Michael Folmer

A powered knee brace for ACL rehabilitation, assisting recovery through controlled actuated joint motion. I designed the mechanism, the custom PCB, the firmware and the control interface. A brushless actuator drives the joint over a CAN bus link while a 9-DOF IMU and decoded motor encoder telemetry track joint angle and angular velocity.

I chose a CubeMars AK series quasi-direct-drive motor specifically for its low impedance: rehabilitation loading needs the joint to stay backdrivable so the device can assist a patient without fighting them. High reduction geartrains give torque but resist the wearer, which is the wrong trade for therapy.

Safety drove the whole design. The firmware runs a strict passive measure, confirm, then assist sequence. On power up it transmits nothing at all and passively decodes encoder broadcasts to learn the wearer's range of motion. It only starts driving torque once the operator explicitly confirms.

Assist goes out as torque only CAN frames with the gain bytes zeroed. I chose that framing because two MIT byte layouts exist in the field, and a frame with bytes 0 through 5 zeroed decodes as zero stiffness and zero damping under both of them. An ambiguous motor firmware cannot turn a damping request into a velocity slam. Torque is hard capped, overspeed trips need consecutive confirming samples so one bad reading cannot trip them, and the watchdog is sized to the motor's real broadcast period.

The wearer operates it from a phone over the brace's own WiFi access point. The microcontroller serves a self contained web panel that streams live telemetry over WebSocket, with an on board OLED and a BLE link as fallbacks.

Capstone project on an actuator whose output torque and speed can be varied through a two-stage transmission, instead of trading motor efficiency against one fixed reduction. It grew directly out of the brace work: a rehabilitation joint needs to move quickly and freely during gait, then hold and push hard during therapy, and one fixed ratio cannot serve both well.

The work covers transmission architecture, the ratio switching mechanism, and design for manufacture of the printed and machined parts. Modeled in Creo as a full parametric assembly so stage ratios and envelope can be re-solved as the design converges.

The switching mechanism itself is redacted pending IP and is being held back from public description.

A compact two-stage planetary reduction delivering 20:1 from a standard NEMA 17 stepper, designed, analyzed and bench validated end to end.

Both stages are sized to a common 54 mm ring pitch diameter so the whole gearbox packages into a single bore. I then used different modules per stage, 0.75 on the input and 1.2 on the output, putting the larger teeth on the higher torque stage while holding the ring geometry constant.

I ran Lewis bending stress hand calculations on the critical stage-two sun gear and validated them against static FEA in Creo. The two disagreed by 69%, which I documented rather than papered over, since knowing an analysis is unresolved matters more than a clean number. Iterating face width against the calculated stress raised the analytical safety factor from 0.86 to 1.72, moving the design from under unity to a real margin.

I built and debugged the motor test bench myself (A4988 driver, ESP32, 1.5 A per phase), tracing an intermittent fault to a floating high impedance driver node.

A drone platform built from nothing: airframe, wiring harness, and flight firmware, taken from loose components to synchronized motor output.

I modeled the full mechanical assembly and manufactured the airframe by 3D printing, validating fit and tolerances across several iterations while correcting structural weaknesses and optimizing motor alignment and component layout. I built and soldered the custom power and signal wiring myself, integrating motors, ESCs, battery and control electronics, then wrote and debugged the embedded control code.

It never flew properly. The airframe and electronics integrated and the motors ran synchronized, but structural stiffness and motor alignment stayed the limiting factors. I am keeping it here because the wiring, the bring-up and the failure analysis are the parts I actually learned from, and because knowing precisely why something did not work is worth more than a project that quietly succeeded.