testing grounds, not the final site yet

Active fin stabilization canard system

Rice Eclipse Rocketry · Avionics-Mechanical subteam · Aug 2025 to present

Four canards near the nose of the rocket, each driven by its own servo, that deflect in flight to correct pitch and roll. The flight computer decides the angles. The mechanical side is mine. I designed the canard actuation mechanism and its housing in Onshape, from the torque path to the module packaging, and I own the CAD on this page.

How it works

1. The flight computer reads the IMU and commands a deflection angle for each canard.
2. Each of the four canards has its own servo. The servo horn drives a connecting rod, and the rod drives a fin horn on the canard shaft. The servos are AGF-RC SA18 low-profile programmable digital servos, 250 oz-in.
3. Each canard shaft rides in a standard ball bearing in the top plate, so the servo only fights aerodynamic load, not friction. Threaded joints use heat-set inserts in the printed parts, and the fasteners are secured with Loctite.
4. The module stacks top to bottom: canard plate with bearings, servo cage, battery, then the PCB compartment for the flight computer. Composite or metal bulkheads close it off on either side, and the whole module fits inside a 3 in airframe.
5. The control law that produces those angles is mine. Attitude error to desired rate, PD to desired angular acceleration, Euler's equations with the inertia tensor to the required moment, then M = qSEδ solved for the four canard deflections.

Design loads

The vehicle design point, from the OpenRocket model. 45 in long, 3.16 in diameter, 63.5 oz on the pad, flown on an I280DM-14 to 3,764 ft, 838 ft/s max velocity (Mach 0.75), 575 ft/s² max acceleration, stability 1.45 cal at Mach 0.3. Three of those numbers drove the mechanism.

• Acceleration. 575 ft/s² is about 18 g. During the burn every part in the stack, the servo cage, the battery carriage, and the PCB floor, carries eighteen times its own weight, and so do the screws holding it together. That is the load the heat-set inserts and the aluminum load path are sized against.
• Aerodynamic load. At Mach 0.75 the dynamic pressure is about 40 kPa, roughly 6 psi. A deflected canard turns that pressure into a hinge moment on its shaft, 0.24 N·m (about 34 oz-in) at full deflection by the MATLAB model, and that moment set the torque range the servos had to hold. It is how we landed on 250 oz-in servos, a 7x margin that still fit the budget, and it is why the canard shafts ride in ball bearings, so the servo spends its torque on aerodynamic load rather than friction. Backlash matters here too, because the correction itself is only two or three degrees.
• Stability. 1.45 cal is deliberately low. A very stable rocket fights the canards and needs bigger deflections to move. A low margin gives the canards authority but leaves less room for error, so the mechanism has to position accurately and hold that position under load.

Validation

What I could check before the hardware flew, what the numbers said, and what is still unproven.

Validated

• Canard loads. A MATLAB model of the canard as a flat trapezoidal fin (3 in root, 2 in tip, 1 in span) using finite-wing lift-line theory. At 250 m/s and 15° deflection, one canard sees 16.9 N of lift, 5.2 N of drag, and a 0.24 N·m hinge moment about its shaft. I reran the script and the numbers reproduce exactly.
• Servo torque. 0.24 N·m is about 34 oz-in. The 250 oz-in servos we bought carry a 7x margin, before the factor of safety of 2 we applied to everything else. Result: the servos are not the weak link, and a smaller servo would have fit if the budget had pushed us there.
• Structural margin. With the same loads, hand calcs on the connecting rod, fin horn, and canard shaft land between 100 kPa and 300 kPa against yields of 30 to 60 MPa for the printed PLA parts. The one number that actually matters is bending in the 1/4 in 6061 shaft, where the canard force acts about an inch off the bearing. That comes to 18.6 MPa against a 276 MPa yield, a 15x margin. Result: nothing in the load path was ever close to breaking, so materials were chosen for manufacturability and wear, not strength.
• Linkage kinematics. A four-bar closure solve for each of the four servo positions, mapping servo arm angle to canard angle across the servo's travel and plotting the transmission ratio, so the flight computer knows how many servo degrees make one canard degree at each station.

Not yet validated

• Closed-loop performance. The RocketPy flight simulation exists, with a fixed-time-step scheduler feeding the PD controller and a drag-versus-Mach study for the fin set, but the controller gains were never tuned in it and no run was recorded. So there is no simulated attitude-error number I can stand behind yet. The reason is simple. The team's time went to building and car-testing the hardware before the sim was finished.
• [placeholder: ANSYS modal analysis of the housing, ANSYS Fluent canard loading at the design point, and the RocketPy disturbance run with the RMS attitude-error reduction. Rerun and add the numbers.]
• Flight. The system has been car-top tested, not flown. Every load and margin above is a prediction until an instrumented flight measures it.

Status

The canard system works on the bench. The flight computer reads the IMU and drives all four canards, which is the video below. It has not been on the car-top rig yet, and it has not flown.

What went wrong, and what comes next

What went wrong

• Every performance number so far is a simulation number. The controller has been exercised against injected disturbances in a model, not in flight. Until the system flies on an instrumented vehicle, the result is a prediction. 
• Backlash in the torque path, and it is a tolerance problem before it is a design problem. Going back over the earlier iteration with the subteam lead, the two things that actually hurt it were materials and tolerances, and specifically the backlash stacking up through the servo and the pushrods. That matters more than it looks here, because the corrections the controller asks for are only two or three degrees. A degree of slop between the servo horn and the canard is a third of the command gone before the fin moves.
• The mechanism was never car-tested. The rig was built and driven, but the canard module never went on it, so the torque path has only ever been exercised by hand on a bench.

How I would improve it

• Bench-test the loaded mechanism, not the free-swinging one. Servo torque and deflection accuracy under a representative aerodynamic load is the number that matters.
• Measure the real first mode with an impact test and compare it to the ANSYS prediction before trusting the margin.
• Take the slop out of the torque path. Tighter tolerances on the horn and rod-end joints, stiffer material where the linkage is printed, and a bench measurement of the backlash itself (command a known angle, measure what the fin actually does) so it is a number instead of a feel.

Next steps

• Car-top test the mechanism. The rig already exists and has been driven, so the next run puts the canard module on it and logs commanded angle against measured angle at speed.
• Fly it. An instrumented test flight with IMU logging to compare measured attitude error against the simulated reduction.
• Hardware-in-the-loop testing so the flight computer drives the real servos against a simulated flight before launch day.
• Push the design point closer to transonic. The I280 gets the vehicle to about Mach 0.75; the next question is what the canard loads, hinge moments, and flutter margin look like at Mach 0.9 and above. [confirm]

Tools

Onshape · OpenRocket · RocketPy · ANSYS Mechanical · ANSYS Fluent

Back to projects