Ghali Tsouli I design complex parts — and I prove they work.

A wind turbine blade sheds a strong vortex off its tip, and the drag that comes with it costs you energy you would otherwise capture. Birds have the same problem. The Andean condor deals with it by spreading its primary feathers so that one strong vortex becomes several weak ones. I wanted to know whether that still works on a 61 metre blade, so I tested it three separate ways: structurally, computationally, and on real hardware.

Design

The blade came first, and rebuilding it properly took longer than the winglets did. The NREL 5-MW definition gives you seventeen radial stations, each with its own chord length, twist angle and aerofoil profile. I placed an aerofoil section at every station, scaled each one to its local chord, rotated it to its prescribed twist, and lofted a surface through the set. The profiles are not all the same: three cylindrical root sections, then DU40 at node 4 running through DU35, DU30, DU25 and DU21 toward mid-span, then NACA 64₂24 from node 12 out to the tip at 61.63 m. Getting section-to-section continuity right at the profile transitions mattered more than anything else, because a discontinuity there produces numerical artefacts in both the FEA and the CFD rather than an obviously broken model.

Chord falls from root to tip to trade structural stiffness against drag, and twist is set so each section sits near its best angle of attack across the operating range. I implemented both distributions straight from the published data without adjustment, so the baseline keeps its original aerodynamic behaviour and any change in the results belongs to the winglet rather than to me.

On top of that I built three winglets differing only in sweep angle: 10°, 25° and 40°. Height stayed at 1 m throughout, roughly 1.6% of blade length and in line with published full-scale practice. Cant angle, aerofoil profile and junction geometry were fixed across all four configurations, which is what lets the results be attributed to sweep and nothing else.

Structural check first

There is no point improving the aerodynamics if the blade cannot carry the load, so I checked that first. All four configurations were run in SolidWorks Simulation at an extreme aerodynamic load of 36,569 Pa using GFRP properties. Peak von Mises stress came out at the mid-span chord transition rather than at the root or the winglet junction, which makes sense once you consider that this is where a high bending moment meets the widest section.

CFD methodology

The domain is a 150 m hemispherical inlet with a 250 m wake extension and a body of influence zone for refinement. Ten inflation layers bring y+ down to roughly 1, which the low-Reynolds k-ω SST formulation needs, and an 18° curvature setting puts cells where they are needed at the leading and trailing edges without me placing them by hand. My first attempt used 15 inflation layers and produced non-positive volume cells at the winglet junction; dropping to 10 fixed it in every case. For mesh independence I ran three refinement levels and saw under 2% variation in C_L and C_D between the medium and fine meshes.

I used two convergence criteria rather than one: scaled residuals down to 10⁻⁶ on velocity, and lift and drag monitors holding within 0.1% over the last 50 iterations. Residuals flattening out on their own is not enough, because the forces can still be moving when they do.

Results

What the winglet actually does

Q-criterion iso-surfaces show what is going on. On the baseline blade there is one tightly wound helical vortex tube. Add the swept winglet and it splits in two: a primary core at the blade-winglet junction and a secondary one at the winglet tip. Two weaker structures instead of one strong one, which is the same thing the condor's feathers are doing. Drag the handle to compare.

Testing it on the rig

CFD results are worth more once something physical agrees with them. I designed three winglet sets for printing, all NACA 0015, 12 mm tall, with sweep offsets of 2.1, 5.6 and 10.1 mm, then FDM printed them and bonded them to a GUNT ET210 rig blade. Print orientation mattered because I wanted the layer lines off the aerodynamic surfaces, and the bonded root joint had to hold up to 3,000 rpm without letting the incidence shift, so this was as much a manufacturing problem as an aerodynamic one. I took 16 steady state points per configuration between 200 and 3,000 rpm at 10 m/s and repeated the runs. Variation between repeats was small compared with the differences between configurations.

The brief

A diaphragm seal is the thin metal membrane that separates a pressure instrument from whatever it is measuring. How stiff it is, how much stress margin it has and how it behaves with temperature all feed directly into the accuracy of the instrument and how far it can safely be pushed. PCI wanted to know which of four corrugated geometries between 32 and 57 mm, all in roughly 0.05 mm cold rolled 316L foil, suited which application, and where each one ran out of road.

Decisions I had to justify

Outcome

Client work — figures withheld. Geometry, plots and product data are confidential. The methodology, the convergence evidence and the reasoning are mine to walk through in full at interview.

A fusion reactor needs pulse power. You store a large amount of energy slowly and then release it very quickly, which is the job JET's flywheels have been doing since 1981. Our brief was a system holding about 4.5 GJ per unit at a maximum rim speed of 230 mph, that would stand up structurally and could be maintained.

Sizing the machine

We compared four architectures on a weighted matrix: four units at 300 rpm, six at 200, three at 400 and eight at 150. The four unit option won on the balance of energy density, redundancy and how manufacturable each component was. I did the sizing from end to end: energy and power needed per unit, then angular velocity, rim moment of inertia, preliminary hoop stress, allowable stress and factor of safety, and finally the bolted joint capacity for the segmented rim. The rim is thick walled and split into segments so it can be made and repaired in pieces, with bolted interfaces I sized against their own factor of safety. It did not arrive in one pass. The geometry went through several revisions as the hoop stress, rim inertia and joint capacity resolved, and each change to the numbers sent me back into the model to make the design satisfy the structural requirement and stay manufacturable at the same time.

Structural and modal FEA

I ran static structural analysis on the rim segment, the spider arms and the shaft under centrifugal loading at operating speed. The symmetric boundary conditions were checked against how the parts are actually constrained rather than assumed, fillets were added at the stress concentrations, and I ran mesh convergence at the arm attachment where the peak sits. Peak von Mises stress there was compared against material yield and I reported a factor of safety for each component. I then ran a modal study to confirm that 300 rpm sits clear of the assembly's natural frequencies, which is the check that matters most on anything that spins.

The drawing set

Every component was drawn to BS 8888 with tolerancing, surface finish callouts, section and detail views, and completed title blocks. The set covers the rim, rim plate, hub, spider arm, shaft and magnetic bearing holder, and finishes with a general arrangement and an exploded assembly drawing with balloon references against a parts list. The point of a drawing set is that someone who has never seen the model can build from it, and that is the standard I worked to.

Drawings shown blurred. I'm waiting on Liverpool John Moores for permission to publish the full set. The originals are complete and I'm happy to take you through them at interview.

The assembly drawings are the ones I would point to first. Producing an exploded view with balloon references tied to a parts list forces you to resolve how the machine actually goes together, in what order, and with what fasteners, which is a different exercise from modelling the parts.