11 min to read

DIY Torque

#Education #Plan

Most digital torque adapters are built around 1/2 inch square drive. This project is for smaller 1/4 inch hex bit work, where the usual automotive adapters are too bulky.

The goal is a compact torque checker/adapter for repeatable workshop use. It is not a certified torque instrument.

Design targets:

Selected hardware:

System overview

The measuring chain is:

torque element -> strain gauges -> Wheatstone bridge -> NAU7802 -> T-QT Pro display

The steel bit holder or extension twists slightly under load. Strain gauges bonded to the steel change resistance as the shaft twists. Four gauges form a Wheatstone bridge. The NAU7802 powers that bridge, amplifies the tiny differential signal, and sends ADC readings to the T-QT Pro over I2C.

How torque becomes a number

Torque is force applied at a distance from a pivot:

\[T = FL\]

For a hanging mass on a calibration arm:

\[T = mgL,\qquad g \approx 9.81 \ \text{m/s}^2\]

Example:

\[T = (5 \ \text{kg})(9.81 \ \text{m/s}^2)(0.50 \ \text{m}) \approx 24.5 \ \text{Nm}\]

A round shaft under torsion has its strongest surface strain roughly 45 degrees from the shaft axis. That is why the gauges are placed at alternating +45 and -45 degree angles. The exact placement is covered in the mechanical build section.

The bridge output is close to zero with no torque applied. After calibration, firmware subtracts the zero reading and divides by a scale factor to estimate torque.

Parts and wiring

T-QT Pro

The T-QT Pro is the first controller choice because it already has the MCU, display, USB-C power/programming, and a Qwiic/STEMMA-style I2C connector. The selected variant is the 4 MB flash + 2 MB PSRAM version.

Community notes for this board point to GPIO43 as SDA and GPIO44 as SCL. The I2C scanner is the sanity check before committing to that pinout.

NAU7802

The NAU7802 is the strain gauge front end. It has a 24-bit ADC, a programmable gain amplifier, bridge excitation pins, and a fixed I2C address of 0x2A.

Typical breakout wiring:

T-QT Pro side NAU7802 side Purpose
3V3 VIN Power
GND GND Common ground
GPIO43 or SDA SDA I2C data
GPIO44 or SCL SCL I2C clock
not required first DRDY Optional data-ready signal

Bridge wiring:

NAU7802 terminal Bridge node
E+ Excitation positive
E- Excitation negative
A+ Signal positive
A- Signal negative

Controls and pin plan

The enclosure uses four external buttons. The onboard buttons are avoided because their placement does not work well for this layout.

Button Short press Double tap 3 s hold
Power / Clear Clear - On / Off
Mode Track / Peak - Backlight
Up +1 Nm +5 Nm -
Down -1 Nm -5 Nm -

Pin plan:

Function T-QT Pro pin Connection Notes
NAU7802 SDA IO43 Qwiic/STEMMA SDA Reserve for amplifier I2C
NAU7802 SCL IO44 Qwiic/STEMMA SCL Reserve for amplifier I2C
Button: Power / Clear IO16 Button to GND Internal pullup
Button: Mode IO17 Button to GND Internal pullup
Button: Up IO18 Button to GND Internal pullup
Button: Down IO48 Button to GND Internal pullup
Buzzer module signal IO35 Module I/O pin Active buzzer module
Buzzer VCC 3V3 Module VCC 3.3 V active buzzer
Buzzer GND GND Module GND Common ground
Battery JST battery pads/connector 1S LiPo T-QT Pro supports battery charge/discharge
Avoided IO0, IO47 Onboard buttons Bad placement for this enclosure
Reserved IO1, IO2, IO3, IO5, IO6, IO10 LCD pins Display

Battery and runtime

The enclosure may force the battery down to about 250 mAh. That is still plausible for short workshop sessions if the firmware has auto shutoff and the display is not left bright.

Assume usable capacity is roughly:

250 mAh x 0.8 = 200 mAh usable

Runtime estimate:

Active current Runtime 30 min sessions
40 mA optimized ~5.0 h ~10 sessions
55 mA good ~3.6 h ~7 sessions
70 mA likely ~2.9 h ~5 to 6 sessions
90 mA high ~2.2 h ~4 sessions
100 mA bad ~2.0 h ~4 sessions

Component drain estimate:

Component Current estimate Notes
ESP32-S3 / T-QT Pro board ~20 to 50 mA Biggest variable
LCD/backlight ~10 to 25 mA Dim aggressively
350 ohm strain bridge ~9.4 mA Continuous while measuring
NAU7802 ~2 mA Small load
Buzzer ~0 idle, ~10 to 30 mA active Average is low unless constantly beeping
Board overhead ~2 to 10 mA Regulator, charger, LEDs, leakage

Expected active draw:

normal: ~60 to 90 mA
optimized: ~40 to 55 mA

For 30 minute sessions with auto shutoff, a 250 mAh cell should be workable. Plan around 4 to 6 real sessions per charge. More is possible if the screen is dimmed and the bridge/display are shut down when idle; less is likely if the display is bright and the ESP32 is left running fast.

Mechanical build

Shaft choice

The first spring element is a high-quality 1/4 inch hex bit holder.

Rationale:

A 1/4 inch hex shank has:

Across flats: 6.35 mm
Individual face width: ~3.67 mm

The working scope is:

Target low end: 3 Nm or lower
Upper working range: 20 Nm

A 1/4 inch steel shank should produce a readable strain signal. The practical limit is whether the shaft stays elastic through the working range. Heat treatment, grooves, detent cuts, transitions, and socket cavities all matter.

Good candidates:

Poor candidates:

A failed zero-return test at 20 Nm points to a stronger torsion section with 1/4 inch hex interfaces.

Gauge selection and prep

The bridge uses four separate two-wire gauges:

\[R_{gauge} = 350 \ \Omega\]
Item Decision
Model direction BF350-0.1AA or equivalent
Approx. substrate size ~1.4 x 2.22 mm
Gauge factor ~2.0
Why Small enough to fit on a 1/4 inch hex flat at 45 degrees

A prebuilt four-wire load cell is the wrong final sensor. It already has its own mechanical element; this project needs gauges bonded directly to the bit holder or extension.

The gauges need bare, clean metal under the active area. Plating only comes off where the gauges sit. The goal is strain transfer, not cosmetics.

Item Decision
First prototype adhesive Thin CA glue
Better adhesive later M-Bond 200
Rejected adhesives Gel CA, thick epoxy, 5-minute epoxy, silicone, rubberized adhesive
Reason The bond layer needs to be thin and hard so shaft strain transfers cleanly into the gauge

Sensor placement

On the BF350, the fine parallel lines inside the active area are the sensing direction.

All four gauge centers sit on one narrow band around the shaft.

Shaft end view:

Placement rule:

Gauge Clock position around shaft Fine-line direction
G1 0 degrees, top +45 degrees
G2 90 degrees, right -45 degrees
G3 180 degrees, bottom +45 degrees
G4 270 degrees, left -45 degrees

A negative reading under tightening torque is only a sign convention problem. A+ and A- can be swapped, or the sign can be inverted in firmware.

Validation and calibration

Software setup

Arduino IDE is enough for the first hardware proof. The needed pieces are:

Software/library Purpose
Arduino IDE Compile and upload sketches
ESP32 board package Adds ESP32-S3 support
Adafruit NAU7802 library Reads the NAU7802
TFT_eSPI library Drives the T-QT Pro display
LILYGO T-QT examples/config Reference display setup

The display can be brought up from the LILYGO examples. The NAU7802 should appear in an I2C scan at 0x2A.

Electrical test

The first electrical test uses a dummy Wheatstone bridge instead of bonded gauges:

\[R = 350 \ \Omega\]

Passing signs:

T-QT Pro powers up
I2C scanner finds 0x2A
NAU7802 raw reading is stable enough to tare
Display updates

Bonded gauge validation

The bonded-gauge check is the first test of the actual sensing element:

  1. Four gauges bonded to one shaft band.
  2. Full bridge wired into E+, E-, A+, and A-.
  3. No-load tare captured.
  4. Raw readings move consistently under torque.
  5. Opposite sign under reverse torque.

Calibration

Calibration can come from a torque wrench or a lever-and-weight setup. A single linear scale factor is enough for the first prototype.

Calibration points:

\[T = [0,\ 1,\ 3,\ 5,\ 10,\ 15,\ 20] \ \text{Nm}\]

The useful data is the raw reading while loading and unloading. Poor zero return points to mechanical hysteresis, a weak gauge bond, wire strain, or overload.

The shaft decision comes from zero return:

tare
apply torque
release
check zero

Torque progression:

3 Nm
5 Nm
10 Nm
15 Nm
20 Nm

A shaft that does not return to zero after release is not suitable at that torque.

Scale factor:

\[\text{countsPerNm} = \frac{\text{raw}_{known} - \text{raw}_{zero}}{T_{known}}\]

Firmware conversion:

\[T = \frac{\text{raw} - \text{zeroOffset}}{\text{countsPerNm}}\]

Remaining validation

The remaining work is mostly validation, not feature design:

Open decisions

Decision Current leaning
Exact bit holder High-quality or impact-rated holder with the cleanest hex section
Final adhesive Thin CA first, M-Bond 200 later if the prototype proves the geometry
Final package Decide after sensor geometry is proven

Sources