Page 1 Lecture 3-4 – MEEN 260 Dr. R.

Tafreshi

Lecture 3-4: Sensor Terminology

Introduction to Statistical Data Analysis:
Probability Density Functions (PDF)
Normal Distribution

Lecture Outline:
• Population vs. Sample
• Probability Density Functions (PDF)
• Properties of a PDF
• “Normalizing” the Normal Distribution
• Examples

Reading Assignments:
• Chapter 3 of textbook by Beckwith

Next Lecture: Statistical Data Analysis (continued)

Page 2 Lecture 3-4 – MEEN 260 Dr. R. Tafreshi

Sensor Terminology
• Since sensors generally output electrical signals, we characterize sensors with some of
the same terms we use for electronic devices
• Unfortunately there is no standard

• Resolution
o Minimum detectable fluctuation in signal
§ D/A conversion can impact this greatly

• Accuracy
o Largest expected error between actual and recorded signal

• Span/Range
o Range of detectable characteristics
§ E.g. 0-5000 psi for pressure gage

• Sensitivity
o Ratio between a small change in electrical signal to a small change in
physical signal.
DOutput
§ High sensitivity is usually good h=
DInput
Page 3 Lecture 3-4 – MEEN 260 Dr. R. Tafreshi

Sensor Terminology…

• Bandwidth
o Frequency range of detectable signals
§ e.g. a normal pressure transducer (sensor) is too slow to detect
pressure fluctuations within an engine cylinder.
A suitable pressure sensor needs to be fast enough to detect
in-cylinder pressure fluctuations.
ü Possible option for diagnosing engine knock

• Hysteresis
o Occurs when the difference in output depends on the direction of signal
change:
o Exists when dependence of the output of a system is not only on its
current input, but also on its history of past inputs
§ Generally, not a desirable quality

http://www.nptel.ac.in/courses/105108072/mod02/lec2.html

• Linearity
o How much does the sensor deviate from a linear representation
§ Non-contact position sensor (+/- 10 cm linear in useful range)

From: Alleyne, A., B. Williams, and M. DePoorter, “A Lateral Position

Sensing System for Automated Vehicle Following,” ASME/IEEE
Transactions on Mechatronics, 3:3, 218-224, 1998.
Page 4 Lecture 3-4 – MEEN 260 Dr. R. Tafreshi

Statistical Data Analysis/Processing

Probability Density Functions (PDF) – Gaussian/Normal Distribution

Question:
What percentage of measurements will fall between ±1.5σ or in general, between arbitrary
x1 and x2?
Answer:
We need to be able to compute the area under the PDF between x1 and x2.
2
1 æ x-µ ö
1 - ç
s ÷ø
p( x) = e 2è
s 2p

….

….

- Instead of calculating the integral we use Table 3.2 of the textbook, by defining an
auxiliary variable z as described next, a process known as “Normalizing” the
Normal Distribution.
 x1 x2 x

In other words
x2 1 3-4 – MEEN x – 260 2
Page 5 P x1 x x2 = Lecture
-------------- exp – ------------------- dx (2) Dr. R. Tafreshi
x1 2 2 2
“Normalizing” the Normal Distribution
R. Langari, 8/23/11 7 of 15

DEPARTMENT OF MECHANICAL ENGINEERING

• Given a random variable x that is normally distributed with mean µ and std. dev.
TEXAS A&M UNIVERSITY
COLLEGE STATION, TX 77843-3123
979 845 1251

of s FAX 979 845 3081

o Denoted by N ( µ , s 2 )
T E X A S A & M U N I V E R S I T Y

•
Then the random variable z is normally distributed with mean 0 and std. dev. of 1.
o Denoted by N (0,1)
Statistical Data Processing
x-µ
z = an auxiliary variable z as
This is done using Table 3.2 of the text, by defining
follows:
s
p x
1
--------------
2 x–
z = ------------

x1 – x2 –
x1 x2 z 1 = -------------- z 2 = --------------
x
x– z2 1
z = ------------ 2
p z P z1 z z2 = ---------- e – z 2 dz
z1 2
1
----------
2

z1 0 z2 z
2
1 æ x-µ ö
1 - ç
s ÷ø
p( x) = e 2è
R. Langari, 8/23/11
s 2p
8 of 15

x ( x−µ )2
1 −
2σ 2
DEPARTMENT OF MECHANICAL ENGINEERING

∫ e dx = TEXAS A&M UNIVERSITY

COLLEGE STATION, TX 77843-3123
979 845 1251

0 σ 2π FAX 979 845 3081

MEEN 260 Notes Page 23 of 134 R. Langari, Fall 2011

Page 6 Lecture 3-4 – MEEN 260 Dr. R. Tafreshi

Example:
2
1 # x−µ &
µ +σ
1 − % ( 1
1 x-µ
z=
2
2$ σ '
P(µ − σ ≤ x ≤ µ + σ ) = ∫ e dx = ∫ e − z /2 dz
µ −σ σ 2π −1 2π s

Table 3.2 does not directly give you the value of the above “z” integral, however, noting
that the integrand is symmetric we have:
1
1 - z 2 /2 æ 1 1 - z 2 /2 ö
ò
-1 2p
e dz = 2 çò
è 0 2p
e dz ÷ = 2(0.3413) = 0.6826
ø

Table 3.2 of textbook (page 50)

Page 7 Lecture 3-4 – MEEN 260 Dr. R. Tafreshi

Example – Suppose that for a large set of data µ = 12.1 and s = 0.5 , find the probability
that a given measurement of x falls between 12.5 and 13 assuming that the data is normally
distributed.

Solution:

P(12.5 £ x £ 13) =

First find the equivalent upper and lower bounds of the integral in terms of:

ì
x-µ ï
z= =í
s ï
î

This means we need to find

1.8
1 - z2 / 2
ò0.8 2p e dz
which is not directly given in Table 3.2. We can however, write this integral as two parts:
1.8 1.8 0.8
1 - z 2 /2 1 - z 2 /2 1 - z 2 /2
ò0.8 2p e dz = ò0 2p e dz - ò0 2p e dz
Both terms on the right hand side can be directly read off Table 3.2 and we thus have
1.8
1 - z 2 /2
ò0.8 2p e dz = = 0.176

In other words, 17.6% of all measurements will fall between 12.5 and 13. Likewise, we can
say that there is 17.6% probability that a given arbitrary measurement of x will be between
12.5 and 13.
Page 8 Lecture 3-4 – MEEN 260 Dr. R. Tafreshi

Example – Suppose that for a large set of data µ = 10 , s = 0.8 . Determine the probability
that a given arbitrary measurement of x will fall between 9.5 and 10.3.

Solution:
First we find the upper and lower bounds for the ‘z’ integral:

ì10.3 - 10
= 0.375
x - µ ïï 0.8
z= =í
s ï 9.5 - 10 = -0.625
ïî 0.8

Then:
2
10.3 1 æ x-µ ö 0.375
1 - ç
s ÷ø 1 - z 2 /2
P(9.5 £ x £ 10.3) = ò e 2è dx = ò e dz
9.5 s 2p -0.625 2p

Note that the lower bound for the ‘z’ integral is negative (see the rightmost integral.) We
can still use Table 3.2 to compute the above as follows:
0.375 0 0.375
1 - z2 / 2 1 - z2 / 2 1 - z2 / 2
ò 2p
-0.625
e dz = ò 2p
-0.625
e dz + ò0 2p e dz
Since Table 3.2 does not list the value of the ‘z’ integral for negative bounds, and knowing
the fact that the PDF is symmetric we simply re-write the above as follows:
0.375 0.625 0.375
1 - z2 / 2 1 - z2 / 2 1 - z2 / 2
ò
-0.625 2p
e dz = ò
0 2p
e dz + ò
0 2p
e dz

There is another difficulty that we must deal with. The table only lists the values of the ‘z’
for ‘z’ values of up to two decimal digits. We need to therefore, either round the values of
0.375 and 0.625 or else interpolate. To simplify things, let us simply round these values to
0.38 and 0.62. The corresponding values of the ‘z’ integral can be read off Table 3.2 as
0.1480 and 0.2324. We therefore have
0.375
1 - z 2 /2
ò
-0.625 2p
e dz = 0.2324 + 0.1480 = 0.3804

In other words, 38.04% of all measurements fall between 9.5 and 10.3
Page 9 Lecture 3-4 – MEEN 260 Dr. R. Tafreshi

Example Problem (Prob. 3.1 of textbook)

• Assume that we know that 𝜇𝑥 = 200, 𝜎𝑥 = 20

• How can we find the range within which 60% of the data will fall, assuming a normal
distribution

Solution:
§ Step 1: find z values
𝐹𝑟𝑜𝑚 𝑡𝑎𝑏𝑙𝑒
𝑧
60% data ∫0 𝑝 (𝑧) = 0.30 𝑧 = 0.8418

§ Step 2: Convert from the standardized normal distribution

<=>?
±𝑧 = @?
𝑥 = ±𝑧𝜎< + 𝜇< 𝑥 = ±(0.8418)(20) + (200)

183.164 < 𝑥 < 216.836

Example Problem #3 (Prob. 3.2 of textbook)

Assume that we (magically) know that 𝜇𝑥 = 303, 𝜎𝑥 = 33

• What is the probability that a measurement will never exceed 350?

§ Probability that the measurement will be between the mean and 350?

350−303 1.424
From before 𝑧𝑢𝑝𝑝𝑒𝑟 = 33
= 1.424 ∫0 𝑝 (𝑧) = 0.4228

§ Probability that the measurement will be below the mean?

§ Total probability
50% + 42.28% 92.3%