Professional Documents
Culture Documents
Uncertainty PDF
Uncertainty PDF
Uncertainty Analysis
Measurement Error
True data
x'
Measured value, x Bias error
x Precision
error in xi
Measurement number
Design-state uncertainty
ud = u02 + uc2
u0 = ±1 / 2 resolution (95%)
yo
Instrument Uncertainty, uc
This information is available from the manufacturer’s
catalog
x
resolution uncertainty
1/2 resolution
Design-Stage Uncertainty Analysis
u0 uc
½ Resolution = 0.125 N el2 + er2 = ± 0.2 2 + 0.32 = ±0.36 N
Design-Stage Uncertainty Analysis
Example: A voltmeter is to be used to measure the output from a pressure transducer that outputs
an electrical signal. The nominal pressure expected will be ~3 psi (3 lb/in2). Estimate the design-
state uncertainty in this combination. The following information is available:
Voltmeter
Resolution: 10 µV
Accuracy: within 0.001% of reading
Transducer
Range: ±5 psi
Sensitivity: 1 V/psi
Input power: 10 Vdc ± 1%
Output: ±5 V
Linearity: within 2.5 mV/psi over range
Repeatability: within 2 mV/psi over range
Resolution: negligible
Known: Instrument specifications
Assume: Values representation of instrument 95% probability
Solution:
Design-Stage Uncertainty Analysis
Design-state uncertainty
ud = (ud )2E + (ud )2P
(ud )E = (u )2
0 E + (u )2
c E
(ud )P = (u0 )2P + (uc )2P
Error Propagation
Computation of the overall uncertainty for a measurement system consisting
of a chain of components or several instruments
Let R is a known function of the n independent variables xi1, xi2 , xi3, …, xiL
R = f ( x1 , x2 , K , xL )
L is the number of independent variables. Each variable contains some
uncertainty (ux1, ux2, ux3,…, uxL) that will affect the result R.
Application of Taylor’s expansion gives, (neglect the higher order term)
R ± ∆R = f ( x1 ± u x1 , x2 ± u x 2 ,..., xL ± u xL ) ≈ f ( x1 , x2 ,..., xL ) +
∂f ∂f ∂f
u x1 + u x 2 + ... + u xL
∂x1 ∂x2 ∂xL
Where R = f ( x1 , x2 ,..., xL )
Error Propagation
The combination of uncertainty of all variables (probable estimate of uR)
2 2 2
∂f ∂f ∂f
u R = ± u x1 + u x 2 + K + u xL
∂x1 ∂x2 ∂xL
L
=± (
∑ i xi
θ u )2
( P %)
i =1
∂f
θi =
∂xi
Error Propagation
Example: For a displacement transducer having a calibration curve y = KE, estimate the
uncertainty in displacement y for E = 5.00 V, if K = 10.10 mm/V with uk = ±0.10 mm/V and uE =
±0.01 V at 95% confidence
Known: y = KE
E = 5.00 V uE = 0.01 V
K = 10.10 mm/V uk = 0.10 mm/V
Solution: Find uy y ' = y ± u y = KE ± u y
u y = ± (θ E u E ) + (θ K u K )
2 2
∂y ∂y
θE = =K θK = =E
∂E ∂K
uE = 0.01 V uK = 0.10 mm/V
uy = ± (Ku E )2 + (Eu K )2
= ± (10.10 mm/V × 0.01 V ) + (5 V × 0.10 mm/V ) = ±0.51 mm
2 2
Sequential Perturbation
A numerical approach can also be used to estimated the propagation of
uncertainty. This refers to as sequential perturbation. This method is
straightforward and uses the finite difference to approximate the
derivatives (sensitivity index)
1) Calculate the average result from the independent variables
R = f ( x1 , x2 ,..., xL )
2) Increase the independent variables by their respect uncertainties
and recalculate the result based on each of these new values. Call
these values Ri+
R1+ = f ( x1 + u1 , x2 ,..., xL ),
R2+ = f ( x1 , x2 + u 2 ,..., xL )
RL+ = f ( x1 , x2 ,..., xL + u L )
3) Decrease the independent variables by their respect uncertainties
and recalculate the result based on each of these new values. Call
these values Ri−
Sequential Perturbation
R1− = f ( x1 − u1 , x2 ,..., xL ),
R2− = f ( x1 , x2 − u2 ,..., xL )
RL− = f ( x1 , x2 ,..., xL − u L )
δRi+ + δRi−
δRi = ≈ θ i ui
2
The uncertainty in the result
1/ 2
L
2
u R = ± ∑ (δRi )
i =1
Error Propagation
Example: For a displacement transducer having a calibration curve y = KE, estimate the
uncertainty in displacement y for E = 5.00 V, if K = 10.10 mm/V with uk = ±0.10 mm/V and uE =
±0.01 V at 95% confidence
Known: y = KE
E = 5.00 V uE = 0.01 V
K = 10.10 mm/V uk = 0.10 mm/V
Solution: Find uy
y ' = y ± u y = KE ± u y
u y = ± (δRE ) + (δRK )
2 2
eij
[
Pi = Pi12 + Pi 22 + ... + Pik2 ]
1/ 2
i = 1, 2, 3
[
B = B12 + B22 + B32 ]
1/ 2
Multiple-Measurement Uncertainty Analysis
The measurement uncertainty in x, ux
u x = B 2 + (tv ,95 P )
2
(95%)
2
3 K 2
∑∑ Pij
i =1 j =1
v= 3 K
∑∑ Pij4 / vij ( )
i =1 j =1
Multiple-Measurement Uncertainty Analysis
Measurement uncertainty, ux
[
u x = B 2 + (tv ,95 P ) ]
2 1/ 2
(95%)
[
P = P12 + P22 + P 3 ]
2 1/ 2
[
B = B12 + B22 + B32 ]1/ 2
[
Pi = Pi12 + Pi 22 + ... + P ]
2 1/ 2
ik [
Bi = Bi21 + Bi22 + ... + Bik2 ]
1/ 2
eij=Pij+Bij
Measurand, x
Multiple-Measurement Uncertainty Analysis
Example: After an experiment to measure stress in a load beam, an uncertainty analysis reveals
the following source errors in stress measurement whose magnitude were computed from
elemental errors
B1 = 1.0 N/cm2 B2 = 2.1 N/cm2 B3 = 0 N/cm2
P1 = 4.6 N/cm2 P2 = 10.3 N/cm2 P3 = 1.2 N/cm2
v1 = 14 v2 = 37 v3 = 8
If the mean value of the stress in the measurement is 223.4 N/cm2, determine the best estimate of
the stress
Known: Experimental error source indices
Assume: All elemental error have been included
Solution: Find uσ
Measurement uncertainty, ux
[
u x = B 2 + (tv ,95 P ) ]
2 1/ 2
(95%)
Consider the result, R which is determined from the function of the n independent
variables xi1, xi2 , xi3, …, xiL
R' = R ± u R ( P%)
L
where
∑ [θi Pxi ]2
L
PR = ±
i =1
BR = ± ∑ [θ B
i =1
i xi ] 2
Where psfa refers to lb/ft2 absolute. Determine a best estimate of the density. The gas constant is
R = 54.7 ft lb/lbm oR
Known: p , S p , T , ST
ρ = P / RT R = 54.7 ft lb/lbm o R
Assume: Gas behaves as an ideal gas
Solution: Find ρ ' = ρ + uρ
Propagation Uncertainty Analysis to a result
[
u ρ = B 2 + (tv ,95 P ) ]
2 1/ 2
(95%) where v =
[(θ P ) + (θ P ) ]
p p
2
T T
2 2
(θ P )
p p
4
/ v p + (θ T PT ) / vT
4
B = ± (θ p B p ) + (θT BT ) P = ± (θ p Pp ) + (θT PT )
2 2 2 2
∂ρ 1 ∂ρ p
θp = = θT = =−
∂p RT ∂T RT 2