TECH NOTE 04 / Signal quality and metrology
Load cell noise and zero drift: diagnose the full measurement chain
Separate changing force, temperature drift and electrical interference before adding a filter.
Separate changing force, temperature drift and electrical interference before adding a filter.
Express noise in input-referred force units with bandwidth and observation conditions.
A local amplifier can shorten the vulnerable bridge-signal run; it does not remove mechanical or thermal errors.
01 / ENGINEERING DECISION
First decide whether the force itself is changing.
Load cell noise filtering is useful only after the measurement task is defined. A cable pulling on a micro-force sensor, a warming fixture or actuator vibration can produce a real force change. Filtering that motion may hide the symptom while also removing the signal the controller needs.
Log raw or minimally processed readings alongside temperature, motion state and excitation where available. Keep the instrument range, sample rate and digital filters in the record. A screen capture without those settings cannot establish the measurement chain’s usable resolution.
| Symptom | Possible cause | Distinguishing check |
|---|---|---|
| Changes with cable motion | Parasitic force or intermittent connection | Restrain the cable and repeat the same motion. |
| Slow change after power-up | Thermal/electrical drift or changing preload | Log temperature, zero and excitation versus time. |
| 50/60 Hz or drive-related bands | Electrical pickup or periodic mechanics | Compare controlled motion/power states and spectrum. |
| Clipped or flat-topped output | Range/common-mode/output saturation | Check raw input and conditioner limits. |
Supporting scope: NI: SNR and excitation in strain measurements ↗
02 / ENGINEERING DECISION
Convert the useful increment into bridge voltage.
Assume a 500 N sensor has a nominal 2 mV/V full-range change at 5 V excitation. The ideal span is 10 mV, so the slope is 20 microvolts/N. An input-referred RMS noise of 2 microvolts would correspond to 0.10 N RMS in the stated bandwidth. Neither the nominal span nor the noise example is an MT-ILC test result.
Use the actual calibrated slope when evaluating a supplied chain. Distinguish RMS noise, peak-to-peak noise, quantization step and accuracy. Their values depend on bandwidth, observation duration and conditions; a 24-bit output word does not guarantee 24 useful force bits.
Force-equivalent noise = input-referred voltage noise / calibrated voltage-per-force slope
| Quantity | Value / calculation |
|---|---|
| Nominal electrical span | 2 mV/V × 5 V = 10 mV |
| Nominal slope | 10,000 microvolts / 500 N = 20 microvolts/N |
| Assumed RMS input noise | 2 microvolts in the stated bandwidth |
| Calculated force-equivalent noise | 2 / 20 = 0.10 N RMS |
Supporting scope: NI-DAQmx: Bridge Sensor Scaling ↗
03 / ENGINEERING DECISION
Compare the three measurement-chain layouts.
Place gain near the sensor when reducing the exposed low-level run is valuable, but verify local power, heat, sealing, available volume and calibration association. A conditioned output can improve transmission robustness while adding its own offset, gain, noise, delay and environmental limits.
Four-wire bridge systems and six-wire remote-sense systems have different compensation arrangements. Sense wires let a compatible instrument observe excitation at the bridge; they are not extra force channels. Follow the instrument’s approved connection when sense terminals or a cable balance module are present.
| Layout | Benefit to evaluate | Evidence to request |
|---|---|---|
| Raw bridge to remote instrument | Simple sensor head and flexible instrumentation | Cable-loss, pickup and input-range evaluation. |
| Local analog conditioner | Shorter low-level signal route | Input noise, zero/span drift, power, output scaling and delay. |
| Local digital conversion | Digital transport after conversion | Per-channel rate, timing, filtering, error handling and calibration. |
| Remote excitation sensing | Compensates approved lead-drop mechanism | Correct four/six-wire diagram and compatible sense input. |
Supporting scope: NI FieldDAQ: Remote Sensing and Shunt Calibration ↗ · Beckhoff ED336x: commissioning a measuring bridge ↗
04 / ENGINEERING DECISION
Change one cause at a time.
Choose a repeatable unloaded or reference-load condition with the machine in a controlled state. Compare mechanically stable, drive-off and drive-on conditions as appropriate to the installation. De-energizing a drive can change holding forces, so do not mistake that mechanical change for an electrical-only test.
A manufacturer-approved bridge simulator or matched dummy can help separate the instrument path from the actual sensor. It must match the input’s electrical requirements; an arbitrary short or resistor is not an equivalent load-cell source. Keep protective grounding intact and follow the sensor/conditioner EMC connection plan.
- Capture zero and known-load behavior with the original settings before changing anything.
- Check for loose fixtures, cable reaction, temperature gradients and overload history.
- Inspect excitation, signal range, sense arrangement, shield/case bonding and cable routing.
- If approved, compare a known compatible electrical source with the actual bridge.
- Compare noise spectra and force-equivalent noise at the same bandwidth.
- Apply one change, repeat the baseline and record both improvements and lost response.
Supporting scope: NI: SNR and excitation in strain measurements ↗ · NI FieldDAQ: Remote Sensing and Shunt Calibration ↗
05 / ENGINEERING DECISION
Worked example B: higher excitation also changes heat.
For an assumed 350 ohm bridge input, total input power is V²/R: about 71 mW at 5 V and 286 mW at 10 V. Doubling excitation doubles an ideal ratiometric bridge’s voltage signal but quadruples this resistive input power. The resulting temperature and drift depend on the actual sensor and installation.
This is not a recommendation to apply 10 V. Respect the exact sensor’s excitation limit and evaluate the warm-up trend and noise together. NI’s excitation guidance explains why a larger electrical signal can be offset by self-heating effects.
| Assumed bridge input | Excitation | Input power |
|---|---|---|
| 350 ohm | 5 V | 25 / 350 W ≈ 71 mW |
| 350 ohm | 10 V | 100 / 350 W ≈ 286 mW |
| Required decision | Within manufacturer limits | Compare useful noise and thermal behavior together. |
Supporting scope: NI: SNR and excitation in strain measurements ↗
06 / ENGINEERING DECISION
Use filtering only with a response budget.
For a hypothetical 10-sample moving average at 1,000 samples/s, the linear-phase group delay is (10−1)/(2×1,000) = 4.5 ms where that group-delay interpretation applies. Startup fill, input filtering, conversion, transfer and control scheduling can add further delay.
Filtering after sampling cannot recover information already aliased into the sampled band. Specify the relevant force bandwidth, suitable acquisition/anti-aliasing arrangement and the entire delay budget. A quieter displayed number can still be too delayed for force control.
Moving-average group delay = (window length − 1) / (2 × sample rate)
| Change | Potential benefit | Check before acceptance |
|---|---|---|
| Lower bandwidth / average | Less in-band noise for slow signals | Step response, phase/delay and lost transients. |
| Higher sample rate | More temporal information when supported | Actual noise, alias filtering and transport rate. |
| More gain | Uses converter range more fully | Clipping, input common-mode range and conditioner headroom. |
Supporting scope: NI: Aliasing and sampling considerations ↗
07 / ENGINEERING DECISION
There is no universal shield connection for every system.
The approved shield and case connections depend on the sensor, instrument, chassis bonding and interference environment. Use their documented EMC plan rather than treating the shield as a signal return or assuming every installation uses the same termination.
Keep power/drive wiring and low-level measurement routing appropriately separated, and record shield continuity and chassis connections during diagnosis. Do not disconnect protective earth as a noise experiment. If a cable contains trimming or compensation components, replacing it can change more than pickup.
Supporting scope: Beckhoff ED336x: commissioning a measuring bridge ↗ · NI: SNR and excitation in strain measurements ↗
08 / ENGINEERING DECISION
Request data for the complete chain, at the intended settings.
Include sensor range, rated sensitivity, excitation, cable length, connector, conditioner location and PLC/DAQ input. State the smallest useful force increment, relevant bandwidth, sample rate and allowable total latency. Request input-referred noise and zero/span drift under stated conditions.
For an MT-AMP-KIT review, ask for the paired sensor and conditioner serial association, output scale and actual sample data. A pre-calibrated chain still requires a stable installation and an application acceptance check.
Resolve the fit, signal and calibration questions.
Will a cable amplifier eliminate zero drift?
It can reduce the exposed low-level run, but its own electronics, heating and the mechanics can still create drift. Evaluate the complete installed chain under stated conditions.
Can I compare two RMS noise values directly?
Only when bandwidth, filtering, sample rate, observation conditions and input-referred scaling are comparable. Otherwise the smaller number can describe a different measurement task.
Should I always increase excitation for more resolution?
No. Stay within the sensor limit and compare useful signal, self-heating and drift. A larger voltage signal alone does not establish a better force measurement.
Do six wires provide two force channels?
No. A typical six-wire bridge connection adds excitation sense leads to the four excitation/signal functions. Follow the exact instrument and sensor diagram.
Can a digital filter remove aliasing?
A filter applied after sampling cannot distinguish an aliased component from a genuine in-band component. Define the acquisition bandwidth and anti-aliasing arrangement first.
Does the 0.10 N RMS worked example specify a product?
No. It is arithmetic using explicit assumed span and input noise. Request calibrated slope and noise data for the actual supplied configuration.
ENGINEERING RFQ / SAMPLE REQUEST
Send the load case.
Define the interfaces.
Include both thread ends, working and peak force, available space and the signal your controller expects.
Company and university email domains are preferred. Prototype and research inquiries are welcome.
inquiry@inlineloadcell.com ↗