Conceptual aircraft design often involves a frustrating bottleneck: traditional CAD tools are too rigid to evaluate dozens of geometric variations programmatically, while low-fidelity aerodynamic tools lack the geometric feedback needed for structural integration and manufacturing.
During my summer internship at nTop, I set out to bridge this gap by building an automated Multidisciplinary Design Optimization (MDO) workflow for unmanned aerial vehicles (UAVs). My goal revolved around building a conceptual sizing workflow robust enough to produce flight-worthy aircraft.
Over the summer, I designed three distinct aircraft configurations, with our Blended Wing Body (BWB) serving as the ultimate test of the design loop - a novel configuration none of us had prior experience with. By combining nTop’s implicit modeling engine, automated execution via nTopCL, and aerodynamic solvers including AVL and AeroSandbox, I progressed from high-dimensional design space exploration to a physical, flight-tested aircraft that successfully demonstrated the gentle handling characteristics targeted by the design.
Integration with AVL for instant aerodynamics feedback
Blended Wing Bodies feature continuous curvature transitions between the central lifting body and the outer wing panels. Due to this seamless blend between the centerbody and outer wings, accurately representing such an aircraft in aerodynamic analysis tools such as AVL becomes tedious and error prone. To eliminate these geometric representation failures, I built a custom Aircraft Slicing Tool in nTop that leverages implicit section bodies to robustly slice the outer mold line at defined spanwise stations.
At each extracted station, my workflow evaluates the implicit cross-section, determining metrics including leading-edge coordinates and chord length. The sliced airfoil profiles are then formatted and saved as standard .dat files. To postprocess this data, I created a Python script that generates an .avl geometry definition file using the extracted leading-edge coordinates and chord lengths, accurately representing the 3D lifting surfaces for Vortex Lattice Method (VLM) analysis.
To execute the solver, the script automatically generates an AVL run case configured with flight conditions specified as variables in nTop, including flight speed, angle of attack, and aircraft Center of Gravity (CG) position. Once AVL completes the aerodynamic evaluation, nTop reads the raw output text files directly, parsing aerodynamic coefficients and stability derivatives (including CL, CD, Cm, Cma, Cnb, Clb and static margin) and displaying the resulting metrics directly within the nTop user interface for real-time visual feedback.
The image below shows the comparison between the additive-manufacturing-ready implicit geometry in nTop (right) and its corresponding AVL geometry (left). The second image demonstrates the same concept applied to a different geometry using AeroSandbox instead of AVL.
Automated Design Space Exploration with nTopCL
Before attempting local numerical optimization, I needed to understand the broader behavior of the BWB design space and establish physical limits for my design variables. Attempting to optimize an unconstrained geometry directly often leads to non-physical solutions or optimizer divergence. To prevent this, I conducted a global Latin Hypercube Sampling (LHS) Design of Experiments (DoE) across 100+ parametric configurations. The pipeline automatically swept key geometric variables defining the aircraft’s geometry, driven headlessly by a Python script via nTopCL.
For each set of input parameters, nTopCL generated the candidate geometry and executed the slicing script, which was then fed into AVL to generate aerodynamic and stability data. Geometrically infeasible configurations - such as those with taper ratios causing the wingtip to collapse - were automatically discarded during the geometry generation phase. I then postprocessed this data to visualize trade-offs between aerodynamic efficiency and stability derivatives. This visualization allowed me to filter configurations and eliminate those that did not meet longitudinal and lateral stability requirements, mapping out a clean, feasible design space.
Optimization Problem Formulation
With the feasible design envelope mapped out by my global LHS sweep, I moved to local numerical optimization to fine-tune the BWB’s planform geometry and flight trim. To solve this problem numerically, I built an MDO script in Python using AeroSandbox, decoupling geometry parameterization from the aerodynamic solver.
Rather than optimizing a simple trapezoidal wing, I parameterized a multi-panel lifting surface tailored specifically to blended wing bodies. I defined a root chord croot and a total span multiplier bmult, splitting the wing into two main spanwise panels. Each panel features independent leading-edge and trailing-edge sweep angles to allow a continuous geometric transition between the high-sweep centerbody and the lower-sweep outer wing.
At the centerbody, I utilized a modified NACA 23118 airfoil with a 1.4x thickness scaling to accommodate payload and electronics volume, blending into a thinner NACA 0014 profile at the wingbreak and tip sections. I also incorporated a fixed 3-degree geometric washout angle along the outer panel to mitigate tip stall risk. Finally, to add lateral stability for this tailless configuration, I added an integrated winglet defined by height, sweep, taper ratio, and a 10-degree cant angle.
Evaluating BWB efficiency accurately requires capturing both induced vortex dynamics and viscous skin friction. Standard VLMs excel at predicting 3D circulation, lift slope, and stability derivatives, but completely ignore viscous effects. Conversely, simple empirical drag models lack dynamic pressure feedback across complex planforms. To solve this, I implemented a hybrid aerodynamic engine combining two complementary solvers: VLM and AeroBuildup.
I used a 3D Vortex Lattice Method to evaluate 3D panel circulation and compute inviscid lift, induced drag, pitching moment, neutral point location, and dynamic stability derivatives. Alongside VLM, I integrated the AeroBuildup engine to estimate viscous profile drag and skin friction drag based on local station Reynolds numbers and cross-sectional thickness distributions. I then coupled both engines into a unified drag polar by summing the induced drag from VLM and the viscous drag from AeroBuildup.
Rather than setting the objective function to simply maximize the lift-to-drag ratio, I formulated the problem to directly minimize cruise power draw. Maximizing L/D in isolation often forces optimizers toward impractically low flight airspeeds where induced drag is minimal, producing an unviable aircraft and flight conditions.
To ensure the aircraft performs reliably throughout its entire flight profile, I evaluated candidate geometries simultaneously across two distinct flight conditions. The cruise condition was evaluated at 200ft across an allowable airspeed range of 15-30m/s, while the takeoff condition was evaluated at sea-level atmospheric conditions and appropriately constrained to prevent takeoff stall. The complete mathematical formulation of the optimization problem is structured as follows:
The Product: Flight Test Result
The optimization script passed these exact numerical outputs back into nTop, updating the implicit model and establishing the foundation for detailed design. With the optimized geometry finalized in AeroSandbox, transitioning from a digital concept to a physical, airworthy prototype was frictionless thanks to nTop’s implicit modeling engine. Because the master geometry was defined implicitly, we were able to generate internal structural features automatically without manual surface repair or broken CAD references. The parametric nature of the implicit workflow meant that any minor last-minute adjustments to the outer mold line were immediately manufacturing ready.
The ultimate test of my computational aircraft design pipeline was taking the assembled aircraft to the field for flight testing. On its first launch, the BWB aircraft, dubbed Vlad the Inhaler, demonstrated outstanding stability, handling, and pitch control as soon as it took off. The aircraft flew smoothly without oscillation and required minimal pilot intervention for steady and level flight, proving that the low-fidelity aerodynamic models accurately captured the vehicle's real-world flight dynamics.
Most remarkably, the aircraft achieved steady, hands-off cruise flight with virtually zero elevator trim deflection. This zero-trim flight performance directly validated my optimization formulation, confirming that the stability bounds, and trim constraints calculated through my conceptual sizing loop translated seamlessly into physical flight. Achieving a successful test flight on the first attempt with a highly stable and well-performing aircraft demonstrated the power of a complete aircraft design loop connecting implicit geometry and gradient-based optimization.
