I was asked to measure the deflection of a welded frame under load. I mounted a strain gauge on the frame, wired it to a signal conditioner, and read the output. The number bounced around ±20 με. The real strain was probably 80 με. The noise was 25% of the signal. I got a better number by rewiring the bridge and cleaning up the grounding. This is how I set up a strain gauge measurement that’s worth reading.
The Wheatstone bridge
A strain gauge is a foil resistor on a backing. When the part strains, the gauge’s resistance changes. The change is tiny — gauge factor 2.0, so a 1000 με strain gives a 0.2% resistance change. For a 120 Ω gauge, that’s 0.24 Ω change. You can’t measure that with a multimeter. You need a Wheatstone bridge.
The bridge has four arms. If you use one active gauge and three fixed resistors, it’s a quarter-bridge. The output voltage is:
V_out = V_exc × GF × ε / 4
Where V_exc is the excitation voltage (typically 2-10V), GF is the gauge factor (2.0), and ε is the strain in microstrain. For V_exc = 5V, GF = 2.0, ε = 100 με: V_out = 5 × 2.0 × 100e-6 / 4 = 0.00025 V = 0.25 mV. That’s 0.25 millivolts. You need an amplifier with a gain of 1000 to read it as 250 mV.
The three bridge configurations
| Configuration | Gauges | Output | Temperature compensation |
|---|---|---|---|
| Quarter-bridge | 1 active + 3 fixed | 1x | No (temperature drifts) |
| Half-bridge | 2 active (opposite sides) | 2x | Yes (self-compensating) |
| Full-bridge | 4 active (two tension, two compression) | 4x | Yes, best |
For a simple deflection measurement on a beam, I use a half-bridge: one gauge on the top (tension), one on the bottom (compression). The output doubles, and temperature drift cancels. For a load cell (which has four gauges on a flexure), I use a full bridge. The output is 4x and temperature-compensated.
The noise problem
0.25 mV is tiny. Any electrical noise in the environment (VFDs, servo drives, fluorescent lights) couples onto the gauge wires. The noise is often 0.1-0.5 mV — bigger than the signal. I do three things:
1. Twisted pair wiring. The gauge wires are twisted. Noise induces equally on both wires. The amplifier measures the difference, cancels the noise.
2. Shielded cable. The twisted pair is inside a shield. The shield grounds at one end (the signal conditioner end), not both. Grounding both ends creates a ground loop that picks up 50/60 Hz hum.
3. Short leads. I keep the gauge leads under 2 meters. Longer leads add capacitance that rolls off the signal. For long distances (over 5 m), I use a Wheatstone bridge amplifier at the gauge (transmit the voltage, not the mV signal).
The shunt calibration
Before I take a measurement, I calibrate the system. I apply a known resistance (a shunt resistor across one bridge arm) to simulate a known strain. For a 120 Ω gauge, a 100 kΩ shunt gives a known microstrain. I read the output, compare to the expected value, and adjust the amplifier gain. This confirms the wiring is correct and the gain is set right. Without shunt calibration, I don’t know if the number is real.
What I learned
The ±20 με noise was from a ground loop. The shield was grounded at both the gauge end (through the part) and the instrument end. I lifted the gauge-end shield ground. The noise dropped to ±2 με. The signal was clean. I could read the 80 με strain confidently. The measurement wasn’t wrong — the grounding was.
The setup I use: half-bridge for bending, twisted shielded leads, shield grounded at one end only, shunt-calibrated before reading. A 0.25 mV signal in a noisy shop needs careful wiring. If the number bounces more than 10% of the signal, fix the grounding before you trust the reading. Strain gauges don’t lie, but their wiring does.
