot-tall stainless steel tower perform a controlled flip and return to its launch site is a masterclass in modern engineering. But for aerodynamics and controls engineers, the question isn't just "Can we do it?" but "How can we make it more efficient, predictable, and robust?".
At Flexcompute, we wanted to push the boundaries of what is possible with physics simulation. By leveraging Flow360, we performed a high-fidelity, time-accurate Delayed Detached Eddy Simulation (DDES) of the Starship booster’s landing flip. We did it entirely without the use of an Overset mesh.
Performing a steady simulation at various altitudes during the landing motion requires making a lot of assumptions. However, for reusable launch vehicles (RLV) undergoing highly dynamic maneuvers like the landing flip, the flow is inherently unsteady, and the history of the motion affects the forces at a given altitude. Plus, due to the presence of ground, capturing the behavior of separated flow with steady simulation is not possible. High-fidelity dynamic simulations allow engineers to:
Capture Exact Force and Moment Histories: Rapid rotations create "memory effects" in the air, where the forces at a specific angle depend on the previous state of the flow.
Predict Pitch Damping: Accurately calculating dynamic derivatives is crucial for tuning flight control laws to prevent oscillations or loss of control during the flip.
Optimize Reliability: By mapping peak torques throughout the entire trajectory, teams can ensure structural margins and the motion of grid fin actuators are sufficient for even the most extreme recovery scenarios.
A major hurdle in Computational Fluid Dynamics (CFD) for moving bodies is how to handle the mesh. Most industry professionals would tell you that simulating a combined translation and rotation maneuver, like the Starship flip would be nearly impossible without an Overset mesh or without remeshing. Typically, Overset meshes involve overlapping multiple meshes, which can be computationally expensive and prone to interpolation errors at the boundaries. Remeshing is definitely not a practical option when you have a long trajectory to simulation in a limited time.
For this Starship simulation, we employed a unique methodology utilizing Nested Sliding Interfaces. The animation below shows three nested rotational interfaces around the starship geometry. The placement of these cylinders, their radii, and the rotational rates have been chosen such that when they move rigidly within each other, the starship moves on a trajectory very close to the actual trajectory that happens during the landing flip. To provide some more insight:
Translational Control: By placing two cylinders with slightly offset centers and rotating them in opposite directions, we can create a specific linear trajectory. The rotation of the outer cylinder moves the object in one direction, while the middle cylinder's counter-rotation cancels out the unwanted lateral movement, leaving a clean vertical or diagonal path.
Rotation Cancellation & Control: To perform the flip (pitching), a third inner cylinder is used. This allows us to prescribe the exact angular velocity needed for the booster to rotate independently of its translation through the domain.
This method allows for a single, high-quality unstructured mesh to be used throughout the entire simulation, ensuring that the solver maintains its speed and accuracy without the overhead of re-meshing or complex Overset connectivity.
The value of this high-fidelity approach is best seen in the load history. As the booster transitions from its belly-flop orientation to a vertical landing stance, the aerodynamic loads fluctuate rapidly.
Below is a representation of the force histories captured during the landing flip sequence.
The simulation reveals significant unsteady hysteresis, providing a 'gold mine' of data for refining landing logic. By capturing accurate aerodynamic insights, engineers can precisely tune dynamic derivatives and optimize maneuvers with greater confidence. The alternatives—expensive, time-consuming wind tunnel testing or steady-state CFD, often fail to capture the true dynamic nature of these forces.
While the Starship’s dynamic landing flip is notoriously complex to model, Flow360 simplifies the setup process without compromising on performance. Using the Flow360 Python API, we generated an 80-million-node mesh and simulated 30 seconds of physical flight time in just 1.5 hours using 8 x NVIDIA H200 GPUs. This rapid turnaround is transformative; it allows engineers to iterate through multiple trajectories and tuning parameters, which is critical for identifying potential failure modes and ensuring mission success.
At Flexcompute, our mission is to push the boundaries of physics simulation. By eliminating traditional bottlenecks, such as the inherent complexity of overset meshes and the prohibitive turnaround times of frequent remeshing, we empower aerodynamics and controls teams to model highly dynamic maneuvers with unprecedented speed and robustness.
Whether it is a Starship booster or the next generation of eVTOL aircraft, Flow360 delivers the high-fidelity physics required to conquer the most challenging flight envelopes.