Statistical Considerations 1997

EXAMPLE 20-7 A round bar subject to a bending load has a diameter d = LN(2.000, 0.002) in. This
equivalency states that the mean diameter is µ,d = 2.000 in and the standard deviation is
ad = 0.002 in. Find the mean and the standard deviation of the second moment of area.
Solution The second moment of area is given by the equation
rrd4
l=-
64
The coefficient of variation of the diameter is
ad 0.002
Cd = - =-- = 0.001
/l,d 2

Answer
Using Table 20-6, we find
Answer 4 2 4
µ,1 = (rr/64)µ, (1 + 6C) = (rr/64)(2.000) [1 + 6(0.001) ] = 0.785 in

a1 = 4µ, Cd [1 + (9/4)CJ 2
= 4(2.000)4(0.001)[1 + (9/4)(0.001) ] = 0.064
4
in

These results can be expressed in the form

I = LN(0.785, 0.064) = 0.785LN(l, 0.0815) in4

20-5 Linear Regression

Statisticians use a process of analysis called regression to obtain a curve that best fits a
set of data points. The process is called linear regression when the best-fitting straight
line is to be found. The meaning of the word best is open to argument, because there
can be many meanings. The usual method, and the one employed here, regards a line
as "best" if it minimizes the squares of the deviations of the data points from the line.
Figure 20-10 shows a set of data points approximated by the line AB. The
standard equation of a straight line is

y = mx + b (20-31)
2 Mechanical Engineering Design

y
Figure 20-10
Set of data points approximated
by regression line AB.
where m is the slope and b is the y intercept. Consider a set of N data points (x;, Yi ). In
general, the best-fit line will not intersect a data point. Thus, we can write

(b)

where Ei = Yi - y is the deviation between the data point and the line. The sum of the
squares of the deviations is given by3

(c)

Minimizing £, the sum of the squared errors, expecting a stationary point minimum,
requires 8£/8m = 0 and 8£/8b = 0. This results in two simultaneous equations for
the slope and y intercept denoted as m and b, respectively. Solving these equations
results in

(20-32)

b
A = I:Yi - m I:x; = y - mx
N (20-33)

Once you have established a slope and an intercept, the next point of interest is to
discover how well x and y correlate with each other. If the data points are scattered all
over the xy plane, there is obviously no correlation. But if all the data points coincide
with the regression line, then there is perfect correlation. Most statistical data will be in
between these extremes. A correlation coefficient r, having the range -1 :::: r :::: +1,
has been devised to answer this question. The formula is
A Sx
r=m- (20-34)
Sy

where Sx and Sy are the standard deviations of the x coordinates and y coordinates of the
data, respectively. If r = 0, there is no correlation; if r = ±1, there is perfect correla
tion. A positive or negative r indicates that the regression line has a positive or negative
slope, respectively.
The standard deviations for m and b are given by

(20-35)

2
Sf, = Sy·x -1 + ---- (20-36)
NL (xi - i)2
where
120-37)
 EXAMPLE 20-8 A specimen of a medium carbon steel was tested in tension. With an extensometer in
place, the specimen was loaded then unloaded, to see if the extensometer reading returned
to the no-load reading, then the next higher load was applied. The loads and extensometer

I
elongations were reduced to stress a and strain E, producing the following data:

a, psi 5033 10 068 15 104 20 143 35 267

E 0.000 20 0.000 30 0.000 50 0.000 65 0.001 15

Find the mean Young's modulus E and its standard deviation. Since the extensometer
seems to have an initial reading at no load, use a y = mx + b regression.

Solution From Table 20-7, i = 0.002 80/5 = 0.000 56, y = 85 615/5 = 17 123. Note that
a regression line always contains the data centroid. From Eq. (20--32)

Answer m = 5(65.229) - o.0028(85 615) = 1. (106) si =

03 E
5(0.000 002125) - 0.00282 3 p
From Eq. (20--33)
b = 0.000 002 125(85615) - 0.002 80(65.229) = 254_ 69 si
_
5(0.000 002 125) - 0.00282 p
From Eq. (20--34), obtaining sx and Sy from a statistics calculator routine,
4
f= msx = 31031597.85(316216310- ) = 0.998
Sy 11601.11
From Eq. (20-37), the scatter about the regression line is measured by the standard
deviation sy·x and is equal to

Sy·x = /-r, y2 - b L y - m L x y
V N- 2

2 004 328 267 - (-254.69)85 615 - 31.03(106)(65.229)

5- 2
= 811.1 psi

Table 20-7

I (J1 psi E
Worksheet for Ex. 20-6 x2 xy y2 (X -
X)2

5 033 0.000 20 0.000 000 040 1.006 600 25 330 089 0.000 000
y x 130
10 068 0.000 30 0.000 000 090 3.020 400 101 364 624 0.000 000
 From Eq. (20--35), the standard deviation of m is
s 811.1
Answer s· - y·x = = 1.086(106) psi = S
E m- .J I:C x - x)2 .Jo.ooo ooo 558
See Fig. 20--11 for the regression plot.

Figure 20-1 1 u
50 000
The data from Ex. 20-8 are
plotted. The regression line
passes through the data
centroid and among the data
points, minimizing the squared
deviations.

Engineering strain e

PROBLEMS
20-1 At a constant amplitude, completely reversed bending stress level, the cycles-to-failure experi
ence with 69 specimens of 5160H steel from 1.25-in hexagonal bar stock was as follows:

L 60 70 80 90 100 110 120 130 140 150 160 170 180 190 200 210

f 2 3 5 8 12 6 10 8 5 2 3 2 0

20-2
 where L is the life in thousands of n of the life for the population from which the sample was drawn.
cycles, and f is the class frequency of
Determinations of the ultimate tensile strength s.1 of stainless steel sheet (l7-7PH, condition TH
failures.
(a) Construct a histogram with class
frequency f as ordinate.
(b) E
s
t
i
m
a
t
e

t
h
e

m
e
a
n

a
n
d

s
t
a
n
d
a
r
d

d
e
v
i
a
t
i
o
998 Mechanical Engineering Design

20-3 A total of 58 AISI 1018 cold-drawn steel bars were tested to determine the 0.2 percent offset
yield strength Sy. The results were

I
Sy, kpsi
:4 68 72 76 80 84 88 92
f 6 6 9 19 10 4 2

where Sy is the class midpoint and f is the class frequency. Estimate the mean and standard
deviation of Sy and its PDF assuming a normal distribution.

20-4 The base 10 logarithm of 55 cycles-to-failure observations on specimens subjected to a constant

stress level in fatigue have been classified as follows:

y 5.625 5.875 6.125 6.375 6.625 6.875 7.125 7.375 7.625 7.875 8.125

f 0 0 3 3
6 14 15 10 2

Here y is the class midpoint andf is the class frequency.

(a) Estimate the mean and standard deviation of the population from which the sample was taken
and establish the normal PDF.
(b) Plot the histogram and superpose the predicted class frequency from the normal fit.

20-5 A -in nominal diameter round is formed in an automatic screw machine operation that is
initially set to produce a 0.5000-in diameter and is reset when tool wear produces diameters in
excess of 0.5008 in. The stream of parts is thoroughly mixed and produces a uniform distribution
of diameters.
(a) Estimate the mean and standard deviation of the large batch of parts from setup to reset.
(b) Find the expressions for the PDF and CDF of the population.
(c) If, by inspection, the diameters less than 0.5002 in are removed, what are the new PDF and CDF
as well as the mean and standard deviation of the diameters of the survivors of the inspection?

20-6 The only detail drawing of a machine part has a dimension smudged beyond legibility. The round
in question was created in an automatic screw machine and 1000 parts are in stock. A random sam
ple of 50 parts gave a mean dimension of d = 0.6241 in and a standard deviation of s = 0.000 581
in. Toleranced dimensions elsewhere are given in integral thousandths of an inch. Estimate the
miss ing information on the drawing.

20-7 (a) The CDF of the variate x is F(x) = 0.555x - 33, where x is in millimeters. Find the
PDF, the mean, the standard deviation, and the range numbers of the distribution.
(b) In the expression rr = F/A, the force F = LN(3600, 300) lbf and the area is A =
LN(0.112, 0.001) in2• Estimate the mean, standard deviation, coefficient of variation, and
distribution of er.

20-8 A regression model of the form y = a1x +a 2x 2 is desired. From the normal equations
 For the data set

x I o.o 0.
2
0.
4
0.
6
0.
8
1.
0
0.1 0.2 0.2 0.1 -
5 5 5 7 0.0

find the regression equation and plot the data with the regression model.

20-9 R. W. Landgraf reported the following axial (push-pull) endurance strengths for steels of differ
ing ultimate strengths:

65 29. 325 114 28 9

6 5
30 23 10 0
29 6
9
0
82 45 8
13 967 5
12 9
4
6 48 0
20 8 0
18 8
8
4
101 51 7
205 7
96 0
213 4
7
119 50 22 99 242 15
195 78 5
325 117 134 06
210 87 355 12 145 0
6
230 105 22 28 227 14
265 105 5 7 1

(a) Plot the data with s; as ordinate and s.as abscissa.

(b) Using the y = mx + b linear regression model, find the regression line and plot.

20-10 In fatigue studies a parabola of the Gerber type

O'a + ( Um ) = l2

s. s.,
is useful (see Sec. 6-12). Solved for ua the preceding equation becomes

This implies a regression model of the form y = a0 + a2x 2 • Show that the normal equations are

LY = nao + a2 Lx 2

and that

and
1000 Mechanical Engineering Design

20-1 1 Consider the following data collected on a single helical coil extension spring with an initial
extension F; and a spring rate k suspected of being related by the equation F = F; + kx where
x is the deflection beyond initial. The data are

x, in 0. 0. 0. 1. 2
4 6 8 0 .
F, lbf 10. 12. 13. 16. 2
3 1 8 2 5.
(a) Estimate the mean and standard deviation of the initial tension F; .
(b) Estimate the mean and standard deviation of the spring rate k.
20-1 2 In the expression for uniaxial strain E = 811, the elongation is specified as 8 (0.0015,0.000 092)
in and the length as I (2.0000,0.0081) in. What are the mean, the standard deviation, and the
coefficient of variation of the corresponding strain E.
20-1 3 In Hooke's law for uniaxial stress, u = eE, the strain is given as e (0.0005, 0.000034) and
Young's modulus as E (29.5, 0.885) Mpsi. Find the mean, the standard deviation, and the coef
ficient of variation of the corresponding stress u in psi.
20-14 The stretch of a uniform rod in tension is given by the formula 8 = Fl/AE . Suppose the terms in
this equation are random variables and have parameters as follows:

F (14.7, 1.3) kip

A (0.226, 0.003) in2
I (1.5, 0.004) in
E (29.5, 0.885) Mpsi

Estimate the mean, the standard deviation, and the coefficient of variation of the corresponding
elongation 8 in inches.

20-1 5 The maximum bending stress in a round bar in flexure occurs in the outer surface and is
given by the equation u = 32M/ nd 3 . If the moment is specified as M (15 000,1350) lbf ·
in and the diameter is d (2.00, 0.005) in, find the mean, the standard deviation, and the
coefficient of vari ation of the corresponding stress u in psi.

20-16 When a production process is wider than the tolerance interval, inspection rejects a low-end scrap
fraction ot with x < x1 and an upper-end scrap fraction f3 with dimensions x > x 2 . The surviving
population has a new density function g(x ) related to the original f (x) by a multiplier a. This is
because any two observations x; and X j will have the same relative probability of occurrence as
before. Show that

a = -----
1
F(x2 ) - F(x1 ) 1 - (ot + {3)
and
f(x) f (x)
g(x ) =;(x2) - F(x1 ) = 1 - (ot + {3)
{ otherwise
20-1 7
An automatic screw machine produces a run of parts with a uniform distribution d =