For a vibrating wire sensor there is no single universal "temperature compensation". You have to calculate separately the effect of temperature on the transducer itself, on the thermistor and cable, and on the measured element. The correction should remove the measurement-chain error, not the structure's response. The coefficient comes from the documentation for the specific sensor and material, and the model is validated on data before and after correction. If you simply subtract a linear temperature trend, you may suppress real stress.
In brief
- The thermistor near the sensor is an input to the model, not a ready-made compensation.
- The wire, housing, structural element and surroundings may have different temperatures and different delays.
- For a strain gauge you need to distinguish the strain indicated by the sensor, the total material strain and the mechanical component related to load.
- A coefficient derived from historical regression is not automatically a physical coefficient. Correlation may describe real seasonal behavior of the asset.
- Good validation includes hot and cold days, different loading phases, raw data, the corrected result and cases where temperature is missing.
A correction that can fix the chart and break the interpretation
A vibrating wire strain gauge is working on a girder. Every sunny afternoon the result shifts by several dozen microstrain, and at night it returns. The team sees a strong correlation with temperature. The simplest idea is to calculate the slope of a straight line and subtract the temperature effect.
That can be correct, but it can also remove the real behavior of the structure. A heated girder expands. If its movement is restrained, stress develops. Uneven solar exposure creates a gradient through the section. A support, bearing or connection may block free length change. The sensor then shows not "noise" but the structure's response.
The GEOKON Model 4200 Series vibrating wire strain gauge instruction manual discusses both effects. Temperature affects the sensor itself, but thermal changes in concrete can also generate real stress, especially when deformation is restrained. The conclusion is uncomfortable, but practical: a prettier chart is not a validity check.
Four layers of temperature influence
Before building the equation, define which layer you want to correct.
| Layer | What changes | Should it be subtracted | How to verify it |
|---|---|---|---|
| transducer | the wire and housing react thermally | yes, according to the manufacturer's model | calibration sheet, test without load change |
| temperature measurement | thermistor, wires and connectors change resistance | correct the T measurement error | resistance check, comparison with a thermometer |
| mounting coupling | adhesive, anchors, cover and substrate have their own inertia | depends on the method | heating test and lag test |
| structure or ground | the material expands, stresses, changes pressure or flow | usually not as an "error" | physical model, reference sensors, load context |
The first two layers belong mainly to the measurement chain. The third sits at the boundary between sensor and object. The fourth is often what you really want to observe.
Wire temperature is not always the same as structural temperature
A thermistor placed in the housing describes the local sensor state well, but it does not guarantee representativeness for the full section. On the surface of a steel element, the cover may heat up faster than the girder core. A sensor embedded in concrete responds more slowly. A piezometer in soil may have a different inertia than the air in the cabinet.
If temperature and result are time-shifted, regression using the same timestamps will underestimate or distort the relationship. Sometimes the peak response appears two hours after the peak housing temperature. A fixed "T now" does not solve that problem.
The thermistor has its own circuit and its own faults
The wire transmits information through frequency, while a common thermistor uses resistance. A long cable and its connections affect these two measurements differently. The RST Instruments VW2100 vibrating wire piezometer instruction manual links unrealistically low temperature to an open circuit or excessive resistance, and unrealistically high temperature to a short circuit or leakage. It also describes checks for connections, corrosion and cable damage.
If the thermistor shows an error, the temperature correction may suddenly shift a correct wire reading. That is why temperature quality should block or flag the corrected result. You must not quietly substitute zero degrees.
Three different questions for a vibrating wire strain gauge
When talking about compensation, first decide which result should go to the dashboard.
1. What the transducer indicated after correction of its own sensitivity
This is the metrological level. From the raw frequency or digits you calculate the change using the calibration coefficient and apply the thermal correction of the sensor itself, as described by the manufacturer. The result should remove the part of the dependence caused by the difference in expansion of the transducer elements.
2. How the material length changed
Here you are interested in the total strain of the element. Free thermal expansion is a real change in length, so it should not disappear just because it can be predicted from temperature. If the report compares the measurement with geometric displacement, this level may be the right one.
3. Which part of the strain comes from mechanical load
In stress analysis, a designer may want to subtract the free thermal strain of the material and keep the part caused by restrained movement and load. That requires the material expansion coefficient, a temperature representative of the element, information about restraints and a consistent sign convention.
These three results are not synonyms. One project can store all of them, but each must have a different name, unit and formula. A channel named simply "strain corrected" leaves too much to guess.
A calculation model without false universality
A general form for many sensors can be written as:
result = readout_function(R, R0, coefficients) + sensor_correction(T, T0)
The meaning of R and R0 must be fixed first. They may mean Hz, period or manufacturer-specific digits, so before building the correction run the test described in the guide Hz, period, digits and vibrating wire sensor calibration.
If the goal is to separate free material strain, another term is added:
mechanical_result = total_result - α material × (T element - T0)
This is a scheme, not a formula to copy and paste. The manual for the specific model determines the sign, readout type, thermal coefficient and range. For a piezometer there are other effects: changes in fluid density, sensor position relative to the reference point, atmospheric pressure in the non-vented version and the temperature of the transducer itself. A strain gauge formula cannot be transferred to pore pressure.
Manufacturer coefficient, material coefficient and statistical coefficient
| Coefficient | Source | What it describes | Typical misuse |
|---|---|---|---|
| sensor thermal | sensor datasheet or manual | change in the response of the measurement chain | replacing it with regression from the asset |
| material expansion | material documentation or test | free change in length | using a catalog value for an unknown section |
| empirical slope | pair of data points from the asset | the observed relationship in a given period | treating it as a physical cause |
| lag | time analysis and test | shift of the response relative to temperature | fixed pairing of samples at the same timestamp |
Regression is an excellent diagnostic tool. It shows whether a strong dependence remains after applying the physical correction. It should not, however, decide on its own what to remove from critical data.
Example: a strut in an excavation and an apparently perfect compensation
Illustrative example. Two vibrating wire strain gauges and thermistors are installed on a steel strut. Over one month the strain shows a clear daily cycle. An analyst fits a line to the whole history and subtracts 5.8 µε for every degree. The correlation coefficient drops to almost zero. The chart is calm, and the number of alarms falls.
A week later an independent force measurement shows that part of the daily change was a real load from expansion of the bracing system between the walls. The regression removed both sensor sensitivity and structural response. Worse, the coefficient was created during the excavation phase, when the load was changing. The model mixed two processes.
The team returns to the raw data. The sensor coefficient is taken from the datasheet. Surface temperature is compared with a second point in the shade, and lag is assessed separately for heating and cooling. The data are split by construction phase. The total result remains on one channel, the sensor self-correction on another, and the mechanical result is published only after the model is approved by the designer.
The alarm does not use the automatically "smoothest" series. The designer selects the quantity that matches the action in the response plan. If the threshold concerns force in a strut, a force model with full documentation is needed, not arbitrarily corrected microstrain.
What revealed the error
- the slope differed between heating and cooling,
- two sensors in the same section had different lag,
- the relationship changed after the next excavation stage,
- an independent method confirmed part of the daily change,
- the regression correction was larger than the effect described for the sensor itself.
None of these signals was visible in a single correlation coefficient.
Procedure for building and accepting the correction
Step 1. Name the final quantity
Write one sentence that says whether you are calculating the sensor-corrected indication, the total strain, the mechanical component, pressure or another quantity. State the unit and the decision supported by the result.
Step 2. Secure the input data
Keep the wire reading, temperature, time, status and reference. If the logger already sends a corrected result, obtain its configuration and version. Without that, double compensation is a real risk.
Step 3. Confirm the sensor documents
Check the serial identifier, coefficient, sign convention, temperature range and any polynomial. Do not transfer values from a replacement sensor or a different type.
Step 4. Collect temperature that is representative of the question
The thermistor in the sensor may be enough for self-correction. To describe the structure, you may need an additional point, a section gradient, air temperature or data from several depths. The choice follows the physics of the object.
Step 5. Assess lag and operating phases
Compare the relationship for different time shifts and separately for heating and cooling. Mark work stages, load changes, rainfall and interventions. A model trained on a mixture of phases is often worthless.
Step 6. Build the model from documents, not from aesthetics
First apply the physical relationship. Use regression to check the residuals and find inconsistencies. If an empirical model is to go into production, it needs a validity range, validation on a separate period and engineer approval. Record the inputs, coefficients, units and version in the card from raw value to engineering unit.
Step 7. Validate the result before and after
Show the raw value, temperature, sensor-corrected result and final quantity in one view. Check events known from the construction log. The correction must not cut out load-related jumps or create jumps when temperature is missing.
Acceptance tests for the contract
| Test | Data | Acceptance criterion |
|---|---|---|
| constant temperature | stable section | the correction does not introduce a trend |
| daily cycle | full heating and cooling | lag described, no arbitrary shift |
| known load | test or documented event | mechanical response remains visible |
| missing thermistor | gap or bad status | no explicitly flagged result, no zero substitution |
| temperature step | field control or event | no non-physical jump in the final quantity |
| change of work phase | two separate periods | the model has a defined validity range |
| two adjacent sensors | common stimulus | the difference is explained, not silently averaged |
| validation period | data not used for fitting | error stays within the accepted criterion |
Do not write only "automatic temperature compensation" into the contract. A better clause defines the source of the coefficients, required inputs, behavior when they are missing, control samples, the approving person and the model version. That requirement can be accepted.
Alarms after correction: a separate decision
Changing the model may shift the current result relative to the thresholds. Do not assume the thresholds remain valid. The designer should specify whether the alarm operates on the raw value, the sensor-corrected value, the total value or the mechanical value. Each answers a different question.
After implementing the correction, check the historical number of warning and alarm entries. Do not send retroactive notifications, but document the difference. Then test threshold crossing, hysteresis, return and missing temperature. Configuration criteria are developed further in the guide on alarm thresholds in structural monitoring. An alarm based on a result that cannot be calculated should not silently fall back to OK.
If all exceedances disappear the day after deployment, treat that as a signal for inspection, not as success. The correction may be good. It may also be eating the measurement.
How it looks in Inclify
Inclify does not have a single built-in switch for "compensate every vibrating wire sensor". The platform accepts automatic input values and temperature over HTTP/JSON, stores time in UTC, and calculations are created in explicit project equations. This makes it possible to reproduce the model from the documentation of a specific sensor instead of forcing a common constant.
On the dashboard you can place the raw value, temperature and corrected series on a shared time axis. The temperature compensation report compares recognized channel pairs before and after correction: it checks the dependence on temperature and assesses whether a strong relationship remains. This is model diagnostics. The report does not prove that the removed dependence was a sensor fault, and it does not confirm the physics on behalf of the designer.
Each channel can have a warning and alarm threshold with hysteresis, and missing required data can be covered by a separate NO_DATA rule. Notifications are sent by email, SMS and in the app according to permissions and preferences. The user can acknowledge the event or mute it until a specified time. There is no rate-of-change alarm or automatic escalation chain.
A configuration change leaves the before and after values in the log. This does not replace a signed model card or calibration certificate. If a new rule is to cover history, a separate controlled run and acceptance are needed, described in the guide recalculating data after a calibration change.
Temperature model review checklist
- [ ] The final quantity has an unambiguous name and unit.
- [ ] The sensor effect is separated from the structural response.
- [ ] The coefficient comes from the document for the correct device or method.
- [ ] The material expansion coefficient and its source are recorded, if used.
- [ ] The thermistor has a range test and quality status.
- [ ] The long cable and connections are assessed separately for the temperature circuit.
- [ ] The time shift of the response is checked.
- [ ] The data are split by work and load phases.
- [ ] The fitting period and the validation period are disjoint.
- [ ] Heating and cooling are analyzed.
- [ ] Known mechanical events remain in the result.
- [ ] Missing temperature does not create a false zero.
- [ ] Thresholds refer to the named quantity after the correct correction.
- [ ] The input series and the copy of the configuration valid before the change are preserved.
- [ ] The engineer approved the physical sense, and the integrator approved implementation compliance.
Limitations
This article does not provide coefficients for a specific sensor. Even two sensors in the same category may require a different relationship, sign or correction. The manufacturer's manual and calibration sheet for the specific device take priority.
A temperature model does not replace measurement uncertainty evaluation. It also does not determine on its own whether the change comes from the structure, ground, water, installation or sensor. For that you need additional points, work context and sometimes an independent method. A broader interpretation tree is in the article temperature or structural change.
FAQ
Does every vibrating wire sensor need temperature compensation?
Every sensor requires an assessment of temperature influence, but the result of that assessment does not always mean an additional term in the equation. In some systems the effects of materials partially cancel out, while in others the manufacturer provides the correction. Regardless of the outcome, it is worth measuring temperature and checking the dependence in the data. The decision concerns the specific sensor, installation and final quantity.
Is the thermistor built into the sensor enough?
Usually it works well for self-correction of the sensor. It does not always describe the temperature of the whole element or the gradient through the section. If the analysis concerns the mechanical behavior of the structure, additional temperature points and a thermal model may be needed. First define the question, then choose the data.
Can the correction be determined only from regression?
You can build an empirical model, but you must not automatically treat the whole correlation as a sensor error. Temperature may drive real strain, pressure or support conditions. Regression requires a validation period, control of load phases, lag and approval of the physical meaning.
What should be done when temperature suddenly shows an extreme value?
Mark the corrected result as unreliable or unavailable and check resistance, connectors, corrosion, cable and comparison with an independent thermometer. Do not automatically substitute ambient temperature or zero. An open circuit and a short circuit can produce opposite extremes, as described in manufacturers' service manuals.
After changing the compensation, do you need to recalculate the whole history?
Not always. The scope depends on the cause. Correcting an error present since installation may justify recalculating earlier data, while a new calibration after replacement applies only from a specific moment. In both cases keep the previous version, recalculate in a controlled run and do not send alarms retroactively.
Which series should the alarm threshold be set on?
On the quantity that matches the decision and has an approved model. A threshold on total strain, mechanical component and raw reading means different things. The designer should name the series in the response plan, define behavior when temperature is missing and recheck threshold values after the model changes.
Sources and further reading
- GEOKON, Model 4200 Series Vibrating Wire Strain Gauges, Data Reduction, the effect of temperature on the sensor and real thermal stress in concrete.
- GEOKON, Model 4200 Series, Temperature Corrections, correction examples and the role of temperature.
- RST Instruments, VW2100 Vibrating Wire Piezometer Instruction Manual, diagnostics of the thermistor circuit, connections and moisture.
- USACE, EM 1110-2-1908: Instrumentation of Embankment Dams and Levees, planning, installation, reading and interpretation of geotechnical instrumentation.
- JCGM, International Vocabulary of Metrology, correction, the metrological meaning of correction and its dependence on the model.
If the daily temperature wave triggers alarms, or if the chart became suspiciously flat after correction, let's review one raw result-temperature pair. With a single sample it is often possible to tell quickly whether the problem lies in the data, the formula or the physics of the structure.