An open, low-cost torsion balance for measuring the spurious forces that dominate small-force and thrust experiments, with printable mechanics, a validated force-noise model and fab-assembled electronics.
The question. On an open, fully SMT-assembled, low-cost torsion balance operating in air, quantify the forces that usually pass for thrust: the Lorentz force of supply-lead currents in the ambient magnetic field, and the thermal and convective forces of a dissipating device. Measure them as functions of current, lead geometry and dissipated power, with the balance's force noise floor, calibration constant and drift fully characterized. The instrument claim is a resolution of 0.4 nN or better (5σ on a 20 nN term) in a 1000 s measurement, at a parts cost to be filled in from the final bill of materials.
| Piece | State |
|---|---|
| G1 measurement goal | confirmed 2026-10-23 |
| G2 mechanics | an FR4 arm board on edge hung from a torsion head, collision-free over the swing, centre of mass solved onto the axis; JLC mechanical boards (the arm, electrode and clamp boards are ordered; the electrode and ground-clamp boards have arrived); printed PLA enclosure (1.2 mm plates with a 2 mm rim, no window); field-modelled eddy damper; cutting jigs; fabrication decisions, assembly steps |
| G3 force-noise prediction | 0.108 nN (1σ, 1000 s, in air): a conditional go |
| WP1–WP3: requirements, 50 JLC-stocked parts, analog block simulations | done |
| WP4 schematic | done as checked netlists (main board + four head boards, 50 JLC parts); readable sheet rendering is in the backlog |
| WP5 main-board layout | done to the DFM gate (120 x 90 mm, 4 layers, routed); fab files in hardware/instrument/fab/ (Gerbers, BOM, CPL, ORDER.md) |
| Main-board order | route decided (Economic PCBA, 2 boards assembled, LIS2DW12 not placed on the first articles); in the JLC cart, held for the Black Friday sale |
| A5: prototype noise measurement | after the board arrives and passes bring-up (docs/bringup.md); it runs on the assembled board, not on breadboards |
The prediction is dominated by air currents (4.8 nN/√Hz), so the verdict flips at about 4 mm/s of air-speed fluctuation; the enclosure is the biggest noise control. The model reproduces six published numbers, from Soni and Roy (Rev. Sci. Instrum. 84, 095103, 2013) and the cryogenic torsion balance of arXiv:2206.02890.
| Path | What |
|---|---|
docs/plan.md |
the plan: goal, mechanics, noise budget, environment control, calibration |
docs/fabrication.md, docs/assembly.md |
how each part is made (fab boards, PLA, bought; JLC orders, coupons and parcels) and how to build it |
docs/bringup.md, docs/procedures.md |
powering up the electronics step by step; the prototype check, calibration and measurement campaign |
docs/paper/ |
the paper skeleton (LaTeX; results are placeholders) |
hardware/balance/ |
parametric CAD (balance.py, STL and STEP in cad/), the FR4 boards for JLC (boards.py, boards/fab/), jigs, analytic model, parameters, checks |
hardware/instrument/ |
the electronics: requirements, pin map, parts table, ngspice simulations, checks |
hardware/tools/jlc_query.py |
search of the JLCPCB parts snapshot |
tools/force_noise_budget.py |
the force-noise prediction and the anchor checks |
docs/TODO-agent.md, docs/TODO-arcticoder.md |
the work lists (TODO-BLOCKED.md, TODO-backlog.md, TODO-completed.md beside them) |
Setup (Linux/WSL) is in hardware/README.md: KiCad 9, a Python venv with CadQuery, SKiDL and ngspice.
python3 tools/force_noise_budget.py # the noise prediction and anchor checks
~/.venvs/lab-hw/bin/python tools/run_all_smoke_tests.py # CAD, pin map, power budget, simulations
- Every electronic part is a machine-placed SMT part from JLCPCB's own library; nothing is hand-soldered and nothing is mailed to the fab.
- USB power only; nothing above 50 V DC or 30 V AC. The device current is capped at 300 mA by the USB budget.
- The balance carries no permanent magnet on the moving arm (a 6 × 3 mm N42 disc alone would add about 7 nN/√Hz of field noise); masses on the arm are brass or plastic.
The bench circuits this experiment uses for its prototype check (lock-in amplifier, optical shadow readout, R-2R signal generator) live in arcticoder/lab, which also holds the shared parts inventory and order ledger.
- Code (MIT, LICENSE): the Python sources, including the CadQuery scripts and the noise model.
- Hardware designs and their documentation (CERN-OHL-P v2, LICENSE-HARDWARE): the CAD outputs (STL, STEP), KiCad
files when they exist, SPICE netlists, parts tables and BOMs, and the assembly and design documents. A CadQuery
.pyfile is code (MIT) whose output is hardware (CERN-OHL-P).