Error Analysis (Uncertainty Analysis) : 16.621 Experimental Projects Lab I
Error Analysis (Uncertainty Analysis) : 16.621 Experimental Projects Lab I
(UNCERTAINTY ANALYSIS)
1
TOPICS TO BE COVERED
4
USES OF UNCERTAINTY ANALYSIS (II)
5
[Adapted from Kline, S. J., 1985, “The Purposes of Uncertainty
Analysis”, ASME J. Fluids Engineering, pp. 153-160]
UNCERTAINTY ESTIMATES AND HYPOTHESIS ASSESSMENT
600
500
400
Distance [cm]
300
200
100
0
0 20 40 60 80 100 120
Mass [g]
600 600
500 500
400 400
Distance [cm]
Distance [cm]
300 300
200 200
100 100
0 0
0 20 40 60 80 100 120 0 20 40 60 80 100 120
Mass [g] Mass [g]
6
HOW DO WE DEAL WITH NOT KNOWING
THE TRUE VALUE?
7
AN IMPLICATION OF NOT KNOWING THE TRUE VALUE
• How about if we don’t know the true value? Can we, by looking at
the data in the slide after this, say that there are bias errors?
8
A TEAM EXERCISE
• List the variables you need to determine in order to carry out your
hypothesis assessment
• Can you make an estimate of the level of the uncertainty in the final
result?
– If so, what is it?
– If not, what additional information do you need to do this? 11
HOW DO WE COMBINE ERRORS?
12
ERRORS IN THE FINAL QUANTITY
• Z=X+Y
• Linear combination
– Z + dz = X + dx + Y + dy
– Error in Z is dz = dx + dy BUT this is worst case
• For random errors we could have
– dz = dx − dy
or dy − dx
– These errors are much smaller
• In general if different errors are not correlated, are
independent, the way to combine them is
dz = dx2 + dy2
13
• This is true for random and bias errors
THE CASE OF Z = X - Y
dz = dx2 + dy2
dx dy
• Say = =ε (say 2%)
X Y
dz 2 dx dx
• = may be much larger than
Z X−Y X
14
ESTIMATES FOR THE TRUE VALUE AND THE ERROR
15
SOME “RULES” FOR ESTIMATING
RANDOM ERRORS AND TRUE VALUE
1 Number of measurements
18
A REVELATION
20
ONE ADDITIONAL ASPECT OF COMBINING ERRORS
21
REPORTING OF MEASUREMENTS
22
[from Taylor, J., An Introduction to Error Analysis]
COMMENTS ON REJECTION OF DATA
• You may not be controlling all the variables that you need to
26
SUMMARY
• Both the number and the fidelity of the number are important in a
measurement
• We considered two types of uncertainties, bias (or systematic
errors) and random errors
• Uncertainty analysis addresses fidelity and is used in different
phases of an experiment, from initial planning to final reporting
– Attention is needed to ensure uncertainties do not invalidate
your efforts
• In propagating uncorrelated errors from individual measurement to
final result, use the square root of the sums of the squares of the
errors
– There are generally only a few main contributors (sometimes
one) to the overall uncertainty which need to be addressed
• Uncertainty analysis is a critical part of “real world” engineering
projects
27
SOME REFERENCES I HAVE FOUND USEFUL
• Baird, D. C., 1962, Experimentation: An Introduction to Measurement
Theory and Experiment, Prentice-Hall, Englewood Cliffs, NJ
29
THRUST STANDS
35