Understanding and Expressing Measurement Uncertainties

associated with Thermodynamic Metrology

(A primer)

Edward Brown
Quality Systems Lab
Boca Raton, FL 33487

Introduction maintain a variety of standards for realization

In Thermodynamic Metrology all too often of temperature depending upon the complexity
personnel are assigned to perform tasks with or level of calibration being performed. Dry
little or no experience. The importance of blocks, liquid baths and fluidized sand baths
sound temperature metrology was of little are all used in the comparison method of
interest or was not given the same respect as calibration, where a calibrated SPRT is
DC Low Frequency or RF microwave. Over compared to an unknown sensor. This method
the past 20 years temperature realization has is typically used by secondary laboratories
become more important due to the effect it has calibrating commercial temperature elements
on other metrology disciplines. There are and is the most common type of
many broad and some minute aspects to this thermodynamic calibration.
discipline. This paper is intended as a primer
to assist the new technician in understanding In the commercial laboratory, a less common
thermodynamic metrology, determination of method is the use of ITS-90 (International
measurement uncertainty and expression of Temperature Scale – 1990) fixed point
uncertainty SI units. apparatus. These devices are based on fixed,
known temperature points which are derived
Measurement of Temperature from physical constants of nature. Properly
A variety of instruments can be used to designed fixed point maintenance apparatus
measure temperature; LIG (liquid in glass) and the associated fixed point cell will
thermometers, RTD’s (resistive temperature maintain a known temperature over many
detectors), thermocouples, SPRT’s (standard hours and in some cases days. The most
platinum resistance thermometers) and a few common fixed point is the triple point of water
others. Most of these with the exception of (TPW), which, when properly prepared can
LIG’s need an external device to read the maintain 0.01°C with better than .0001°C of
signal from the sensing element. uncertainty.

Sources of uncertainty in this type of There are many sources of uncertainty in a heat
arrangement are the sensor, the indicating source; axial uniformity, radial uniformity,
device, and in many cases, the connection loading effect, temperature stability, stem
method. conduction error, reference probe, and many
others. We will discuss each of these briefly
Heat Source and how to arrive at a combined measurement
All heat sources introduce measurement errors uncertainty and an expanded uncertainty (k=2).
as a result of their thermodynamic properties.
This measurement error should be quantified to
determine the heat sources contribution to the
measurement uncertainty. All laboratories
providing temperature calibration services

Page 1 of 4
 Loading effect
The number of probes impacts the amount of
heat drawn away from or into a bath or well.
This effect can be minimized by allowing
sufficient time for equalization and saturation
of the work load before taking measurements.

Other Considerations
Drift
Control sensor and reference sensor drift will
vary depending on the care and frequency of
Figure 1. Diagram of comparison calibration
use of any bath or well. This drift can be
determined with regular calibration and
Uncertainties associated with comparison intermediate checks.
calibrations
The major sources of uncertainty are Control Sensor
uniformity, stem conduction, loading, bath When using a bath or well without an external
instability, reference thermometer accuracy, reference the control sensor becomes another
and unit under test accuracy. source of uncertainty. Minimize your
uncertainty by using a reference sensor
Uniformity whenever possible.
The vertical gradient in a bath or dry well is
termed “axial uniformity”. The surface of a Hysteresis
bath or dry well is exposed to ambient Hysteresis is the difference in indicated
environment and to a controlled temperature temperature dependant on the direction the
along a portion of the vertical length. This bath or well approaches set point.
axial uniformity can be minimized be aligning
the reference probe and UUT sensing centers. Uncertainty Budget
Any uncertainty budget should be an aggregate
The horizontal gradient in a bath or dry well is of the most important contributors from the
termed “radial uniformity”. The bath or dry previously mentioned sources of error. The
well construction as well as the distance most common method is to use the “Type B”
between sensors and the heating source can all approach. In this approach, the uncertainties
lead to radial measurement uncertainty. published in the instrument manuals are used.
Minimize this type of uncertainty by using a Some of the values needed for an uncertainty
reference sensor of the same diameter as the budget can be found on the calibration report
UUT and placing the sensors as close together for the instruments used. If uncertainties are
as possible. required at points between those expressed on
the calibration report, then rigorous calculating
Stem conduction error methods are needed to propagate uncertainties.
Stem conduction is the parasitic loss of heat The calibration lab can use the law of
along the length of the thermometer stem. This Propagation of Uncertainties as described in
affects both the reference and UUT. the GUM to achieve these uncertainties.
Following the recommended depth for However, a less complex method for
minimum thermometer insertion (20 X probe establishing the uncertainties between
OD) will minimize stem conduction error calibration points would be a technique called
resulting in very little uncertainty.

Page 2 of 4
Linear Interpolation. In this technique we standard uncertainties are then combined
assume that the uncertainty changes linearly through a process called Root Sum Square or
between the points of interest. RSS. This means that each of the standard
uncertainties is squared before adding all of the
Combining uncertainties squared components together. The square root
We must first convert to standard uncertainty of the result is taken as the total combined
in order to evaluate the total uncertainty. At the standard uncertainty.
same time we must ensure that all values are
expressed in the same unit of measure, see “SI
Units” below. U= S12+ S22+ S32...
There are two types of uncertainties; The expanded uncertainty is obtained by
uncertainties based on known or assumed multiplying the resulting value U by a factor of
probability distribution and those based on two (k=2). This expanded uncertainty is
limits of error. When uncertainties are stated shown in the uncertainty statement of scope of
with a known distribution with a coverage accreditation.
factor of k=2, it means a normal, or Gaussian,
distribution has been assumed. To convert this
type of uncertainty to a standard uncertainty
we simply need to divide by two. For other
uncertainties types such as limits of error that Example
were assigned without probability Illustrated below is a theoretical uncertainty
distributions, a rectangular distribution is table for a metrology bath with a standard
assumed. To convert a rectangular distribution thermometer as the readout. This example is at
to a standard uncertainty, the value is divided 0°C and does not take the UUT into
by the square root of three. The individual consideration. Please remember this is
theroretical, the math is sound the uncertainties
are without basis.
Table 1

Uncertainty Specification (°C) Probability Uncertainty (°C)

Contributors @ 0.0°C Distribution
Stability .002 Rectangular .0011
Loading effect .001 Rectangular .0005
Radial uniformity .002 Rectangular .0011
Axial uniformity .004 Rectangular .0023
Reference probe .004 Rectangular .0023
calibration
Reference probe drift, .002 Rectangular .0011
including hysteresis
Reference probe stem .001 Rectangular .0005
conduction
Thermometer readout .001 Rectangular .0005
accuracy
Combined Standard Uncertainty (RSS method) .003868
Expanded Uncertainty (k=2) .007736

Page 3 of 4
Rounding
Rounding should only be performed once all
calculations have been performed. As seen in
Table 1 the expanded uncertainty is reflected to
6 places after the decimal. This number can
now be rounded to a more practical value
before being expressed on a scope of
accreditation or customer certificate of
calibration. The final expanded uncertainty
should reflect the resolution of the laboratories
indicating device. In the example above the
rounded expression would be, ±. 0.008°C for a
three decimal place digital display.

SI Units
When starting the process of determining
uncertainty, always convert non-standard units
to SI units. The Kelvin and degree Celsius are
the accepted units for expressing uncertainty.
This conversion should be performed before
any calculations are done. It should be noted
that Kelvin is not expressed as °K, the Kelvin
is an absolute value of temperature and should
always be expressed as mK or K.
Note: 0°C = 273.16K.

Summary
There are many factors to take into
consideration when expressing total and
expanded uncertainties in thermodynamic
metrology. This paper is intended to give the
new technician some insight into the
complexity of thermodynamic metrology. This
paper does not, by any means, cover the
diversity and complexity of Thermodynamic
Calibrations in today’s metrology laboratory.
However, more extensive information can be
gleaned from websites such as NCSLI.org or
NIST.gov.

Page 4 of 4