Capgo datalogger, data logger, datalogging and data logging.




Introduction to Temperature Measurement

Temperature is the most commonly measured parameter, yet in many respects it is the least understood. It is a surprisingly difficult parameter to measure with the precision that one might reasonably expect.

To obtain accuracies better than 0.2°C (0.4°F) great care in needed. Errors occur due to the presence of temperature gradients, drafts, sensor nonlinearities, poor thermal contact, calibration drifts, radiant energy and sensor self heating.

Generally the accuracy of all sensor types can be greatly improved by individual calibration. For more information, refer to the appropriate page on each sensor type.

The information in this section is oriented towards electronic thermometers - those with an electrical output that can be connected to a measuring instrument, such as: a data acquisition system, a data logger, a control system or a chart recorder.

However, there is also a wide range of thermometers that can be used for manual temperature measurement. These include: the glass thermometer, various gas thermometers, pressure based thermometers, bimetallic thermometers and temperature sensitive paint or film thermometers.

Is temperature measurement difficult?

The answer depends on the temperature, the material being measured and your expectations of accuracy . The table below summarises the difficulty of temperature measurement over a range of temperatures:

Accuracy Required
care needed
very difficult
0°C to 50°C
care needed
very difficult
care needed
very difficult
extremely difficult
almost impossible
very difficult
extremely difficult
almost impossible
don't even try

In a laboratory with appropriate standards and equipment, it is possible to measure temperature to 0.001°C (1°mC) or even better. This is typically done by interpolation (estimation of the values) between two standards, using a quality platinum temperature sensor and / or a Type S thermocouple.

When measuring temperature it is important to keep your goals in mind. Identify exactly what is to be measured and the accuracy needed. If accurate temperature differences are of prime importance, then consider using the thermopile to avoid the need for closely matched sensors.

Sources of temperature measurement error

In using temperature sensors it is helpful to think of where heat flows. This applies to both sheathed and unsheathed sensors. Understanding the thermal resistances and where they are located is especially useful in identifying potential errors sources.

The following diagram indicates some of the complexity in temperature measurement. Note the presence of thermal gradients in the material being measured. These gradients can be particularly troublesome when measuring the materials with poor thermal conductivity, such as plastics and even stainless steel.

Below is a description of temperature measurement error sources and some suggestions on minimising these errors.

1. Sensor calibration

Sensors calibration errors can be due to offset, scale and linearity errors. In addition, each of these errors can drift with time and temperature cycling. Hysteresis (where a value depends on the direction from which it was approached) can be noticed with some sensors, but the effect is usually small with the exception of the bimetallic strip where it may be several degrees. Platinum RTD's are considered the most accurate and stable of standard sensors. However, individually calibrated thermocouples can come close over the same temperature range. The platinum based thermocouples can be just a stable as platinum RTD's and cover a higher temperature range.

Sensor interchangeability is often the decisive factor. It refers to the maximum temperature reading error likely to occur in replacing a sensor with another of the same type without recalibrating the system.

Choosing a practical calibration reference can be an issue. For professional purposes, a high quality platinum RTD (Class A, band 5) is best, along with an appropriate indicator. Other references include iced water bath, traditional glass thermometers (especially laboratory grade) and medical thermometers. In general, the defining reference points of the ITS-90 are not practical for routine calibration purposes.

2. Thermal gradients

Thermal gradients are often a major source of measurement error. This is especially true when measuring materials with poor thermal conductivity such as: air, most liquids and non-metallic solids.

In the case of fluids it important that the fluid be stirred. An unstirred ice bath (a mixture of ice and water) can have a vertical temperature gradient of several degrees. If stirring is not practical, gradients can be minimised by insulating the system being measured, to prevent heat transfer into or out of, the system. Employing multiple sensors for spatial diversity and averaging the readings is another solution.

3. Heat conduction in sensor leads

All sensors with the exception of non-contact and maybe the fibre optic types require that wires be brought to the sensor. These wires are usually copper, an excellent heat conductor. The placement of these wires can have a significant impact on accuracy.

The wires allow heat flow into or out of the sensor body, requiring the sensing element to be in better thermal contact with the material being measured than would otherwise be needed. When measuring the temperature of thermal insulation materials, this can be a major source of error.

There are three solutions, all of which are good standard practice:

  • Use as thin wires as is practical for sensor hook-up. (Note: this contradicts good practice for high temperature thermocouple measurement where the reverse rule applies - use the thickest wire that is practical)
  • Place the wires in or against the material being measured, so that they actually assist in transferring the temperature of the material into the sensor

  • Attempt to minimise the thermal gradient along the sensor wires by placing the wires at an angle to the gradient. This ensures a higher thermal resistance because of a longer length of wire.

4. Radiation

Radian heat can be a major source of error in measuring air temperature. A sensor in sunlight is almost certain to read significantly higher than the actual air temperature. To avoid this error the sensor must be shielded from source of radiant energy. The sun is the most obvious source, however just about any object that is at a different temperature to the air is a potential source (or sink) of troublesome radiant energy.

The best solutions are the following:

  • Ensure that the sensor's outside surface is highly reflective of infrared radiation ( is painted white or has a bright metal finish)
  • Ensure the sensor is thermally 'well connected' to the air by having a good surface area-to-volume ratio. Small sensors are generally better.
  • Place the sensor in a vented radiation shield that also has a highly reflective surface on the outside and inside
  • Ensure the sensor has a high surface area-to-volume ratio to ensure good thermal 'contact' with the air

Radiant heat loss can be a source of sensor error when measuring elevated temperature. Again, the same rules apply. Use reflective surface finish on the sensor, shield the sensor if possible, and ensure a good thermal contact with the medium being measured.

5. Sensor self-heating

Thermistors, RTDs and semiconductor sensors require the application of an excitation power in order that a reading may be taken. This power can heat the sensor, causing a high reading. The effect depends on the size of the sensing element and the level of power. Typically, the magnitude of the self heating effect is between 0.1°C and 1.5°C.

The best solutions are the following:

  • Calibrate out the self-heating effect. This is perhaps the easiest solution. However, the equipment must be allowed time to 'warm up', and different calibrations are required for mediums with different thermal characteristics e.g. air and water
  • Use the lowest possible excitation power. However, a compromise between self-heating and sensitivity (and signal-to-noise ratio) must be made
  • Avoid unnecessarily small sensing elements - they will self-heat more than larger elements
  • Switch the excitation power off between readings. This is the best solution if the readings can be made quickly, before the sensor has time to warm, and if there is adequate time between readings for cooling
  • Avoid self-heating sensor types - use thermocouples. However this is not necessary as simple as that, as the measuring device is likely to use a reference junction temperature sensor that is itself prone to self-heating.

5. Thermal contact

Obviously, thermal contact with the material being measured is important, but the degree of contact required is dependent on other parasitic thermal connections to the surroundings which are likely to have a significant impact on heat flow. These parasitics include: lead conduction, direct contact with other material (e.g. air) and radiant energy transfers.

If there is no temperature gradient in the vicinity of the sensor, the thermal contact of the sensor can be poor and the sensor will still provide accurate readings.

6. Thermal time constant

When the temperature changes, it takes time for a sensor to respond. Some sensors respond quickly, some in less than a second, while others take minutes or even hours.

The time taken to reach 63% of the way to the new temperature is referred to as the 'thermal time constant'. Most sensors have one dominant time constant. However, sometimes, there are minor, but longer, time constants present that can confuse the measurement process. See the Measurement Methods section for further details.

Obviously if the temperature is changing more quickly than a sensor is able to track, the measurement will be in error.

The best solutions include the following:

  • Use a more rapidly responding sensor
  • Improve thermal contact
  • Reduce the sensors thermal mass, by minimising material in contact with the sensing element that is not associated with improving thermal contact
  • Compensate the readings using an inverse matching filter. If the thermal characteristics of the system are constant and known, it is possible to predict the temperature dynamically.

Sometimes long time constants are useful in providing an averaging effect on a rapidly fluctuating temperature. If this effect is to be exploited, care needs to be taken to compensate for the phase (time) delay in the response.

7. Read-out errors

The measuring device connected to the sensor is never perfect. The measuring device, be it a: meter, chart recorder, data acquisition board or data logger, can have calibration, linearity and temperature dependent errors.

These errors can be reduced by:

  • Calibration of the readout device against known references
  • Calibration of the total sensor - readout system, using a reference temperature or against a precision thermometer (glass or RTD+readout)

Temperature effects on the readout device can be a subtle source of error. It is recommended that a test be conducted where the temperature of the sensor is held constant, but the readout device be placed in an oven or freezer. This is particularly important for thermocouple read-out devices, as their performance can be greatly impacted by temperature gradients and the quality of the internal reference junction sensing.

8. Electrical noise or interference

Electrical noise can induce errors in systems with poor noise rejection. Use the standard procedures as described in the Measurements Methods section to minimise the impact. These include the following:

  • Use shielded twisted pair cable
  • Keep wiring away from power cables, transformers and electrical machinery
  • Install low pass filters into the measuring device
  • Avoid ground loops

In some industrial processes, electrical noise can be so intense that non-contact or fibre-optic sensors are the only option.

9. Condensation

Sometimes in situations where temperatures are frequently cycled through the dew point, condensation in the sensor and wiring can collect and become an electrical leakage path, causing errors. Prevention is better than cure.

Preventative measures include the following:

  • Ensure that the sensor and its wiring is sealed. When an air volume is sealed, ensure that the air is dry
  • Moisture can wick along wires by capillary action. This can be inhibited by sealing wires at the sensor end, with a medium-viscosity super glue, or using a penetrating oil from both ends.

Condensation can be a source of intermittent errors and may go unrecognized for a long time. It also can lead to corrosion, accelerated by sensor excitation power, ultimately leading to complete failure. Semiconductor sensors can be particularly prone to moisture penetrating the metal-plastic interface of plastic packages.

Evaporating condensation can also lead to measurement errors due to evaporative cooling effects - a subtle but real error source.

10. Sensor mechanical stress

Many temperature sensor elements can respond to mechanical stress. For example, film type RTD has the appearance of a strain gauge and will behave like one given the opportunity. Again, prevention is better than cure.

Preventative measures include

  • Ensure that sensing elements are not subjected to deformation in the way they are mounted
  • Avoid using adhesives in attaching sensors to the surfaces to be measured. The differences in the thermal coefficients of linear expansion will induce mechanical stress
  • Use sensor that are less sensitive to stress - for example, the thermocouple
  • Wound (as opposed to film) RTDs can be prone to vibration damage. Take great care in the selection and mounting of sensors in high-vibration environment
  • Use grease in preference to adhesive to ensure thermal connection. Avoid potting sensor elements in epoxy


If funds are available, special purpose temperature calibrators can be purchased. However, these are expensive and not always as accurate as one might expect.

To calibrate over the normal environment range, there are two low cost standards that can be effectively employed to achieve 0.2°C accuracy. These are ice-water mix made from distilled water and a standard medical thermometer. These points (0.0°C and about 37°C) are sufficiently far apart to provide a useful calibration over the -20°C to +60°C range.