AeroQubIt

Motivation

Intelligence, surveillance and reconnaissance (ISR) platforms require high endurance, making lift coefficient (CL), drag coefficient (CD) and lift-to-drag ratio (L/D) critical design parameters. Unlike conventional combustion-powered aircraft, electric propulsion offers lower noise, reduced thermal signature and lower maintenance. Distributed Electric Propulsion (DEP) further improves design flexibility but introduces complex aerodynamic interactions that are computationally expensive to simulate. This project investigates whether machine-learning surrogate models and quantum-assisted methods can provide reliable aerodynamic predictions at much lower computational cost while remaining consistent with CFD and experimental results.

The five stages

Geometric modelling & configuration comparison

The NASA X-57 Mod IV was selected as the reference (V0) and modelled in OpenVSP. Six configurations (V0–V5) varied motor number and placement while keeping the wing and fuselage fixed. They were analysed in VSPAERO under cruise conditions using CL, CD, and L/D. The reference model achieved 90.27% agreement with published NASA data, with a 9.73% mean absolute percentage error.

Machine-learning design optimisation

Using OpenVSP–Python integration, wing parameters (taper, twist, thickness ratio) and motor placement were swept into 486 configurations, each analysed in VSPAERO. A Random Forest surrogate (R² = 0.972, 5-fold CV) then screened 1,000 candidate designs, and SHAP analysis quantified each variable's effect. The model is served live via a Streamlit interface.

CFD & multidisciplinary performance analysis

The ML-optimised design was meshed with blockMesh/snappyHexMesh and solved in OpenFOAM at Re = 1.92×10⁶ using the k-ω SST turbulence model. Rotors were represented with the actuator-disk approach as momentum sources. The stage covered aerodynamic forces, rotor interactions, energy–power budgets, convective thermal management and a full flight-envelope assessment.

Quantum-assisted flow modelling

A quantum–classical hybrid represented the flow in a reduced 2-D state space. Ten airfoil sections (K0–K9) were solved with the Lattice-Boltzmann Method (D2Q9) for reference data, amplitude-encoded into Hilbert space, and fed to a PyTorch + Qiskit variational quantum circuit (Ry, Rz, CNOT gates) trained via the parameter-shift rule. A low-depth, NISQ-compatible circuit predicts CL and CD for a section in about 22 seconds, against 4.5 hours for a full solution, a ratio of 720. Accuracy across the ten spanwise stations averages 87 %.

Visualisation, experiment & prototyping

Results were turned into 3-D Blender animations of DEP integration and the CFD workflow. A 12×12×70 cm steam-flow wind tunnel visualised stall behaviour in 2° steps, and a 1:9.8 scale, 120 cm-wingspan prototype — 3D-printed PLA wings, foam-board fuselage, drone motors — formed a coupled numerical-experimental validation platform.