Technical Guides · Measurement Protocol
Calibration and Validation: A Practical Measurement Protocol
A force plate’s calibration matrix turns electrical outputs into forces and moments; validation asks whether that conversion still behaves when the plate encounters known loads at known positions. A 1,000 N vertical load at the plate’s center should read close to 1,000 N vertically, while horizontal forces and moments remain near zero within the system’s stated tolerances.
Published October 6, 2026
Calibration is not the same as validation
Calibration establishes the relationship between a plate’s sensor outputs and the quantities reported by its software. In a multiaxial system, that relationship is commonly represented by a calibration matrix. Validation tests the finished measurement chain against independent, known inputs: applied loads and moments that were not simply used to determine the matrix in the first place.
The distinction matters. A plate can fit its calibration data beautifully and still perform poorly at an off-center loading point, under a horizontal force, or after installation. A center-only vertical check may detect a gross scale error, but it says little about whether the plate resolves the center of pressure accurately near its corners. A useful validation protocol therefore varies load magnitude, direction, and position—and records enough information for another person to reproduce the test.
This guide focuses on that protocol: preparing the measurement chain, applying known loads safely, inspecting residuals and cross-talk, assessing repeatability, and documenting uncertainty. It is a practical check, not a substitute for the manufacturer’s calibration procedure or a laboratory’s quality-management requirements.
What the calibration matrix is doing
A force plate’s transducers produce electrical signals—often voltages or digitized counts—that respond to loading. The plate’s electronics and software combine those signals to estimate forces and moments in a defined coordinate system. In simplified form, the transformation can be written as:
y = Cx
Here, x represents the vector of raw sensor channels, C is the calibration matrix, and y is the output vector of force and moment components. The exact channel count, units, sign conventions, and matrix arrangement depend on the instrument. A typical output might include force components in newtons (N) and moment components in newton-metres (N·m).
The matrix accounts for more than a simple conversion factor. Real transducers can respond to loads in coupled ways: a force in one direction may influence more than one sensor channel. Matrix terms compensate for those relationships. That is why a validation test should examine the full output, not only the component that the operator intended to load.
For example, a nominally vertical load should produce a strong vertical-force response. A small horizontal reading may be caused by alignment, fixture friction, cross-talk, or an actual horizontal component in the applied load. The reading is not automatically evidence of a faulty plate; it is a prompt to examine the setup and the instrument together.
Before applying a reference load
Start with the manufacturer’s documentation. Confirm the plate’s rated capacity, permitted loading directions, coordinate axes, sign convention, specified warm-up period, and whether its software applies filtering or other processing by default. Never exceed the plate’s rated load to make a test more decisive. A validation load must be large enough to measure reliably and small enough to remain within the permitted operating range.
Inspect the installation. The plate should be mounted as specified, with the surrounding floor or mounting frame clear of contact that could bridge around the sensing structure. Check that cables are secure and that no cover, trim, or debris touches the plate in a way that carries load. A rigid object touching both plate and surrounding floor can create a parallel force path: some load bypasses the sensors, and the measured value may be misleading even if the electronics are functioning correctly.
Allow the system to stabilize for the interval recommended by its manufacturer. Record the room or equipment conditions when they are relevant, and acquire an unloaded baseline before testing. A zero offset is not a calibration, but a substantial or drifting baseline can compromise a comparison with known loads. If the baseline changes during the sequence, stop and investigate rather than subtracting the discrepancy without explanation.
Use a reference load whose value is known and whose uncertainty is documented. A calibrated mass provides a gravitational force that can be estimated from F = mg. For a 10.00 kg mass at standard gravity, 9.80665 m/s², the corresponding force is approximately 98.07 N. In a real laboratory, use the local gravitational value when the required accuracy warrants it, and account for the mass certificate, any additional loading fixture, and how the force is transmitted to the plate.
A hand-held push or a body-weight estimate is not an adequate reference for a quantitative validation. It may be useful for a quick functional check, but it does not provide the known input needed to estimate measurement error.
A practical validation sequence
Plan the test points before collecting data. Choose several load magnitudes within the working range, multiple application locations, and, where the setup allows, different loading directions. Include a center position and off-center positions distributed across the usable surface. Avoid placing a load dangerously close to an edge unless the manufacturer permits that procedure and the fixture is secure.
- Record the configuration. Note the plate model and serial number, software and firmware versions if available, mounting arrangement, coordinate convention, reference-load identification, and date.
- Capture an unloaded baseline. Record the duration, mean output, and variability for each force and moment channel. Check whether zeroing or tare functions are active and document when they were used.
- Apply a center load. Place the load at a measured position near the defined origin. Begin with a moderate load and allow the signal to settle before extracting a steady interval.
- Repeat at different magnitudes. Use more than one known load to look for scale error or nonlinearity. Where safe, include a sequence that increases and then decreases the load.
- Move the application point. Test representative positions across the surface. Measure the point relative to the plate’s coordinate origin rather than describing it only as “left side” or “near the corner.”
- Test other directions when feasible. A controlled horizontal load can reveal behavior hidden by a vertical-only check. Use an appropriate fixture and account for its geometry and friction.
- Repeat selected trials. Remove and reapply the load at least several times at important test points. Repetition helps distinguish a stable bias from inconsistent placement or loading.
Keep each application controlled. Do not drop a weight onto the plate: an impact produces a transient force that can greatly exceed the static weight and may damage the instrument. Use a stable loading frame or fixture designed for the task, and keep people clear of suspended or unstable masses.
For dynamic applications, static checks alone are not enough to establish the quality of a full measurement. A plate may pass a steady-load test while the acquisition chain, mounting, or signal processing still affects rapid events. Validate dynamic behavior with an appropriate method if the intended work depends on transient forces, and follow the manufacturer’s limits and laboratory procedures.
Sampling rate and the data you actually test
A validation result is meaningful only if the recorded data correspond to the intended measurement. Record the sampling rate and the signal-processing settings used during the test. For example, a static load sampled at 1,000 Hz may provide many observations, but those observations are not necessarily independent: filtering and the electronics’ bandwidth shape the signal. The sampling rate should match the stated test configuration, not be added later as an afterthought.
For a steady reference load, define a stable analysis window after the load has settled. Report how long that window is and whether you used a mean, median, or another statistic. Do not select a convenient-looking portion of the trace without a rule. For dynamic validation, document the acquisition rate, filter type and cutoff frequency, and any processing applied before comparison. These settings can change peak values and the timing of an event even when the underlying plate is unchanged.
Keep raw or minimally processed data when practical. A processed summary can be useful, but retaining the original channel data and the exact settings makes it possible to review unexpected results and reproduce the analysis later.
Position matters: moments and center of pressure
An off-center vertical load creates a moment about the plate’s axes. In a simple case, a vertical force Fz applied at horizontal distance x from the origin produces a moment magnitude related to Fz × x. The sign depends on the coordinate convention and the axis definition. This makes known loading positions especially useful: they test whether force and moment channels behave together as expected.
Consider a 1,000 N vertical load applied 0.20 m from the origin along the positive x direction. The corresponding moment about the relevant horizontal axis has a magnitude of 200 N·m, subject to the plate’s sign convention. If the software reports the force accurately but the expected moment is markedly different, check the position measurement, fixture geometry, coordinate origin, and calibration before concluding that the plate has failed.
Position-dependent errors matter because center of pressure is derived from force and moment signals. A plate can show a plausible vertical force while producing an inaccurate moment, and that moment error can shift the calculated center of pressure. A central load alone offers little protection against this failure mode; several measured positions provide a stronger test of the relationship.
Residual error, cross-talk, and repeatability
For each trial, compare the measured output with the expected output. The residual is the measured value minus the reference value. For the intended force component, calculate both the absolute error and, where the reference is sufficiently far from zero, the relative error:
- Absolute error: measured force − reference force, reported in N.
- Relative error: (measured force − reference force) ÷ reference force × 100%, reported as a percentage.
Relative error becomes unstable or meaningless when the reference value is near zero, so use an absolute error for unloaded channels and small cross-axis responses. Do not hide a substantial cross-talk signal behind a percentage whose denominator is nearly zero.
Cross-talk is an unwanted response in an output axis when the applied input is intended to load another axis. If a 1,000 N vertical reference load produces 1.5 N in a horizontal channel, the simple ratio is 0.15% of the applied vertical load. That ratio is a descriptive calculation, not a universal pass criterion. Compare it with the manufacturer’s stated performance, the laboratory’s acceptance limits, and the uncertainty of the setup.
Repeatability describes how closely the plate returns to the same result under repeated conditions. Report the individual readings, their mean, and a measure of spread such as standard deviation or range. If the readings drift with time or differ according to whether the load is being added or removed, investigate settling, hysteresis, mechanical interference, or fixture movement. Averaging inconsistent trials does not make the underlying problem disappear.
Look at the pattern of residuals, not just one summary number. A similar percentage bias at every magnitude suggests a scale-factor issue. Errors that grow disproportionately at higher loads may suggest nonlinearity or an overloaded component. A good center result paired with increasing errors toward one side points toward a possible position-dependent issue, installation constraint, or location-measurement error. These patterns guide troubleshooting; they do not, by themselves, identify a single cause.
A worked example
Suppose a plate is tested with a reference vertical load of 1,000 N at its center. The steady-window mean is 992 N. The vertical residual is −8 N, and the relative error is −0.8%. During the same trial, the horizontal outputs are +1.5 N and −2.0 N. Those readings should be recorded, not discarded merely because the test was intended to check vertical force.
Next, the same reference load is applied 0.20 m from the origin. If the expected moment magnitude is 200 N·m and the plate reports 194 N·m, the moment residual is −6 N·m, or −3% relative to the expected magnitude. That result deserves investigation even though the vertical-force result is relatively close to its reference. The moment discrepancy may affect position-derived outputs such as center of pressure.
These figures illustrate calculations, not recommended universal tolerances. Whether −0.8% or −3% is acceptable depends on the instrument specification, the reference-load uncertainty, the procedure, and the measurement question. A study tracking a large change in a robust force outcome may have different requirements from a study estimating small shifts in center of pressure.
Document uncertainty, not just pass or fail
Every reference comparison has uncertainty. The reference mass or load may have a stated uncertainty; the force estimate may depend on gravitational acceleration; the application point may be measured imperfectly; and the fixture can introduce friction or misalignment. The plate’s own repeatability and resolution also contribute. Treating the reference value as exact can make a comparison appear more decisive than it is.
For an off-center load, location uncertainty affects the expected moment. If a 1,000 N load is believed to be 0.200 m from the origin but the point is uncertain by 0.002 m, the location contribution to moment uncertainty is approximately 2 N·m, before accounting for uncertainty in force and other sources. That is material when judging a 6 N·m difference.
Build a concise uncertainty budget around the test’s purpose. Identify the principal contributors, record their sources or estimates, and state whether the result is a simple comparison or an uncertainty-aware evaluation. If the decision is close to an acceptance threshold, the uncertainty can change the conclusion. A bare “pass” conceals that ambiguity; a result with its method, conditions, and uncertainty is more useful to a researcher deciding whether the plate is suitable for a specific protocol.
Use acceptance limits supplied by the manufacturer or defined by the laboratory before examining the results. Setting limits after seeing the data invites selective interpretation. If no suitable limit exists, report the measured performance transparently and ask whether it is adequate for the intended measurement rather than inventing a universal tolerance.
Troubleshooting a failed or unexpected check
When a result falls outside an established limit, first verify the test rather than immediately adjusting the calibration. Confirm the reference value and units, the load’s position, the coordinate origin, the sign convention, and the software settings. Check that the fixture is not touching surrounding structures and that no cable, cover, or debris is creating a parallel load path.
Repeat the measurement after removing and carefully reapplying the load. A result that changes substantially with placement suggests setup sensitivity or poor repeatability. A stable offset across repeated trials points toward a different issue, but still does not prove that recalibration is the answer. Record the original data before changing any settings, and follow the manufacturer’s service and recalibration instructions.
If the system has been moved, remounted, overloaded, repaired, or exposed to an unusual impact, document the event and treat it as a reason to perform appropriate checks before collecting research data. A new center reading alone may not be enough when the event could have affected alignment or position-dependent measurement.
A record that can stand up to review
Preserve a compact record with enough detail to reproduce the validation. At minimum, include the date, operator, plate identification, acquisition configuration, sampling rate, filter settings, reference-load value and uncertainty, load-application locations, number of repeats, steady-window definition, and results for all measured force and moment channels. Attach raw data or a clear summary table when the work warrants it.
Record deviations as well as successes: a load that slipped, a baseline that drifted, a fixture that needed repositioning, or a trial excluded under a stated rule. Do not quietly remove inconvenient observations. A validation history allows a laboratory to distinguish gradual changes from one-off setup errors and supports a defensible decision about whether data collected on a particular day remain fit for purpose.
Calibration establishes the conversion; validation tests its performance under defined conditions. For a biomechanics laboratory, the most informative check is not the one that produces a reassuring single number. It is the one that varies magnitude and position, exposes unwanted axis responses, quantifies repeatability, and makes uncertainty visible before the first participant steps onto the plate.
Protocol checklist
- Confirm manufacturer limits, coordinate conventions, warm-up, and processing settings.
- Use a documented reference load and a safe, stable loading fixture.
- Record unloaded baseline and sampling rate.
- Test multiple load magnitudes and measured application positions.
- Inspect intended force and moment outputs, residuals, cross-talk, and repeatability.
- Document reference and positioning uncertainty, deviations, and acceptance criteria.
- Retain the original results before making adjustments or recalibrating.