Vous êtes sur la page 1sur 39

THE UNCERTAINTY OF MEASUREMENTS

TRUTH IN MEASUREMENT

THE UNCERTAINTY OF MEASUREMENTS

q The concept of uncertainty of measurement is not new

q It has been used in calibration laboratories long before the ISO GUM

q It has been normal practice whenever a value is reported an uncertainty is quot

q What is new is the formalising of the process with the ISO GUM and

q the requirements in ISO 17025

THE UNCERTAINTY OF MEASUREMENTS


AS ISO/IEC 17025

4.6.1 A calibration laboratory, or a testing laboratory, performing its own calibrations hall have and shall apply a procedure to estimate the uncertainty of measurement for a alibrations and types of calibrations

(Ex 1)

THE UNCERTAINTY OF MEASUREMENTS

I USED TO BE UNCERTAIN BUT NOW IM NOT SURE

THE UNCERTAINTY OF MEASUREMENTS

q What is uncertainty? q Does it mean we dont really know? q Does it mean we should not report our result?

THE UNCERTAINTY OF MEASUREMENTS

NO!
Most definitely not

THE UNCERTAINTY OF MEASUREMENTS

q Uncertainty is a component of our measurement. q It is a component that can be analysed and quantified. q When quantified it provides confidence in our result.

THE UNCERTAINTY OF MEASUREMENTS

HOW DO WE DETERMINE OUR UNCERTAINTY?

THE UNCERTAINTY OF MEASUREMENTS

The easiest way is by creating an

UNCERTAINTY BUDGET

q What is an uncertainty Budget?

THE UNCERTAINTY OF MEASUREMENTS

q An uncertainty budget is a convenient means of itemising, tabulating and calculating details of uncertainty. q How does it work?

THE UNCERTAINTY OF MEASUREMENTS

q Uncertainty analysis is simply a means of focussing attention on the individual components that may affect the final result. q The uncertainty budget is a means of capturing this information in logical steps. q Such as ..

THE UNCERTAINTY OF MEASUREMENTS


q Step 1: list the components that may affect our result. q Step 2: determine the type of distribution. q Step 3: determine the value of the semi range to be used. q Step 4: decide on our level of confidence in the value we are using.
(Ex 2)

THE UNCERTAINTY OF MEASUREMENTS


TYPICAL SOURCE COMPONENTS
0-25mm External Micrometer Source Ref. gauge block tol. Ref. gauge block uncert. Anvil geometry Therm effects Resolution/parallax (2), (3) Repeat./random effects(2), (3) Units m m m C m m

THE UNCERTAINTY OF MEASUREMENTS


DISTRIBUTION TYPES q A Normal Distribution - may be either Type A or a Type B uncertainty. A normal distribution may represent a random series of readings where the majority of the readings occur near the mean value. If these readings are sufficient in number and graphed, a bell shaped curve will result.

THE UNCERTAINTY OF MEASUREMENTS

q A Rectangular Distribution - is one in which the actual value may occur anywhere within the distribution with equal probability. It is important to define the limits of the distribution with a high degree of confidence.

THE UNCERTAINTY OF MEASUREMENTS

q A Triangular Distribution - is similar to the rectangular distribution with the difference being that there is a lower probability that the actual value will be at the limits of the range.

THE UNCERTAINTY OF MEASUREMENTS


DISTRIBUTION EXAMPLES
Source Ref. gauge block tol. Ref. gauge block uncert. Anvil geometry Therm effects Resolution/parallax (2), (3) Repeat./random effects(2), (3) Units m m m C m m Dist. Rect B Norm B Rect B Rect B Rect B Norm A

Ex 3)

THE UNCERTAINTY OF MEASUREMENTS


UNCERTAINTY TYPES q A Type A uncertainty is one which is evaluated by statistical means. This is generally only practical and meaningful with a reasonable number of repeated readings. q A Type B uncertainty is one which is evaluated by other than statistical methods.

THE UNCERTAINTY OF MEASUREMENTS


DETERMINING THE SEMI RANGE AND STANDARD UNCERTAINTY q Normal A Distribution - The scatter of a number (n) of repeated measurements will be found to have a normal distribution. We use the population standard deviation (s) of these readings as the semi-range and divide by the square root of the number of readings to obtain the ESDM and standard uncertainty.

THE UNCERTAINTY OF MEASUREMENTS

q If only a small number of readings are taken during the calibration then a pre-characterisation as shown in the ISO GUM in example H.1.3.2 may be appropriate. q For example..

THE UNCERTAINTY OF MEASUREMENTS

q The pooled experimental standard deviation characterising a comparison was determined from th variability of 25 (n1) independent repeated observations and was found to be 13m (s). In comparison with this example 5 (n2) repeated observations were taken.

q The standard uncertainty (u) is then..

THE UNCERTAINTY OF MEASUREMENTS

s ui = n2

or

13m ui = 5

= 5.8m

and the degrees of freedom are based on the number of repeated observations.

Vi = n1 1

i.e. 25-1 = 24

THE UNCERTAINTY OF MEASUREMENTS

q Normal B Distribution - The expanded uncertainty given on a calibration certificate is divided by the coverage factor (k) to obtain the standard uncertainty.

U 95 Uc = k

THE UNCERTAINTY OF MEASUREMENTS

q If a 95% confidence level is quoted on the calibration certificate without a coverage factor then it may be assumed that the divisor is 1.96 or 2 standard deviations . i.e. k = 2 (rounded to 2)

THE UNCERTAINTY OF MEASUREMENTS

q If a standard deviation is quoted; for example an uncertainty of 5m at the 3 standard deviation level, the standard uncertainty is obtained by dividing the quoted uncertainty by 3.

0.005 Uc = 3

= 1.7m

THE UNCERTAINTY OF MEASUREMENTS

q Rectangular B Distribution - The only information that we know about the distribution in this case will be the limits. These limits need to be selected critically with a high degree of confidence.

THE UNCERTAINTY OF MEASUREMENTS

q The ISO GUM advises that: There is no substitute for critical thinking, intellectual honesty, and professional skill

THE UNCERTAINTY OF MEASUREMENTS

q For the calculation of the standard uncertainty u for a rectangular distribution the formula is: a u= 3 q Where: u is the standard uncertainty. a is the semi range of the limits of the uncertainty component.

THE UNCERTAINTY OF MEASUREMENTS


STANDARD UNCERTAINTY Revision

Normal Distribution:

s ui = n a ui = 3 a ui = 6

(ESDM)

Rectangular Distribution:

Triangular Distribution:

Ex 4)

THE UNCERTAINTY OF MEASUREMENTS


DEGREES OF FREEDOM AND CONFIDENCE q Type A Uncertainties - The number of degrees of freedom for each Type A uncertainty is generally one less than the number of readings. Vi = n-1

THE UNCERTAINTY OF MEASUREMENTS

q RectangularType B Uncertainties - The number of degrees of freedom for each Type B uncertainty with a Rectangular Distribution is determined from the confidence in the limits. For example if the relative confidence level is 90% (a one in ten chance that the true value is outside the limits selected) then the degrees of freedom is 50.

THE UNCERTAINTY OF MEASUREMENTS

q The determination of the number of degrees of freedom for each Rectangular Type B uncertainty can be calculated simply by:

(10) 2 Vi = 2
= 50

THE UNCERTAINTY OF MEASUREMENTS

q Normal Type B Uncertainties - typically a calibration report, the confidence level and the k value may be provided. Referring to the Students t Distribution tables, the approximate degrees of freedom can be determined. If a k value is not provided then an infinite number of degrees of freedom may be assumed.

THE UNCERTAINTY OF MEASUREMENTS

Source Ref. gauge block tol. Ref. gauge block uncert. Anvil geometry Therm effects Resolution/parallax (2), (3) Repeat./random effects(2), (3)

Units m m m C m m

Dist. Rect B Norm B Rect B Rect B Rect B Norm A

Value U or a 0.30 0.24 0.25 2.00 1.00 0.12

Divisor Confidence % 1.7321 95 2.0000 95 1.7321 95 1.7321 80 1.7321 95 1.7321 95

Vi 200 30 200 12 200 9

(Ex 5)

THE UNCERTAINTY OF MEASUREMENTS


SENSITIVITY COEFFICIENTS q The sensitivity coefficient (c) describes how the output estimate varies with changes to the value of the input estimates. For example converting temperature in C to a common unit with other components of the budget.

THE UNCERTAINTY OF MEASUREMENTS


EXPANDED UNCERTAINTY q The expanded uncertainty (U)is the final estimate for the uncertainty

THE UNCERTAINTY OF MEASUREMENTS


CONVERSION BETWEEN CONFIDENCE LEVELS q For a normally distributed confidence level of 99% dividing the expanded uncertainty by 2.6 (k = an infinite number of degrees of freedom) will provide an approximation of the combined standard uncertainty. q Spreadsheet..

THE UNCERTAINTY OF MEASUREMENTS


UNCERTAINTY BUDGET 0 to 25 mm external micrometer Date:

: al Number:

Maximum length (mm) = Coeffecient of expansion (ppm / C) =

25 11.5

Source Units gauge block tol. m gauge block uncert. m l geometry m m effects C (2), (3) olution/parallax m (2), (3) eat./random effects m

Dist. Rect B Norm B Rect B Rect B Rect B Norm A Sums

Value U or a 0.30 0.24 0.25 2.00 1.00 0.12

Divisor 1.7321 2.0 1.7321 1.7321 1.7321 1.7321

Confidence % 95 95 95 80 95 95

Vi 200 30 200 12 200 9

ui 0.1732 0.1200 0.1443 1.1547 0.5774 0.0693

ci 1 1 1 0.2875 1 1

ui ci 0.1732 0.1200 0.1443 0.3320 0.5774 0.0693

(u i c i )^2 0.0300 0.0144 0.0208 0.1102 0.3333 0.0048 0.5136 0.7166 166.5300 1.9744

[(u i c i )^4 4.5000E 6.9120E 2.1701E 1.0122E 5.5556E 2.5600E 1.5839E

Combined Standard Uncertainty, U c Effective Degrees of Freedom, V eff Coverage factor, k = Student's t for V eff and CL 95% Expanded Uncertainty, U=ku c +/-

1.4149

THE UNCERTAINTY OF MEASUREMENTS


EXTRACT FROM THE GUM

4.8. Although this Guide provides a framework for assessing uncertainty, it cannot ubstitute for critical thinking, intellectual honesty, and professional skill.

he evaluation of uncertainty is neither a routine task nor a purely mathematical one; it epends on detailed knowledge of the nature of the measureand and of the measuremen

he quality and utility of the uncertainty quoted for the result of a measurement therefo timately depend on the understanding, critical analysis, and integrity of those who ontribute to the assignment of its value.

-----oooOooo-----

Vous aimerez peut-être aussi