Aerospace PhD candidate at Stanford. Uncertainty quantification for machine learning, applied to aircraft design.
The jam happened at the moment of engagement, so I removed the engagement.
The competition car's starter threw a pinion into a spinning flywheel to crank the engine, and about 10% of starts jammed. A jam during a timed run ends the run.
Replaced the throw-in pinion with a dog gear arrangement, in which a PETG driving gear stays permanently meshed and nothing has to engage under load. Debugged the redesign through 3 successive failure modes before it held.
A starter that neither jammed nor broke across 1.5 seasons of testing, against roughly 10% of starts before the redesign.
Designing a failure out beats defending against it: the event the jam depended on no longer exists, so that mode cannot return. The 3 failure modes that followed were new ones, and testing was the only thing that found them.
Asked everyone what annoyed them about the old one, then reproduced every complaint myself before believing it.
A load-bearing rear wheel adapter that was heavy, awkward to adjust, and could not be serviced without dismantling more of the car than the job warranted.
New to the team, started by asking other members what was wrong with the assembly and then reproducing their complaints rather than taking them on trust. Set 4 explicit objectives before drawing anything.
A 52% lighter assembly, serviceable with a hex key and manufacturable in the team's own shop, with a safety factor of 2 on the vertical load case in FEA.
Checking cornering and braking was worth the time even though neither governed - the car travels slowly enough that the vertical case dominates, and knowing which case governs is what let the mass come out.
Most of the drag was frontal area, and the only thing stopping me cutting it was the driver's eyeline.
An outgoing body shell on a vehicle whose entire purpose is fuel efficiency, where drag is most of what there is to win.
Ran the whole aerodynamic analysis and shape optimisation for a new generation of car, cutting drag principally by reducing frontal area while holding the driver visibility requirement.
A predicted drag reduction of nearly 35% against the outgoing car, in CFD.
The 35% is a CFD prediction. The body was manufactured but the car never raced it, so the number I would most like to have is track data to put against the model.
The shell was sagging at competition. Stiffening it was not going to happen that weekend, so I carried the load somewhere else.
The carbon monocoque sagged in competition, to the point the car visibly sat wrong.
Rather than attempt to stiffen the shell, carried the load with a separate structure - square steel tube with a 0.125 inch wall - designed and fabricated at the 2022 competition rather than beforehand, and fitted in under 4 hours.
The sag was eliminated and the car sat correctly again, for a weight penalty of about 10 kg, roughly 7% of car and driver together. 3 subsequent generations of the car adopted the approach.
A 7% mass penalty on an efficiency vehicle is a real cost, and it was the right trade only because the alternative that weekend was not running at all. With time, the load belongs in the shell rather than beside it.
Type an address into a browser; the aircraft flies there.
Getting a small payload to a street address quickly, dispatched by someone who is not a drone operator.
Built 4 subsystems: MAVLink telemetry including arm and disarm, address-to-coordinate resolution and autonomous navigation, a cellular link, and live video to the ground station.
A drone deployable to an address from any computer over a cellular link, with live video and telemetry, flown autonomously 4 times.
It flew autonomously 4 times and no first-responder organisation ever trialled it. The lesson I took is that the hard part of this project was never the software.
Reasoned from the lift equation: if you cannot change density or angle of attack, change the wing.
A vertical take-off aircraft wants a small wing to hover and a large wing to cruise efficiently.
Recruited and led a team of 4, taking the design, analysis and prototyping myself. Worked from the lift equation to the configuration, then built and flew 2 prototypes.
A cruise lift-to-drag ratio of 2.4 in CFD, and an estimated 2.5 hours of endurance unloaded.
Vertical flight worked and forward flight worked; the handover between them did not - the aircraft could not build enough airspeed during the rotation. That is where the design would have to change, not in the wing.