How To Choose The Right Temperature Sensor

Introduction

This paper is the first of two parts and examines the tradeoffs associated with selecting a conduction type (as opposed to radiation type) temperature sensor for your application. The construction and operation of Thermocouples, Thermistors, and RTD's and the limitations of each in terms of durability, range of operation, inherent accuracy, and susceptibility to external noise influences are reviewed. In part 2, we explore how a temperature sensor can be calibrated to minimize error and how the sensor measurement capability may or may not be affected by neighboring heat sources, placement, and environmental conditions including transient phenomena that contribute to measurement uncertainty. Final sensor selection will depend on the details of the environment where it is to be used.

---

The Application

Each sensor type has its own strengths and weaknesses. So, to help with the selection process, we first need to look at how the sensor will be used in terms of:

• Temperature Range
• Accuracy
• Repeatability
• Endurance and Durability with respect to handling and installation
• Calibration
• Grounding
• Environment of Use:
  • Atmospheric composition, ambient temperature, thermal gradients
  • Local electrical, magnetic, RF, and acoustic Noise
  • Local static and transient heat sources and sinks

---

Thermocouples

Thermocouple Construction

Thermocouples are constructed by joining two dissimilar wires together at a junction. Physically, mineral-insulated, sheathed thermocouples can be as small as 0.01" while an unsheathed (a bare-wire junction) can be made from small gauge wire or ribbon. The adjacent figure shows several options of how a thermocouple may be constructed. Thermocouples exhibit the greatest ability to endure thermal and mechanical extremes. They are limited only by the oxidation and melting temperatures of the alloys used. For that reason, they can often be manufactured inside stand-alone heater assemblies and can survive short duration welding operations inside a metal encasement. Thermocouples are the smallest, simplest, and most enduring of the three sensor types.

Operation

Thermocouples passively sense the temperature difference between hot and cold junctions. So, in general, to know the hot junction temperature, you must already know the cold junction temperature by some other means. Thermocouples are generally the least sensitive of the three types of sensors and, though they have very low source impedance, care must be taken to ensure external noise does not contaminate the desired signal. Type E and T thermocouples are the most sensitive of the standard types at roughly 60µV/ C and 40µV/ C respectively. If the application requires sensitivity to 0.1 C, the signal from these sensors can tolerate (at most) noise levels of about 1µVrms. While the self-noise of these devices are can be ignored when the signal is properly band limited, thermocouple signals can be disturbed by induced currents from electromagnetic sources or offset voltages from nearby potentials. This problem is magnified when extremely small temperature resolution is required because the voltage gradient induced by the temperature difference between hot and cold junctions is actually generated along the entire length of the wire, not at the junctions (it is a very low impedance path).

Mitigation from these problems can be obtained through using short runs of insulated and shielded thermocouple wires with balanced, low-pass filtered differential amplifiers (to avoid common-mode voltage offsets). We note here also that differential filtering in the electronic signal-conditioning path can also be a source of error. The table below shows the typical temperature ranges for various thermocouple types.

Thermocouple Strengths and Weaknesses

Small, Durable, and Robust

Thermocouples are robust and durable passive sensors that can withstand very high temperatures and mechanical punishment. They are some of the smallest and simplest sensors to make and operate. Because of their relative smallness, they are fast reacting and the sensing junction can often be placed very close to the desired point of measurement. Their durability and simplicity make them ideal for embedding into other devices.

Cold Junction Error

As previously stated, two measurements are required to obtain absolute temperature when utilizing thermocouples. If the temperature at the cold junction is uncertain, then this adds to the total uncertainty of the hot junction temperature. Modern temperature measurement devices typically terminate the two thermocouple wires on an electrical block and measure the temperature of the block with either an RTD or thermistor as a reference sensor. So, errors associated with the reference sensor contribute to the total uncertainty of the system. However, there may also be undesired thermal gradients between any two of the three sensing points (i.e. the two terminals of the thermocouple block and the location if the reference sensor) that can induce additional error. For instance, if the cold junction sensor is located nearer to a heat source than the reference junction, then there will be a small temperature difference between them that will translate to the measured temperature (at the hot or sensing junction) to read higher than it should.

High Temperature Hysteresis and Deterioration

For measuring temperatures above about 650C, thermocouples are typically your only choice for in-contact temperature sensing. However, when thermocouples are operated at high temperatures and/or deeply cycled from low to high temperature, they may exhibit two kinds of problems. First, Hysteresis - that is, the voltage they produce for a certain temperature may be dependent upon the time-temperature history of the sensor. While repeatable, the thermoelectric voltage generated by the difference between cold and hot junctions may be different when the temperature is falling compared to when is rising. Second, if operated for prolonged periods near their temperature limits, thermocouples are subject to oxidation. This deterioration causes the sensor to deviate from its initial voltage/temperature relationship. Thermocouple standard types B, K, R, S, and the special types are typically used in high performance applications. They are typically enclosed in special insulation materials to meet demanding application temperatures such as in manufacturing, engine/propulsion, nuclear, and other furnace-like environments.

Homogeniety Weaknesses

All of the standard thermocouple types utilize at least one alloy as a conductor. Because the thermoelectric voltage of a thermocouple is generated along the entire length of the wires, deviations in metal purity and alloy homogeneity along the length of the wire can result in thermocouple temperature profiles deviating from NIST standards. Homogeneity problems can also be compounded when long runs are needed. When high accuracy is needed without calibration, use a thermocouple type which consists of a minimum number of elements like a type T, J or G.

---

Thermistors

Thermistor Construction

Thermistors are constructed from sintered metal oxide in a ceramic matrix which changes electrical resistance with temperature. The most common form of the thermistor is a bead with two wires attached. The bead diameter can range from about 0.02" to 0.2'', but since the sintered metal oxide material is prone to damage by moisture, glass or epoxy is used to encapsulate it. As a result, the diameter of a thermistor is limited to about 0.1". Thermistors cannot endure high temperatures or high mechanical stresses, so they are unlikely to survive embedding into a cartridge heater. Use of a thermistor in this application would therefore be limited to cold bonding on to the thermal block, external to the heater.

Thermistor Operation

Thermistors change resistance logarithmically in response to temperature and have a maximum sensitivity typically around room temperature. A typical 1K ohm glass-bead type thermistor can exhibit a 2:1 change in resistance over a 15-30?C range, averaging 7ohms per 0.1?C change. Self-heating has a minimal impact on the measurement if current is held to 1ma or less, making the voltage sensitivity 100 times better than the best thermocouple . Consequently, thermistors are the least susceptible to external sources of contamination of the three types of sensors. Additionally, the internally generated noise of a thermistor is primarily dependent upon its resistance, which is low enough that it can be ignored. There are no standards for thermistors such as there are for thermocouples and RTDs. So calibration curves and fitted equation parameters are usually provided. While thermistors can be made quite uniformly in batches, batch-to-batch uniformity can be a problem if high precision is required. Thermistors are typically used over a range of -100 to 300?C but their sensitivity to temperature diminishes as temperature gets further away from their area of maximum sensitivity. There are special versions of thermistors capable of working to temperatures of 1000?C. While there are both positive temperature coefficient types (PTC) and negative temperature coefficient types (NTC), here we are primarily describing NTC types. NTC thermistors become less resistive as temperature goes up. Because of the thermistor's inverse relation between resistance and temperature, there may be more potential for thermal runaway when using NTC thermistors as a feedback sensor in a temperature control situation.

Thermistor Strengths and Weaknesses

High Sensitivity, Accuracy, and Signal Strength

Thermistors are best used in a narrow temperature range where very high precision is required. Although the maximum sensitivity point of a thermistor is normally "tuned" near room temperature, special versions are available with maximum sensitivity tuned to other temperatures. Near their maximum sensitivity point, small changes in temperature produce relatively high changes in resistance. Away from the maximum sensitivity point, thermistors are less able to resolve changes in temperature. While a 2:1 change in resistance is typical in the 15-30?C range, a 15 degree change near maximum temperature produces a resistance change of closer to 1.2:1.

Simple Electronics

Thermistor interface electronics can be as simple as connecting the thermistor through a pull-up resistor and measuring the resulting voltage across the thermistor. This is often possible because the thermistor resistance usually dominates resistances associated with wiring and connection. For more precise measurements however, additional care must be taken to compensate for the non-thermistor resistances. When anticipated temperature range is relatively small, such that the resistance varies by less than a factor of 100 (say from 10K to 100ohms), then the voltage source may effectively be replaced with a current source.

Self Heating

Current flowing through the thermistor dissipates power inside the sensor which consequently increases the temperature inside the sensor. When used with a voltage source and a pull-up resistor, current can be effectively limited by the pull-up resistor to reduce self-heating. In general, power dissipated by the thermistor should be limited to a few milliwatts to maintain temperature accuracy. This can be accomplished by either limiting the continuous current through the thermistor or applying a current duty cycle such that, for instance, current is only flowing through the thermistor during measurement time. A 10% measurement duty cycle will reduce the power dissipated inside the sensor by a corresponding amount.

No Standards

Thermistors do not follow NIST-traceable standards and so there may be large variations in response from batch to batch or from manufacturer to manufacturer.

Fragile

Generally, the sintered metal oxide thermal element is brittle so mechanical stresses can alter or destroy its performance. To compensate for this, the sensor may be encased in a protective metal enclosure at the cost of thermal responsiveness.

Response Time

Thermal response is tied to the size and mass of the sensor and its encasement. For a thermistor to approach the response time of a thermocouple it would need to be roughly the same size and mass. Typically, thermistors are larger and more massive than thermocouples.

Averaging Sensor

Thermistors change resistance because the Temperature Coefficient of Resistance (TCR) is active for the entire thermistor element. Therefore, the total resistance change of a thermistor will always be indicative of a temperature that lies somewhere between the extremes of high and low temperatures that exist across the element. As these elements become concentrated and small, they become more accurate temperature measurement requires a uniform temperature across the RTD.

---

RTDs

RTD Construction

The figure below shows three types of RTD elements. The smallest RTD elements are usually thin films of platinum on a ceramic chip substrate. If encapsulated, they are surrounded by an insulator and enclosed in a sheath of metal. Their construction in this respect is similar to a cartridge heater. Platinum is normally used for the element, but nickel, copper, and nickel-iron are available as well. RTDs are very delicate and, while the melting temperature of an RTD element is sufficiently high to survive many high temperature manufacturing operations, they tend not to survive many mechanical operations such as compaction. The minimum size of a typical RTD is about 0.1" which is about 10 times larger than the smallest thermocouple.

RTDs are used in industrial applications as well as laboratories. RTD elements are typically more accurate, stable, and repeatable then thermocouple elements and they maintain that accuracy over a longer period of time. They are generally used from -200C to 650C. Because of their size, an RTD is typically slower in response than a comparable thermocouple.

RTD Operation

Several commonly used metals exhibit a quasi-linear change in resistance as a function of temperature as shown in the table below. RTDs utilize that property to measure temperature and for a platinum-based RTD, the element resistance changes about 0.0039 ohms/ohm/?C. While this equates to an output perhaps two orders of magnitude less than a thermistor at its most sensitive area, the accuracy, and precision of an RTD often exceeds that of a thermistor or thermocouple.

Besides the standard 2-wire configuration, RTDs may come in 3-wire, or 4-wire versions as shown below:

For a typical 100ohm RTD, wire and termination resistance can become a significant source of error. Therefore, to achieve the highest accuracy, three or four-wire RTDs are often used. Resistance is measured as is typically done in 4-wire ohm meters where a current source is applied between terminals 1 and 3 and a high impedance voltage measurement is made between terminals 2 and 4, or 2 and common. If a typical 1-milliamp source is used to drive an RTD (to limit self-heating), detection of a 39milliohm change will occur at voltages roughly equivalent to that generated by a thermocouple. Therefore the same kinds of induced and potential noise from external sources can create measurement problems when using an RTD. Noise problems may be mitigated in much the same way as thermocouples - i.e. using differential, ungrounded, and shielded elements.

RTD Strengths and Weaknesses

Accurate, Precise, and Consistent

RTDs follow NIST-traceable standards, and with good tolerance specifications, off-the-shelf RTDs will be very consistent with one another independent of the batch they come from. RTDs also do not exhibit much hysteresis - that is, they will measure the same temperature regardless of whether they approach it from below or above the setpoint. Platinum RTDs are generally preferred over Nickel (which is more non-linear) or Copper (which is subject to more oxidation) because of their long term stability and repeatability.

Lead Resistance Compensation

The use of 3 and 4-wire RTDs are intended to reduce errors associated with long lead lengths and multiple connections. For a 100ohm RTD, 1 ohm of lead resistance or connection resistance can become a major source of error. So, electronics are constructed to dynamically remove the error associated with lead resistance that includes any lead resistance changes due to temperature or aging. The trade-off for this is in both cost and the number of wires that must be run to perform a measurement.

Self-Heating

For high resolution applications, self-heating can be a problem for RTDs. Unlike thermocouples, RTDs need to be driven by an external power supply to determine their resistance (and so determine their temperature). They are therefore subject temperature measurement error due to the heat dissipated in the sensor. When ultra-high precision is required such as to achieve accuracies to 0.1?C, care must be taken not to over drive the sensor. A 10-milliamp current into a 100ohm RTD dissipates 10 milliwatts inside the sensor if left on continuously. An imperfect or low thermal conductivity sensor bond can cause the internal temperature of the sensor to rise well above the actual temperature to be measured and will continue to rise until the equivalent amount of power is dissipated through the casement to the environment. Typically, 100 watts (1milliamp drive current) is acceptable to limit self-heating. Optional electronics are also available to perform 10% duty cycle measurements and so limit self-heating power without reducing signal strength. The trade-off for utilizing low level signals (power) to drive an RTD is that extra measures may need to be taken to ensure that external noise does not become a factor.

Averaging Sensor

Because an RTD consists of a relatively long length of wire or film and because the Temperature Coefficient of Resistance (TCR) is active for every part of the RTD element, the total resistance change of an RTD element will always be indicative of a temperature that lies somewhere between the extremes of high and low temperatures that exist across the element. Therefore, accurate temperature measurement requires a uniform temperature across the RTD.

Fragile

While RTDs can handle relatively high temperatures, they are usually mechanically fragile due to the low mass substrates the thin film or wire is mounted to. As such, they are often difficult to embed reliably into custom mechanical devices. As a result, many manufacturers can provide metal sheathed assemblies that remove the fragility at the cost of response time.

Response Time

Thermal response is tied to the size and mass of the sensor and its encasement. For an RTD to approach the response time of a thermocouple it would need to be roughly the same size and mass. Some new specialty RTDs used for medical applications can come near this.

---

Conclusions

The thermocouple is the smallest, fastest, and most durable temperature measurement solution and the closest to a "point" sensor, but it is also the sensor type most subject to accuracy, noise, and precision error. For extreme accuracy and precision, many of the thermocouple shortcomings may be compensated for by using short wire runs connected into an isothermal block and by using relatively complex calibration procedures. The type T thermocouple is recommended as the second highest output of the standard types with the least amount of homogeneity problems of the standard types.

The thermistor is the recommended choice where high sensitivity is required over a relatively narrow range of temperatures (less than 300?C). Thermistors are less subject to the different types of error. Local signal conditioning is still recommended, but it is much simpler. The larger size of this sensor means that it reacts slower and may be subject to additional location and heat transfer error relative to an equivalently placed thermocouple. Because there is no NIST standard for thermistors and they are subject to batch-to-batch variations, arrays should be formed from a single manufacturing batch to help with accuracy. A relatively complex calibration procedure may be required to obtain actual temperature curves.

An RTD is recommended where extreme stability and precision are needed and where accuracy over a prolonged time is important. The electronics and the wire-count to support an RTD are generally more than the other sensors, but using NIST traceable units eliminates the batch-to-batch problems of thermistors.

The following table summarizes the strengths and weaknesses of each type.

	Thermocouple	Thermistor	RTD
Maximum Operating Temperature(°C)	2300	1000	850
Minimum Operating Temperature(°C)	-200	-100	-200
Temperature Curve	linear	non-linear	linear
Point or Area measurement	point	are	are
Maximum Sensitivity Region	No	Yes	No
Measurement Parameter	Voltage	Resistance	Resistance
Temperature Measurement	differential	absolute	absolute
Stimulation Electronics Required	None	Yes	Yes
Self-heating Issues	No	Yes	Yes
Typical Output Levels per degree C	volts	millivolts	volts
NIST standards	Yes	No	Yes
Manufacturing Variances	Inhomogeniety	batch-to-batch	lowest
Typical small size	0.01"	0.1"	0.1"
Typical fast thermal time constant	0.01 sec	0.1 sec	0.1 sec
Long Term Stability and Accuracy	OK	good	best
Noise immunity	OK if shielded	best	good
Fragility/Durability	best	OK	good
High thermal gradient environment	best	OK	OK
Typical Temperature Coefficient	Positive	Negative	Positive
Corrosion Resistance	low	good	good
Extension Wires	same alloy	any kind	any kind
Long Wire Runs from Sensor	no	OK	no
Cost	low-medium	low-medium	high