Electric Bike (v3)

System Architecture

By age 15, I wanted the bike to have more torque. Instead of the motor spinning the tire directly, I built a "mid-drive" system that used a chain and sprocket. This meant the motor could help the bike climb hills much better.

The finished v3 model. You can see the large motor and the chain drive system.

Mechanical Design & Fabrication

Drivetrain & Gearing

I removed the pedal chain entirely and mounted a 10-tooth sprocket on the 6384 outrunner motor, driving the stock 33-tooth rear wheel sprocket. That gives a 3.3:1 reduction. At 120KV and 6S (roughly 22V), the motor spins at around 2640 RPM at no load, which drops to a usable wheel speed around 45 km/h once the reduction is applied.

The gears were cheap Chinese chainrings sourced online. They worked, but the chain tension was never stable because of the mount.

10T motor sprocket to 33T rear — 3.3:1 reduction.

Motor Mount

The motor was attached to the frame using the water bottle cage mounts as anchor points, plus hardware store L-brackets. To clamp the L-brackets securely to the frame tubes, I cut a groove in each bracket and threaded a metal screw tie through it, then cinched it around the tube. It held, but it was not rigid under dynamic torque — the brackets still deformed over time.

For the first time I also used rivnuts. I had no proper rivnut tool, so I fabricated a contraption from whatever hardware I had at hand to pull the mandrel and set them. It worked well enough to be a permanent fixture. The rivnuts gave me threaded inserts in the frame without welding, which was the right call given the tools available.

I also modeled up a 3D-printed PLA bracket for the battery tray from the frame fitment reference photo, and mounted the VESC at the front of the frame.

Motor mount, PLA battery tray, and VESC visible. Functional, barely.

Photo used as reference geometry for the 3D printed bracket.

Electrical & Firmware Implementation

Battery Construction

I also finally stopped soldering directly to battery cells. I bought a cheap battery spot welder from AliExpress. This uses high-current pulses to fuse a thin nickel strip to the cell without heating up the entire battery.

Learning how to spot weld. It's much safer and more reliable than soldering.

It took some practice to get it right, but this was a huge step toward building professional-quality battery packs.

Pack interior — foam padding and duct tape, consistent with the v1/v2 build standard.

Dashboard

Before adding the BLE bridge, I built a local dashboard: a 0.91-inch OLED wired to an ESP32, displaying speed, battery voltage, and SOC in real time. I knew voltage alone tells you the state of charge on a lithium cell when the pack is at rest — voltage under load is a different story, but for a rough fuel gauge it was sufficient.

The OLED was the reason the first VESC died. While wiring the ESP32 to the VESC UART connector, I accidentally shorted a data pin to ground. That was enough to kill the BEC. Not a component quality failure — a wiring mistake. It was only after I replaced the VESC and looked more carefully at what happened that I understood the root cause.

BLE Telemetry

For the first time, I added real-time telemetry. I flashed an open-source project — VescBLEBridge — to an ESP32, which bridges the VESC UART output to Bluetooth Low Energy. This let me read voltage, speed, and current from my phone while riding. First time using PlatformIO and VSCode for an ESP32 project (I had used PlatformIO before at 13 for flashing Klipper to an Ender 3, but this was different — compiling and uploading actual firmware logic, not just a config).

Validation & Failure Analysis (Post-Mortem)

This build killed two VESCs:

VESC #1 — same 75100 model, bought cheap. The internal BEC failed. The actual cause was my own fault: while wiring the OLED ESP32 to the VESC UART, I shorted a data pin to ground and killed the BEC. I did not realize this at the time and assumed it was a component failure.

VESC #2 — the internal BEC failed on a second unit. I did not establish whether this had the same cause as the first failure or was a separate hardware fault, so I cannot conclude that the controller design was the cause.

VESC #3 (current) — sourced from a more reputable seller at $75.00 CAD. Still has a known defect: the ADC reads voltage approximately 1V higher than actual. Liveable at$75.00 CAD. In all other functions it works correctly. You get what you pay for.

This build was a hard lesson in power electronics. The first VESC's BEC failed after I shorted a data pin to ground while wiring the OLED. I did not establish the cause of the second failure, so I cannot confidently attribute it to the cold motor or a current spike. In the end, this version reached 45km/h, but it was often broken. It taught me that building a powerful bike is much harder than just making a fast one.

Bill of Materials