Applied Engineering Statistics

Propagation of Error

Dr. Ahmed Asker

Production and Mechanical Design Engineering
Department
Mansoura University
2021/2022
 Introduction

 Scientists and engineers often perform calculations with

measured quantities, for example, computing the density of an
object by dividing a measurement of its mass by a measurement
of its volume.
 The measured values generally differ somewhat from the true
values that are being measured.
 When a calculation is performed, the errors in the measurements
produce an error in the calculated value (error is propagated ).
 In general, the error in a measurement as being composed of two
parts, the systematic error, or bias, and the random error.
 The bias is the error that is the same for every measurement.
 The random error, varies from measurement to measurement, and
averages out to zero in the long run.

Mansoura University Applied Engineering Statistics Dr. Ahmed Asker 2

 Accuracy and Precision

 Any measurement is the sum of the true value plus contributions

from each of the two components of error:
Measured value = true value + bias + random error
 We are interested in two aspects of the measuring process
(Accuracy and Precision)
 Accuracy: is determined by the bias, which is the difference
between the mean measurement 𝜇 and the true value being
measured.
 The smaller the bias, the more accurate the measuring process.
 If the mean 𝜇 is equal to the true value, so that the bias is 0, the
measuring process is said to be unbiased.

Mansoura University Applied Engineering Statistics Dr. Ahmed Asker 3

 Accuracy and Precision

 Precision: refers to the degree to which repeated measurements

of the same quantity tend to agree with each other.
 If repeated measurements come out nearly the same every time,
the precision is high.
 If they are widely spread out, the precision is low.
 The precision is therefore determined by the standard deviation 𝜎
of the measurement process.
 The smaller the value of 𝜎, the more precise the measuring
process.
 Engineers and scientists often refer to 𝜎 as the random
uncertainty or statistical uncertainty in the measuring process.
 We will refer to 𝜎 simply as the uncertainty.

Mansoura University Applied Engineering Statistics Dr. Ahmed Asker 4

 Accuracy and Precision

(a) Both bias and uncertainty are small. (b) Bias is large; uncertainty is
small. (c) Bias is small; uncertainty is large. (d) Both bias and uncertainty are large.

Mansoura University Applied Engineering Statistics Dr. Ahmed Asker 5

 𝐵𝑖𝑎𝑠 and Uncertainty

 Let 𝑋1 , … , 𝑋𝑛 be independent measurements, all made by the

same process on the same quantity.
 The sample standard deviation 𝑠 can be used to estimate the
uncertainty.
 Estimates of uncertainty are often crude, especially when based
on small samples.
 If the true value is known, the sample mean 𝑋 can be used to
estimate the bias: 𝐵𝑖𝑎𝑠 ≈ 𝑋 − 𝑡𝑟𝑢𝑒 𝑣𝑎𝑙𝑢𝑒.
 If the true value is unknown, the bias cannot be estimated from
repeated measurements.
 In the rest of the lecture, we will assume, unless otherwise stated,
that bias has been reduced to a negligible level.
 We will describe measurements in the form:
𝑀𝑒𝑎𝑠𝑢𝑟𝑒𝑑 𝑣𝑎𝑙𝑢𝑒 ± 𝜎

Mansoura University Applied Engineering Statistics Dr. Ahmed Asker 6

 Linear Combinations of Measurements

 If 𝑋 is a measurement and 𝑐 is a constant, then

𝜎𝑐𝑋 = 𝑐 𝜎𝑋
 If 𝑋1 , … , 𝑋𝑛 are independent measurements and 𝑐1 , … , 𝑐𝑛 are
constants, then
𝜎𝑐1 𝑋1 +⋯+𝐶𝑛 𝑋𝑛 = 𝑐12 𝜎𝑋21 + ⋯ + 𝑐𝑛2 𝜎𝑋2𝑛

Mansoura University Applied Engineering Statistics Dr. Ahmed Asker 7

 Example 1

 The radius of a circle is measured to be 3.0 ± 0.1 cm. Estimate

the circumference and find the uncertainty in the estimate.
Solution
 Let R denote the radius of the circle. The measured value of R is
3.0 cm, and the uncertainty is the standard deviation of this
measurement, which is 𝜎𝑅 = 0.1 cm.
 The circumference is given by 𝐶 = 2𝜋𝑅. The uncertainty in C is
𝜎𝐶 , the standard deviation of C. Since 2𝜋 is a constant, we have
𝜎𝐶 = 2𝜋 𝜎𝑅 = 6.28 0.1 𝑐𝑚 = 0.63 𝑐𝑚
 The circumference is 18.85 ± 0.63 𝑐𝑚.

Mansoura University Applied Engineering Statistics Dr. Ahmed Asker 8

 Example 2

 An item is formed by placing two components end to end.

 The lengths of the components are measured independently, by a
process that yields a random measurement with uncertainty 0.1
cm.
 The length of the item is estimated by adding the two measured
lengths. Assume that the measurements are 4.10 cm and 3.70 cm.
 Estimate the length of the item and find the uncertainty in the
estimate.

Mansoura University Applied Engineering Statistics Dr. Ahmed Asker 9

 Example 3

 A surveyor is measuring the perimeter of a rectangular lot. He

measures two adjacent sides to be 50.11 ± 0.05 m and 75.21 ±
0.08 m. These measurements are independent.
 Estimate the perimeter of the lot and find the uncertainty in the
estimate.

Mansoura University Applied Engineering Statistics Dr. Ahmed Asker 10

 Example 4

 In previous example, the surveyor’s assistant suggests computing

the uncertainty in P by a different method.
 He reasons that since 𝑃 = 𝑋 + 𝑋 + 𝑌 + 𝑌,
 Do you agree with him?

Mansoura University Applied Engineering Statistics Dr. Ahmed Asker 11

 Repeated Measurements

 If 𝑋1 , … , 𝑋𝑛 are n independent measurements, each with mean 𝜇

and uncertainty 𝜎, then the sample mean X is a measurement with
mean
𝜇𝑥 = 𝜇
 and with uncertainty
𝜎
𝜎𝑋 =
𝑛
 The average of several repeated measurements has the same
accuracy as, and is more precise than, any single measurement.

Mansoura University Applied Engineering Statistics Dr. Ahmed Asker 12

 Example 5

 The length of a component is to be measured by a process whose

uncertainty is 0.05 cm.
 If 25 independent measurements are made and the average of
these is used to estimate the length, what will the uncertainty be?
 How much more precise is the average of 25 measurements than
a single measurement?

Mansoura University Applied Engineering Statistics Dr. Ahmed Asker 13

 Uncertainties for Functions of One Measurement

 If X is a measurement whose uncertainty 𝜎𝑋 is small, and if U is a

function of X, then
𝑑𝑈
𝜎𝑈 ≈ 𝜎𝑋
𝑑𝑋
𝑑𝑈
 In practice, we evaluate the derivative at the observed
𝑑𝑋
measurement X.
 The uncertainties computed by previous equation are often only
rough approximations.
 For this reason, these uncertainties should be expressed with no
more than two significant digits.

Mansoura University Applied Engineering Statistics Dr. Ahmed Asker 14

 Example 6

 The radius R of a circle is measured to be 5.00 ± 0.01 cm.

 Estimate the area of the circle and find the uncertainty in this
estimate.

Mansoura University Applied Engineering Statistics Dr. Ahmed Asker 15

 Relative Uncertainties for Functions of One Measurement

 If 𝑈 is a measurement whose true value is 𝜇𝑈 , and whose

𝜎𝑈
uncertainty is 𝜎𝑈 , the relative uncertainty in 𝑈 is the quantity .
𝜇𝑈
 The relative uncertainty is a unitless quantity. It is frequently
expressed as a percent.
 In practice 𝜇𝑈 is unknown, so if the bias is negligible, we
𝜎𝑈
estimate the relative uncertainty with .
𝑈

Mansoura University Applied Engineering Statistics Dr. Ahmed Asker 16

 Example 7

 The radius R of a circle is measured to be 5.00 ± 0.01 cm.

 Estimate the area and find the relative uncertainty in the estimate.
Solution
 The area 𝐴 = 𝜋𝑅2 was computed to be 78.5 ± 0.3 𝑐𝑚2 .
 The absolute uncertainty is therefore 𝜎𝐴 = 0.3 𝑐𝑚2 , and the
relative uncertainty is
𝜎𝐴 Τ𝐴 = 0.3Τ78.5 = 0.004
 We can therefore express the area as 𝐴 = 78.5 𝑐𝑚2 ± 0.4%.

Mansoura University Applied Engineering Statistics Dr. Ahmed Asker 17

 Uncertainties for Functions of Several Measurements

 If 𝑋1 , 𝑋2 , … , 𝑋𝑛 are independent measurements whose

uncertainties 𝜎𝑋1 , 𝜎𝑋2 , … , 𝜎𝑋𝑛 are small, and if 𝑈 =
𝑈(𝑋1 , 𝑋2 , … , 𝑋𝑛 ) is a function of 𝑋1 , 𝑋2 , … , 𝑋𝑛 , then
2 2 2
𝜕𝑈 𝜕𝑈 𝜕𝑈
𝜎𝑈 = 𝜎𝑋21 + 𝜎𝑋22 + ⋯+ 𝜎𝑋2𝑛
𝜕𝑋1 𝜕𝑋2 𝜕𝑋𝑛
 In practice, we evaluate the partial derivatives at the
𝑋1 , 𝑋2 , … , 𝑋𝑛 .

Mansoura University Applied Engineering Statistics Dr. Ahmed Asker 18

 Example 8

 Two resistors with resistances 𝑅1 and 𝑅2 are connected in

parallel. The combined resistance R is given by
𝑅1 𝑅2
𝑅=
𝑅1 + 𝑅2
 If 𝑅1 is measured to be 100 ± 10 Ω, and 𝑅2 is measured to be
20 ± 1Ω,
 Estimate 𝑅 and find the uncertainty in the estimate.
 If we replace the 100 ± 10 Ω resistor can be replaced with a more
expensive 100 ± 1 Ω resistor. How much would this reduce the
uncertainty in the combined resistance? Is it worthwhile to make
the replacement?

Mansoura University Applied Engineering Statistics Dr. Ahmed Asker 19