Windpack

01What the competition actually scores

what I was thinking Reliability ranks above marginal Cp gains.

The Collegiate Wind Competition puts a turbine that fits in a 45 cm cube into a wind tunnel and scores five things. Each one pushes on a different part of the machine.

One rule frames everything. Any visible failure, a crack, a part coming off, excessive movement, sparking, stops the test with no further points. A blade that leaves its root scores nothing, however good its airfoil. So the design starts from the worst case, not the best one.

02Season one, as it actually happened

What I was thinking: 6th of 10 in the least competitive region means there is real room to grow. Treat the placement as a floor, not a ceiling.

Windpack started in August 2025 with ten people and no turbine. We split into a mechanical team and an electrical team because that was the only way ten people could work in parallel, taught ourselves QBlade, and read the reports of the 2025 winners. The design report went in on April 23, 2026. Competition was May 13 to 16 at Penn State.

What we built, from that report:

• An A18 airfoil picked in QBlade for its lift to drag ratio, exported to SolidWorks. Blades printed in PLA first, which was too brittle, then in ABS.
• Each blade slid over a 6 mm shaft and was held by one M3 set screw. That was the whole root.
• Pitch by a helicopter style swash plate first, which was not stiff enough for three blades, then bevel gears inside a 40 mm hub driven by a 499:1 gearmotor. Its stall torque was about 58 times the 0.052 N·m the three blades push back with at 14 m/s.
• A carbon steel tower TIG welded to a steel base disc, with the nacelle on an aluminum rail so the generator and the pitch housing could slide for balance. First bending mode estimated at 61.2 Hz by a lumped mass method, above the 58 Hz of 3,500 RPM.
• A D6374 brushless motor run backwards as the generator, 7 pole pairs, about 1 V per 150 RPM. A three phase short circuit brake, after a single phase brake shook the drivetrain, wired so it needs a constant HIGH signal to stay off: lose the controller and the rotor stops by itself. A donated solar MPPT into a 5 Ω, 250 W resistor. An Arduino Uno reading an INA226 over I2C behind optocouplers.

Then the tunnel. We had one practice slot. The rotor vibrated badly on spin up, so we hit the kill switch rather than lose the printed hub. The cause was rotor eccentricity: printed parts that did not line up, from small oversights in the CAD. At the event the turbine did not pass technical inspection, so no scored tunnel task was run. Sixth of ten in the East region came from the design report, the project development report, and the two presentations. The composite blades on our slides, glass fiber in epoxy laid up in a printed mold, were a bench trade study on our dynamometer. They never flew.

What broke

03The retro said process, not parts

What I was thinking: Almost every item is a process problem rather than a technical one. Nobody wrote "the blades were wrong."

We photographed the four whiteboards from the end of season retro and typed them up. Deadlines had lived in nobody's calendar. The report was finished late. Decisions were made once instead of converged on. No part had a history. Sub teams did not talk to each other. The turbine was downstream of all of that.

Three answers. A gated sequence: requirements, a high level design with faculty sign off, a low level design with manufacturing sign off, then a first prototype in the fall and a second in the spring. One source of truth, a Notion workspace, so the club's memory stops living in my inbox; if something is decided in Slack and not written down, it did not happen. And a decision log that records what we considered and discarded, and why, so the next class does not relearn it.

Then the advisor check in that changed the plan. The lab's fan wall is for research, so no unofficial testing, and the large tunnel is good for one or two supervised runs a year. If we cannot iterate on someone else's equipment, the rigs we own are the iteration loop. The rotor test stand and the generator dynamometer moved onto the critical path.

04Start from the rules, then from the physics

What I was thinking: A design pipeline that only runs forward once is not a design pipeline.

Every requirement on the year two turbine traces to a rule section or to our own analysis, with a verification method and a status. The ones that drive the mechanical design:

The physics starts from four lines. Power in the wind the rotor captures, the rotor speed a tip speed ratio implies, the torque that leaves, and the thrust that pushes back:

P = ½ ρ A C_p v³,  ω = λ v / R,  Q = P / ω,  T = ½ ρ A C_T v²

With a 225 mm tip radius, a C_p of 0.35 and a design tip speed ratio of 4, all placeholders until the final QBlade blade is set, the rotor makes about 34 W at 10 m/s, at about 1,700 RPM and 192 N·mm of torque. That is the operating point.

The case nobody had written down is runaway. Lose the load, the E stop or the pitch, and the rotor accelerates until the blades make no torque, at roughly twice the design tip speed ratio. At 13 m/s that is 520 rad/s, about 4,970 RPM, which matches what Cal Poly and Cal Maritime measured on their turbines. Try it:

05The root: size it for the pull, not the bend

What I was thinking: Actually! the hub's lip will be load bearing... hold on let me get back to you on that.

At runaway each 22 g blade pulls outward with F = m ω² r_cg = 593 N, about sixty kilograms, and that pull grows with the square of speed. The flapwise bending moment at the root, from a conservative 30 N of rotor thrust shared by three blades, is 1.22 N·m. The pull is the design load by a wide margin, and it was the load our year one root had never been checked against.

Before sizing anything I read the four public design reports that describe their root joint in enough detail to compare. They all run 45 cm rotors, so their loads scale straight to ours.

Six things fell out. Design for runaway, not for power. Centrifugal dominates, so every gram near the tip costs root load. The plastic interface fails, the metal does not, so load has to enter the plastic over an area, not through a point fastener. Separate the load paths: thrust to a thrust bearing, bending to a radial pair, pitch torque to the gear and nothing else. Backlash is a control problem, not a tolerance footnote. And spin test to overspeed before the tunnel.

So the blade is printed in a tough resin with a 4 mm pultruded carbon rod bonded inside, root to about 35 to 45% span, in a root thickened to 18 to 20% so the rod fits with a 1 mm wall. The diameter came from σ = 32M / πd³ at 1.22 N·m:

Three millimeters meets the safety factor with no margin. Five does not fit a root that is only about 7 mm thick. The rod is not there for static strength: the resin alone would see 8.7 MPa at the root, which passes a 40 MPa static check and fails a 12 MPa fatigue one, and the tower shadow loads the root once per blade pass for the life of the machine. The rod is for fatigue, stiffness, and a clean load path into metal.

Holding the blade on costs a bond length, L = SF F / (τ π d). Epoxy into the resin bore at 5 MPa needs 19 mm, so it gets 25. Epoxy onto the abraded rod inside a steel sleeve at 15 MPa needs 6 mm, so it gets 12. A set screw on bare carbon crushes the rod at about 150 N and was rejected. From there the pull goes blade, bond, rod, bond, sleeve, sleeve flange, thrust bearing, housing lip. The lip is the piece I had forgotten was load bearing.

06The hub: argue about it on a table

What I was thinking: I'm making a scaled up solidworks model and tryna get it printed tonight, so we can look at it in-person to analyze.

Four hub decisions, each with the number that made it.

Ball bearings, not bushings. Friction torque is μ F r. Under 593 N a thrust ball bearing adds 11 N·mm to the pitch torque; a bronze bushing adds 534. Bushings would make the pitch drive 5.7 times stronger, 2.37 N·m against 0.42 at the three blade gears, and they stick and slip, which shows up as pitch hunting in the rated power task. On the drive shaft a bushing would eat 11% of the rotor's torque at 5 m/s, which is where cut in is decided.

Keep the tip, accept the hub. Power per unit span grows with radius, so the fraction lost to a hub of radius r_h is about (r_h/R)². Growing the hub exit from 22 to 32 mm to fit the bearings costs one percent of available power. The tip radius is what matters, and the cube fixes it at 225 mm.

Bevel gears, not miter gears. Three equal 1:1 gears around one central gear sit √3 R_c apart, which is 1.73 of a blade gear radius, less than the 2 they need. They hit. A 30:20 set clears by 2 mm at module 0.5. Lewis bending on the teeth gives 64 MPa at twice the pitch torque, so the gears are steel: tough resin is good for about 12 MPa and printed teeth creep, which is backlash.

Machine the housing. The lip that carries the pull sees 15.8 MPa of shear at twice the load. In 6061 aluminum that is a safety factor of 10; in printed PLA it is 1.9. So the real hub is one piece of aluminum, bored from both ends on the shop's mill.

Then the part I think mattered most. I wrote the hub as a script, printed it at three times scale that night with color coded placeholders for the bearings, thrust washers, spacers and circlips, and left it in the storage room so the pitch sub team could take it apart. Four days later the drive was rebuilt as the helix Penn State uses: a nut that slides on the shaft, spins a lead screw keyed into the central gear, and pitches all three blades. Thirty millimeters of stroke gives 108 degrees of pitch at model scale, and the actuator only ever pushes along the shaft axis, which is what gets around the swash plate buckling that cost us a year ago.

07The actuator question nobody had run

What I was thinking: Starting up was only a problem last competition for us because we had SOOOO much friction in our driveshaft. If we make it spin as smooth as a gust of wind, I don't see cut-in speed being a problem.

The team's first draft carried Penn State's actuator, a 22 N linear actuator, because Penn State's worked. Nobody had run the number it has to beat. A spinning blade wants to twist toward flat pitch so its mass lies in the rotor plane. That centrifugal twisting moment is ½ J_c ω² sin 2θ, so it peaks at 45 degrees of pitch, which is exactly the angle a feathering blade has to pass through.

Through the helix and the gears, that comes to about 10 N at the nut for a shutdown that starts at 1,700 RPM, and 13 to 23 N once helix friction is in. At a 13 m/s runaway it is 70 N ideal and 90 to 160 N with friction. No version of that actuator delivers the second number. So the design has a decision in it that nothing downstream can skip:

• Feather, turning the blades edge on against the twisting moment, and buy the force: a 200 N actuator of the same family, or a 21.5 kg·cm servo on a fork, the way a helicopter tail rotor slider works.
• Pitch to stall, where the twisting moment helps instead of fights, and keep the light actuator, if QBlade shows the stall angle takes the rotor below 10% RPM everywhere from 5 to 13 m/s. Penn State shut down this way.

There is also a race. When the organizers disconnect the load, the rotor starts toward runaway while the nut is still moving, and the force it needs rises with ω². Three ways to win: an actuator that can feather at runaway speed, shutting down toward stall, or a brake resistor on the turbine side that holds the rotor while the blades move.

Open at build freeze, owned by the pitch lead with the controls lead: feather or stall, the gear ratio and pitch range, and the QBlade check that stall actually reaches 10% RPM. Until a printed helix exists, the friction coefficient is a guess, which is why the force ranges are so wide. The first thing the bench test does is push on a printed helix with a scale and get a real number.

08Four bearings on one shaft is three too many

What I was thinking: Instead of it just saying, "decided to pick to use...", it could be, "We calculated that our moment on the driveshaft from the blades to be X, and the RPM to be Y, and with these bearing our driveshaft loss due to friction would be Z%."

The drivetrain is direct drive: hub, shaft, generator. The proposed generator is an outrunner with its own two bearings, and the shaft sits in two more. Two bearings define an axis. If the other two are not on it, and printed housings cannot hold four bearings concentric, the shaft bends to make up the difference. The numbers for a 6 mm aluminum shaft: a 0.1 mm offset across a 60 mm span forces about 100 N into the bearings and 69 MPa of bending that reverses every revolution, against a fatigue strength near 96 MPa for 6061-T6. That is the year one misalignment, written as a stress.

So, in order of preference: a flexible coupling between the shaft and a stub on the generator, a floating generator held only against rotation, or the generator's bearings as the rear support with a single bearing up front. Then one main bearing fixed and one free to slide, so thermal growth and the actuator's reaction have somewhere to go.

The shaft itself is small for the job: 192 N·mm of torque is 4.5 MPa in a 6 mm steel shaft. What sizes it is imbalance. A gram of mass at 99 mm makes about 27 N of rotating force at runaway, fully reversed every turn. At 0.5 g the Goodman fatigue safety factor is 2.7; at 1 g it is 1.7. That is why blades get weighed and matched to a tenth of a gram, and why the first critical speed, 16,600 RPM, about 3.3 times runaway, is worth a line in the document.

The draft also said "full ceramic bearings" with no reason attached. The standard I asked the team to write to is the one in the thought bubble: the load, the speed, and what the choice costs. The review that came back asked the same question in its own words.

09The generator, and two questions nobody had asked

What I was thinking: I feel like Penn State would definitely flag this fly wheel as an energy storage element.

The proposed generator is an off the shelf robotics outrunner: 100 RPM per volt, 20 pole pairs, 840 grams. At rated turbine torque it draws about 2.3 A, roughly 6% of its 40 A continuous rating, so it carries the mass, inertia and iron of a much larger machine, and that inertia has to be stopped within 10 seconds. Open circuit at runaway is 49.7 V line to line, over the 48 V limit at the PCC, so voltage regulation is not optional and keeping the rotor out of runaway protects the meter as well as the blades. Its cogging torque is not published, and cogging is what decides whether the rotor starts at 5 m/s, in the bins that carry the heaviest weights. Those are the three things the generator's own trade study has to answer before it is ordered.

Two assumptions came with us from year one, and the handbook rule is that both get an answer in the decision log either way. First, is a three phase generator actually required by the rules, or did we inherit it. Second, the flywheel. The engineering director proposed a clutched flywheel to carry the rotor between tunnel runs. I looked up the rule instead: energy storage elements may not store bulk energy on the turbine side of the PCC, and a flywheel is bulk energy storage with a shaft on it. Vetoed. Our cut in problem in year one was never energy. It was friction.

10Reviewed, and what came back

What I was thinking: We should try and make this into comments into our CDR, resolve them, and then call that revision b.

On Oct 12 I folded the sub teams' drafts into one Critical Design Review document, WP-CDR-001, with the structure a design review needs: 18 rule requirements and 9 derived ones with verification methods, load cases and a safety factor table, an interface list, a mass budget and a loss budget, 14 tests, 17 risks, and 24 open items with owners and need by dates. Every subsystem carries a maturity label, concept, rough CAD, or proposed, so nobody mistakes a drawing for a decision. Reviewers tag comments MAJOR, MINOR, or Q and cite the requirement, risk or test ID. I sent it out for review on the 13th.

The comments came back on the 17th, and the headline was fair: it reads like a preliminary design review, not a critical one. Shaft alloy, bearing size, coupling, tower, hub body and shutdown strategy were all still open against a Nov 17 build, and the build freeze date was itself TBD. Five items were marked major, and each one has an answer now:

• Shaft material. Precision ground 6 mm steel shafting instead of aluminum. It is cheap, the tolerance is real, set screws and collars do not chew it, and it is stiffer against the critical speed with 840 g hanging on it.
• The ceramic bearings. Full ceramic carries lower load ratings, costs more, and hates misalignment and shock, which is our best documented failure mode. Plain shielded steel deep groove bearings unless a reason appears.
• Overconstraint. Agreed with the ranking: a beam coupling first. Do not ask printed housings to hold four bearings on one axis.
• Feather or stall. Everything downstream hangs on it, so it gets decided first, even provisionally.
• Set screws into prints at 5,000 RPM. No. Heat set inserts at a minimum, clamps where possible, and the same for anything structural threaded into a print.

The printed part advice went straight into the test plan: print the fit coupons in the same orientation as the real parts, because a bore printed vertically is not the same bore printed flat; measure before and after post cure; ream every bearing seat on purpose instead of as a backup; and run the rod pull out test before designing around 25 mm of bond, because the epoxy to resin shear strength is the least trusted number in the whole document. Build freeze was set for Oct 23, with the off the shelf parts order ready to send the same day.

11Test before the tunnel

What I was thinking: Blades were our most common failure last season, so the stand tests them to failure before competition, not at it.

Lab time is one or two supervised runs a year, so everything that can be learned on a bench is learned on a bench first.

Three rigs make that possible. The blade dynamometer, which I pulled a three person team together to build after year one: a motor spins a blade to 2,500 RPM against a load cell and an RPM pickup on an ESP32, with the frame and motor mounts checked in ANSYS for vibration and in SolidWorks FEA so the stand's own shaking does not end up in the torque and thrust numbers. It is where the composite layups were tested against the printed blades, and it is the rig for the overspeed spin. The rotor test stand, a low friction shaft on a sliding carriage pressing a load cell, reads free spin RPM and thrust so two blade designs can be compared in an afternoon. And the generator dynamometer: a drill spins the generator against an adjustable load while it sits in a pivoting cradle with a torque arm resting on a scale, which gets torque out of cheap parts instead of a sensor.

12The other half, and the program

What I was thinking: Deliverables and timelines make people show up. Vague asks get vague results.

Half the competition is not a turbine. The project development contest asks for a wind farm: a site, a wind resource assessment, a layout and energy yield, environmental and permitting review, supply chain and risk, a financial model, and a lease bid defended in front of a different set of judges. Ours in year one was Steuben Valley Wind Farm, and it is the half that carried us. A fourth contest scores industry partnerships, outreach and recruiting with real numbers, and it is where being organized is the whole skill. Year two gives that half its own leads, its own weekly meeting, and its own faculty reviews, staffed with statistics, economics and environmental students next to the mechanical majors.

The program around the engineering: weekly sub team working sessions, not status reports; an officer meeting for cross team blockers; a monthly all hands so the sub teams do not become separate clubs; and a design review at each gate with faculty on both halves. Two prototypes, so there is a semester to iterate between them. The department's parts funding arrived in October. There is still no cash sponsor; the pitch lead did secure a gear analysis software license by asking the vendor.

13Where this stands, Oct 23 2026

What I was thinking: Honest maturity labels. Nothing on this page pretends to be done when it is not.

Still to write

• Measured numbers: the first coupon fits, the pull out force, the cogging torque, the first spin down.
• The overspeed spin test, and what the blades look like afterwards.
• The first tunnel run of prototype one, whichever way it goes.
• Shop and bench photos. Year one was photographed at competition and almost nowhere else, which is its own lesson.