Ethan
Combs

An industrial-grade materials test machine, designed and built on a student budget.

Two linear actuators drive a crossbar through an aluminum extrusion frame, loading the specimen in tension or compression. An S-beam load cell, an optical encoder, and an IMU feed an Arduino Nano on a custom PCB, and firmware I wrote turns the raw readings into engineering stress–strain curves.

Build details

• Designed and built the machine to a 3,000 N rating in both tension and compression, integrating dual linear actuators, an optical encoder, an IMU, an Arduino Nano, and a custom PCB.
• Developed the data acquisition and processing firmware that calculates engineering stress–strain curves at 97% accuracy.
• Checked the load cell mount with a SolidWorks static study before building it into the frame.
• Designed the machine safety systems, including a shrapnel shield and emergency stops, to contain specimen failure.

A battery-powered gauge cluster that lives inside a dashboard vent and reads the car's data live.

An ESP32 talks over Bluetooth to an ELM327 module, polling CAN bus and OBD-II data at high frequency and calculating acceleration on the fly. I designed the menu system, wrote the display firmware, and designed the enclosure in SolidWorks to fit the vent.

It runs on an 18650 battery, so power draw mattered: when the car reports zero speed and the data stream stops, the cluster puts itself to sleep.

A four-wheel robot that drives itself around a scale city course, following lanes and reacting to what's in its path.

A forward camera feeds a Radxa ROCK 5C Lite, which handles lane following and runs an object detection model on its NPU. I trained that model myself on an RTX 2060 GPU before deploying it to the board. When the detector picks up an obstacle, the car stops or steers around it. The detections stream live to a browser dashboard, so I could watch what the car was seeing while it drove.

A Hot Wheels car retrofitted to drive over Wi-Fi from a browser, a phone, or a sim racing setup.

I fit a dual-core ESP32 control board and a motor driver into the car and modeled the chassis in CAD to carry them. The ESP32 runs its own Wi-Fi stack, so any device on the local network can send it full directional control, from a browser tab to a sim racing rig.

A servo gimbal that holds an FPV drone camera steady, with gains placed by root locus instead of trial and error.

The gimbal rolls, pitches, and yaws to keep a 30 g FPV camera on target while the drone maneuvers. We modeled the gimbal arm and camera as a second-order plant, chose 5% overshoot and a 0.5 s settling time, placed the closed-loop poles in the s-plane, and solved for PD gains. A Seeed XIAO ESP32-C3 runs the controller.

Team project with Rebecca Moretta.

Design details

• Modeled each axis as J·θ̈ + c·θ̇ = τ, using a 50 g, 80 mm arm and a 30 g camera at 70 mm for J = 0.000254 kg·m², with damping c = 0.002 N·m·s/rad.
• The 5% overshoot and 0.5 s settling targets give ζ = 0.69 and dominant poles at s = −8 ± 8.39j.
• Proportional control alone only reached ζ = 0.339, so we added derivative action and solved for Kp and Kd.
• Wrote a MATLAB script for the step response and root locus, and checked the ESP32-C3's servo signal on an oscilloscope while it tracked a simulated sinusoidal drone sway.

A compact linear regulator board, designed in KiCad, that steps a 5 V supply down to a clean 3.3 V rail.

I drew the schematic, laid out the board, and checked the fit in KiCad's 3D viewer. A SOT-23-5 low-dropout regulator sits between input and output decoupling capacitors, with a bypass capacitor on its noise-reduction pin. An LED on each rail shows at a glance that 5 V is coming in and 3.3 V is going out.

Four fin layouts, one forced-air cooling case, and a clear winner.

We compared zig-zag plate, cylindrical pin, curved plate, and rounded-rectangular fin arrays in copper, cooling from 80 °C in a 76 °F room under 22 mph (9.8 m/s) airflow. I designed the cylindrical pin and rounded-rectangular arrays, attempted to sand-cast the fins, and built an Arduino temperature-logging rig for physical tests.

When the casting stock turned out to be copper-plated steel rather than solid copper, the comparison moved to a Python model built on adiabatic-tip fin efficiency, the Churchill–Bernstein crossflow correlation, and lumped-capacitance transient cooling.

EML 4140 Heat Transfer, with Mitchell Parrish. Written up as an ASME-format report.

Predicted performance, copper, 22 mph forced air

Fin layout	Initial heat dissipated	Time to cool	Fin efficiency
Zig-zag plate	52.8 W	114.1 s	93.8%
Cylindrical pin	59.0 W	36.9 s	81.4%
Curved plate	41.5 W	92.5 s	92.3%
Rounded rectangularBest	238.9 W	12.2 s	89.6%

Time to cool is from 80 °C to within 1 °C of room temperature; shorter is better. The rounded-rectangular array is a 40 × 40 × 10 mm sink with 130 fins; the other three use 25 mm fins.