Beruflich Dokumente
Kultur Dokumente
© Dr, J. Chen, School of Engineering and Design, Brunel University 1 ME1302 Analytical Methods and Skills
Statistical Analysis of Experimental Data
having the form should exist between the variables. 8.7 2.50
PV γ = C 10.8
11.3
0.57
0.32 K
⎧⎪ ⎛ u ⎞ K ⎫⎪
11.8 0.14 K⎛u⎞
12.3 0.19 p = f (u ) = ⎜ ⎟ exp ⎨ − ⎜ ⎟ ⎬
C , γ are parameters to be found.
– where 12.9
13.4
0.09
0.07
u ⎝C ⎠ ⎪⎩ ⎝ C ⎠ ⎭⎪
13.9 0.05
14.4
15.4
0.03
0.01 • Need to find parameters K and C.
17.0 0.01
© Dr, J. Chen, School of Engineering and Design, Brunel University 2 ME1302 Analytical Methods and Skills
Statistical Analysis of Experimental Data
© Dr, J. Chen, School of Engineering and Design, Brunel University 3 ME1302 Analytical Methods and Skills
Statistical Analysis of Experimental Data
Ø Fix b: dJ ( a ) = 0
da
J* b* Ø Fix a: dJ ( b )
dy b =0
= f ' ( x*) = 0 a* db
dx x*
a
x* x
• Find minimum with partial derivatives:
dy ∂J ( a ,b ) ∂J ( a ,b )
• Find minimum: = f ' ( x*) = 0 =0 =0
dx x* ∂a ∂b
© Dr, J. Chen, School of Engineering and Design, Brunel University 4 ME1302 Analytical Methods and Skills
Statistical Analysis of Experimental Data
Fitting: n n n
– The sum of the squared errors for all the data points a ∑ xi2 + b ∑ xi = ∑ xi yi ( A)
i =1 i =1 i =1
n n
J ( a ,b ) = e =
i
2
(axi + b − yi )2 ∂ J ( a ,b ) n ∂ (axi + b − yi ) n
i =1 i =1 = ∑ 2(axi + b − yi ) = ∑ 2(axi + b − yi ) = 0
∂b i =1 ∂b i =1
• We now choose a and b to minimize J .
⎧n 2⎫
n n
a ∑ xi + bn = ∑ yi (B)
min J ( a ,b ) = min ⎨ axi + b − yi ⎬ i =1 i =1
a ,b a ,b
⎩ i =1 ⎭
© Dr, J. Chen, School of Engineering and Design, Brunel University 5 ME1302 Analytical Methods and Skills
Statistical Analysis of Experimental Data
(∑ x ) − n (∑ x )
2 2
i
n i i =1 i =1 i =1 S xx
1
S yy = ∑ ( yi − y ) 2 = ( ∑ yi ) − ( ∑ yi2 )
2
• The resulting line (linear function) is called the least- n
squares best fit to the data represented by (xi, yi).
• Another commonly used measure of how well the least-
S xy = ∑ ( xi − x )( yi − y ) = ∑ xi yi −
1
( ∑ xi )( ∑ yi ) x=
∑x i
squares line will represent the data is called the standard
n n error of estimate, given by:
S xx = ∑ ( xi − x ) 2 = ( ∑ xi ) −
2 1
( ∑ xi2 ) y=
∑y i
S=
SSE
=
yi2 − b yi − a xi yi
n n n−2 n−2
y = ax ⎧ n 2⎫
Experiment x y ∑ x = 1 + 2 + 3 + 4 = 10
min J ( a ) = min ⎨∑ ( axi − yi ) ⎬
i
(1) 1 6
a a
⎩ i =1 ⎭
(2) 2 7
∑ y = 6 + 7 + 9 + 12 = 34
i
dJ ( a ) n d ( axi − yi ) n ∑ x y = 1× 6 + 2 × 7 + 3 × 9 + 4 ×12 = 95
= ∑ 2 ( axi − yi ) = ∑ 2 ( axi − yi ) xi = 0 (3) 3 9 i i
da da
∑x
i =1 i =1
(4) 4 12 2
= 12 + 22 + 32 + 42 = 30
i
n n
2a ∑ xi2 − 2∑ yi xi = 0 n=4
i =1 i =1
n
a=
( ∑ x )( ∑ y ) − n ( ∑ x y ) = 10 × 34 − 4 × 95 = 2
i i i i
∑yx i i (∑ x ) − n (∑ x ) i
2
10 − 4 × 30 2
i
2
a= i =1
n
∑x 2
i ∑ y − a∑ x 34 − 2 ×10
i =1 b= i i
= = 3.5
n 4
© Dr, J. Chen, School of Engineering and Design, Brunel University 6 ME1302 Analytical Methods and Skills
Statistical Analysis of Experimental Data
b=
(∑ x )(∑ x y ) − (∑ x )(∑ y ) = 7.5 × 13.94 − 13.75 × 7.66 = 0.0295
i i i
2
i i
• The following table gives experimental values of the
(∑ x ) − n(∑ x )
2 2
7.5 − 6 × 13.752 pressure P of a given mass of gas corresponding to various
values of the volume V.
i i
•The resulting best fit line is then: V (m3) 54.3 61.8 72.4 88.7 118.6 194.0
y = 0.9977x + 0.0295 P (N/m2) 61.2 49.2 37.6 28.4 19.2 10.1
or V = 0.9977L + 0.0295
• According to thermodynamic principles, a relationship
•The standard error for these data is obtained by:
having the form should exist between the variables.
(∑ y ) − b (∑ y ) − a (∑ x y )
2 PV γ = C
S=
i i i i
n−2 C, γ
– where are constants.
14.137 − 0.0295 × 7.66 − 0.9977 × 13.94
= = 0.0278
6−2
© Dr, J. Chen, School of Engineering and Design, Brunel University 7 ME1302 Analytical Methods and Skills
Statistical Analysis of Experimental Data
Example Example
a) Find the values of and C. 70
• Solution: PV = C 50
b=
(∑ x )(∑ xy ) − (∑ x )(∑ y ) = 11.6953 × 16.8543 − 23.0059 × 8.7975 = 4.2
2
(∑ x ) − n (∑ x )
2 2
11.6953 − 6 × 23.0059
2
© Dr, J. Chen, School of Engineering and Design, Brunel University 8 ME1302 Analytical Methods and Skills
Statistical Analysis of Experimental Data
1 y1 = ax + b
0.9
1.7 1.8 1.9 2 2.1 2.2 2.3 2.4 We now have a form in which y1 (ln(y)) is a linear function of x.
• Geompertz curve: y = pq b
x
⇒ log y = b x log q + log p
9.3
9.8
0.968
0.980 ln( − ln(1 − P )) = K ln(u ) − K ln(C )
10.3 0.985
10.8 0.991
1 1 11.3 0.994
y a x b
• Logistic curve: y= ⇒ = ab x + 5 11.8 0.995
ab + 5
x
y 12.3
12.9
0.997
0.998
y = ax + b
13.4 0.999
© Dr, J. Chen, School of Engineering and Design, Brunel University 9 ME1302 Analytical Methods and Skills
Statistical Analysis of Experimental Data
∑x
n
rxy = i =1
i ∑y i
n n
x= i =1
y= i =1
? n n
x x x ( xi − x ) 2 ( yi − y ) 2
(a) i =1 i=1
line of best fit: Yes (b) line of best fit: NO (c) line of best fit: UNSURE
(a) A strong relationship between y and x. • Measure the strength and direction of a linear
(b) No functional relationship between y and x. relationship between two random variables.
(c) Uncertain.
© Dr, J. Chen, School of Engineering and Design, Brunel University 10 ME1302 Analytical Methods and Skills
Statistical Analysis of Experimental Data
∑x
n
rxy = i =1 i ∑y i
• A value of +1 would indicate a perfectly linear relationship between
the variables with a positive slope (i.e., increasing x results in
n n x= i =1
y= i =1
n increasing y).
( xi − x ) 2
( yi − y ) 2 n
i =1 i =1 • A value of -1 indicates a perfectly linear relationship with negative
slope (i.e., increasing x produces decreasing y).
Alternative Formula: • A value of zero indicates that there is no linear correlation between
the variables.
n xy − x y – Even if there is no correlation, it is unlikely that rxy will be
rxy = exactly zero.
2 2
n x2 − x n y2 − y y y
rxy = 1 rxy = −1
x x
relationship will offer some approximation of the true and the level of 20
25
0.299
0.265
0.378
0.337
0.444
0.396
0.516
0.462
0.561
0.505
confident that a linear functional relationship exists. 200 0.091 0.116 0.138 0.163 0.180
© Dr, J. Chen, School of Engineering and Design, Brunel University 11 ME1302 Analytical Methods and Skills
Statistical Analysis of Experimental Data
Example x2 y2
Solution x y xy
• It is thought that lap times for a race car depend on the 40 65.3 2612.0 1600 4264.09
ambient temperature. The following data for the same car 47 66.5 3125.5 2209 4422.25
with the same driver were measured at different races: 55 67.3 3701.5 3025 4529.29
62 67.8 4203.6 3844 4596.84
Ambient 66 67 4422.0 4356 4489.00
40 47 55 62 66 88
temperature (°F)
88 66.6 5860.8 7744 4435.56
Lap time (s) 65.3 66.5 67.3 67.8 67.0 66.6 ∑ ∑ ∑ ∑ ∑
70 358.00 400.50 23925.4 22778 26737.03
• Does a linear
n ∑ xy − ( ∑ x )( ∑ y )
68
relationship exist
rxy =
between these two
66
n∑ x2 − (∑ x ) n∑ y 2 − (∑ y )
2 2
© Dr, J. Chen, School of Engineering and Design, Brunel University 12 ME1302 Analytical Methods and Skills
Statistical Analysis of Experimental Data
n ∑ xy − ( ∑ x )( ∑ y )
4
rxy =
n∑ x 2 − ( ∑ x ) n∑ y 2 − ( ∑ y )
2 2 2
0
0 10 20 30 6 × 272.12 − 96 × 23.5
= = −0.8406
(6 × 2084.8 − 96 ) ( 6 × 119.87 − 23.5
Mass (m, kg)
2 2
)
Example Example 1
x= y= f
• For a 99% confidence level, α=1-0.99=0.01. Transformation m
• For six pairs of data, from Table 1, we obtain a m 2.0 7.6 13.2 18.8 24.4 30.0
value of rt, of 0.917 . x=
1
0.7071 0.3627 0.2752 0.2306 0.2024 0.1826
m
rxy = 0.8406 < 0.917 = rt y=f 8.5 4.3 3.2 2.8 2.4 2.3
10.0
5.0
exists.
2.5
0
0.15 0.30 0.45 1 0.60 0.75
m
© Dr, J. Chen, School of Engineering and Design, Brunel University 13 ME1302 Analytical Methods and Skills
Statistical Analysis of Experimental Data
G ⎛ 2 G ⎞ 1
6 × 10.0025 − 1.9607 × 23.5 f = kd 2 = ⎜ kd ⎟
= = 0.9996 mnD 3 ⎝ nD 3 ⎠ m
(6 × 0.8348 − 1.9607 ) ( 6 × 119.87 − 23.5
2 2
)
© Dr, J. Chen, School of Engineering and Design, Brunel University 14 ME1302 Analytical Methods and Skills
Statistical Analysis of Experimental Data
© Dr, J. Chen, School of Engineering and Design, Brunel University 15 ME1302 Analytical Methods and Skills
Statistical Analysis of Experimental Data
© Dr, J. Chen, School of Engineering and Design, Brunel University 16 ME1302 Analytical Methods and Skills
Statistical Analysis of Experimental Data
© Dr, J. Chen, School of Engineering and Design, Brunel University 17 ME1302 Analytical Methods and Skills
Statistical Analysis of Experimental Data
Experiment Example xi xi − x ( x i − x )2
5.3 -0.3130 0.0980
• The length of a steel bar is measured 10 5.73 0.117 0.0137
times and results are: 6.77 1.157 1.3386
5.26 -0.353 0.1246
x(cm) 5.30 5.73 6.77 5.26 4.33 5.45 6.09 5.64 5.81 5.75
4.33 -1.283 1.6461
5.45 -0.163 0.0266
• Average: xi 56.13 6.09 0.477 0.2275
x= = = 5.613 ( cm ) 5.64 0.027 0.0007
n 10
• Standard Deviation: 5.81 0.197 0.0388
5.75 0.137 0.0188
( xi − x ) 2 3.5334 ∑x ∑(x − x)
2
S= = = 0.627 (cm) i i
n −1 10 − 1 56.13 3.5334
Standard Deviation for Weighted Mean 3.2 Dealing with large number of data:
• The numbers x1, x2, ..., xn are associated with Organise data by groups
weighting factors (or weights) w1, w2, ..., wn.
n • When summarising large number of raw data, it is
wi xi often useful to distribute the data into groups (classes),
w x + w2 x2 + wn xn or categories, and to determine the number of
x= 1 1 = i =1
individuals belonging to each case, called the class
w1 + w2 + + wn n
wi frequency.
i =1
• A tabular arrangement of data by classes together with
n the corresponding class frequencies is called frequency
n1 wi ( xi − x ) 2 Here n1 is the number of distribution.
S= i =1
n
non-zero weights.
(n1 − 1) wi
i =1
© Dr, J. Chen, School of Engineering and Design, Brunel University 18 ME1302 Analytical Methods and Skills
Statistical Analysis of Experimental Data
© Dr, J. Chen, School of Engineering and Design, Brunel University 19 ME1302 Analytical Methods and Skills
Statistical Analysis of Experimental Data
Class Frequency
Class mid-
Number of Measurements
Class Frequency 1085 ≤ T <1090 1 20
point 12
1090 ≤ T <1095
5
3 16 P= = 0 .2
1085 ≤ T <1090 1087.5 1 60
1095 ≤ T <1100 12 12
1090 ≤ T <1095 1092.5 3
1095 ≤ T <1100 1097.5 12 1100 ≤ T <1105 21 8 12
© Dr, J. Chen, School of Engineering and Design, Brunel University 20 ME1302 Analytical Methods and Skills
Statistical Analysis of Experimental Data
Relative Frequency
1090 ≤ T <1095 5% 26.7%
P = 0. 2
5
Relative Frequency
1100 ≤ T <1105 35% 0.0533
5
the random
Normalised
13.3% Area=0.2
1105 ≤ T <1110 23.33% 6.7%
12
60
= 20% 0.004 variable in each
1110 ≤ T <1115 11.67% 0.0267 class is the area of
1115 ≤ T <1120 3.33% 1085 1090 1095 1100 1105 1110 1115 1120
the corresponding
Temperature (° C ) 0.0133
Sum Total 1 bar in the
• The relative frequency of a class is the frequency
1085 1090 1095 1100 1105 1110 1115 1120
Temperature (° C )
histogram.
of the class divided by the total frequency of all
classes and generally expressed as a percentage.
0.0533 Aera=0.2
Normalised
D a ta R anges N um bers
0.004
205 – 209 2
0.0267
210 – 214 8
0.0133 215 – 219 20
220 – 224 39
1085 1090 1095 1100 1105 1110 1115 1120
Temperature (° C ) 225 – 229 42
© Dr, J. Chen, School of Engineering and Design, Brunel University 21 ME1302 Analytical Methods and Skills
Statistical Analysis of Experimental Data
n = f 1+ f 2 + + fK
© Dr, J. Chen, School of Engineering and Design, Brunel University 22 ME1302 Analytical Methods and Skills
Statistical Analysis of Experimental Data
Determination of Probability
• Probability is a numerical value expressing the • Probabilities: S = { X = 1, 2,3, 4,5, 6}
likelihood of occurrence of an event relative to all
possibilities (sample space) for an experiment. 1
• Probability is a mathematical quantity that is linked to A1 = X | X = 2 = 2 P ( X = 2) = ,
6
the frequency with which a certain phenomenon
occurs after a large number of trial. 2
A2 = X | X ≥ 5 = {5, 6} P( X ≥ 5) = ,
P(A ) =
number of possible outcome favorable to the event A 6
total number of possible equally likely outcomes
3
An event (A) is an subset of sample space S.
A3 = X | X is even = {2, 4, 6} P ( X is even) =
6
© Dr, J. Chen, School of Engineering and Design, Brunel University 23 ME1302 Analytical Methods and Skills
Statistical Analysis of Experimental Data
• Probability is the chance of occurrence of an event in • The probability of an event happening usually lies
an experiment. somewhere between two extreme values, 0 and 1.
• The probability is obtained by dividing the number 0≤ P≤1
of successful occurrence by the total number of trails.
• Probability is based upon relative frequency. • Success: an event occurs A e.g . x ≥ 0
• Values of probability lie between 0 and 1, where 0 • Fail: an event fails to occur A e.g . x < 0
represents an absolute impossibility and 1 represents
an absolute certainty. P( A) + P ( A) = 1
• The probability of an event happening usually lies
somewhere between two extreme values, 0 and 1. P ( A) = 1 This event A definitely occurs A S
© Dr, J. Chen, School of Engineering and Design, Brunel University 24 ME1302 Analytical Methods and Skills
Statistical Analysis of Experimental Data
S = { X = 1, 2,3, 4,5, 6}
• Probability Addition Law: A B A : X is odd= 1,3,5
and B are NOT mutually
B : X ≥ 5 = 5, 6
exclusive events, the
S A
probability of either event A ( A or B) = ( X ≥ 5) or ( X is odd) = 1,3,5, 6
or event B occurs (not both) The outcome of dice throw is either odd or not less than 5.
is: ( A and B ) = ( X ≥ 5) and ( X is odd) = 5
P ( A or B ) = P ( A) + P( B ) − P( A and B)
The outcome of dice throw is odd and not less than 5.
The outcome of dice throw is either odd or not less than 5. 4 1
P ( A or B) = P( A and B) =
A : X is odd P( A and B) = P(outcome is odd and not less than 5) = 1 6 6
6 Check:
B: X ≥5
3 2 1 4 3 2 1 4
P ( A or B) = + − = A and B = 5 P( A) + P( B) − P( A and B) = + − = = P( A or B)
6 6 6 6 6 6 6 6
A or B = {1,3,5, 6}
© Dr, J. Chen, School of Engineering and Design, Brunel University 25 ME1302 Analytical Methods and Skills
Statistical Analysis of Experimental Data
• Independent event: Events are independent when the • Independent event: Events are independent when the
occurrence of an event does not affect the occurrence occurrence of an event does not affect the occurrence
of the second event. of the second event.
Event A = {throwing a six} when rolling a die A bag of 100 balls (80 white and 20 black). Student 1
Event B = {drawing an ace} from a pack of 52 cards pick up one ball randomly and then Student 2 randomly
pick one ball.
1 4 1
P ( A ) = P( X = 6) = P ( B ) = P(Y = ace) = = Event A = {Student 1 picked black ball}
6 52 13
Event B = {Student 2 pick black ball}
• Multiplication law of probabilities: The probability of Event A and Event B are independent if Student 1 puts back
the occurrence of both events A and B is the ball before Student 2.
1 1 1 Event A and Event B are not independent if Student 1 does not
P ( A and B ) = P( A) P( B) = × = put back the ball when Student 2 picks the ball.
6 13 78
© Dr, J. Chen, School of Engineering and Design, Brunel University 26 ME1302 Analytical Methods and Skills
Statistical Analysis of Experimental Data
• The probability distribution of a discrete random • The probability of a continuous random variable x
variable x is a list or table of the distinct values of x is described by the probability density function
and the probabilities associated with those values. (PDF).
x 1 2 3 4 5 6
P( x ) = 1
P(x) 1/6 1/6 1/6 1/6 1/6 1/6
© Dr, J. Chen, School of Engineering and Design, Brunel University 27 ME1302 Analytical Methods and Skills
Statistical Analysis of Experimental Data
• For entire range of possible value of X (from a to b), the • Example: The lifetime of an electronic component (in
probability density function is constant. thousands of hours) is a continuous random variable
1 with pdf: ⎧ 1 − 2x
f ( x) = ⎧ 1 1
b−a ⎪ a≤ x≤b ⎪ e x≥0
f ( x) = ⎨ b − a f ( x) = ⎨ 2
⎪⎩ 0 ⎪ 0 x<0
otherwise ⎩ 0 x
0 a b Find the proportion of components that last longer than 6000 hours.
Random variable’s value, x 6
P ( X > 6) = 1 − P ( X ≤ 6) = 1 − f ( x) dx
The measurement of length use an ordinarily ruler, the round- −∞
off error X, could take any value between -0.05 cm to 0.05cm. 6
6 1 − 2x ⎡ − 2x ⎤
10 20 30 40 50 = 1− e dx = 1 − ⎢ −e ⎥
60 0 2 ⎣ ⎦0
⎧10 −0.05 ≤ x ≤ 0.05
f ( x) = ⎨ 6
= 1 − ⎡⎣ −(e −3 − e0 ) ⎤⎦ = e −3 ≈ 0.05
⎩0 otherwise 0
Approximately, 1 in 20 components lasts longer than 6000 hours.
© Dr, J. Chen, School of Engineering and Design, Brunel University 28 ME1302 Analytical Methods and Skills
Statistical Analysis of Experimental Data
• For a random variable with the pdf f(x), the man and ⎧α e −α x x≥0
f ( x) = ⎨
standard deviation can be determined as: ⎩ 0 x<0
Mean: +∞ +∞ +∞ +∞ de −α x
μ= xf ( x) dx μ= xf ( x)dx = α xe −α x
dx = − x dx
−∞ −∞ 0 0 dx
Standard +∞ Mean: +∞ +∞ dx
deviation: σ= ( x − μ ) 2 f ( x) dx = − ⎡⎣ xe −α x ⎤⎦ + e −α x dx (integrating by
−∞ 0 0 dx parts)
+∞
= 0+ e −α x dx
0
1 +∞ 1
=− ⎡⎣e −α x ⎤⎦ =
α 0 α
+∞ 1
σ= ( x − μ ) 2 f ( x) dx =
−∞ α
0.8 σ=0.5
• The normal (or Gaussian) distribution function is a mean
σ=0.6
0.6 μ=2
simple distribution function that is useful for many
σ=1
common problems involving random variables.
f(x)
0.4
σ=2
1 ⎡ − ( x − μ )2 ⎤ μ mean 0.2
f (x) = exp ⎢ ⎥
2 πσ ⎣ 2σ
2
⎦ σ s tan dard deviation
0
-1 0 1 2 3 4 5
Random variable, x
1 X ~ N (μ, σ 2 ) • There is only one maximum and it occurs at the mean value.
f(x)
2π σ • The density function is symmetrical about the mean value.
• The width of the density function is directly proportional to the
standard deviation, σ.
mean μ • The maximum value of the density function is inversely
Random variable, x
proportional to the standard deviation σ.
© Dr, J. Chen, School of Engineering and Design, Brunel University 29 ME1302 Analytical Methods and Skills
Statistical Analysis of Experimental Data
t F
• The probability that a random variable falls within one or more ν=∞
standard deviations (σ’s) of the mean. 0.4 N (0,1) ν=5
ν=1 v is degree of freedom
f(v,t)
Range Probability f (v, t )
f(x) 1 0.2
1σ 68.26% 2πσ
t the value for a
2σ 95.44% random variable
0
0 t
3σ -3 -2 -1 1 2 3
99.74%
1.645σ 90% x
1.96σ 95% μ−3σ μ−2σ μ−σ μ μ+σ μ+2σ μ+3σ
68.26%
2.33σ 98% 95.44% d1 and d2 are
99.74% degrees of
2.58σ 99%
freedom
3.3σ 99.9%
f(z)
f(x) be evaluated 0.2
P ( x1 ≤ X ≤ x2 ) analytically, and as z1 z2
a result, the pdf 0
-3 -2 -1 0 1 2 3
x
integration can be x1 x2 Random variable, z
© Dr, J. Chen, School of Engineering and Design, Brunel University 30 ME1302 Analytical Methods and Skills
Statistical Analysis of Experimental Data
Z ~ N (0,1)
z
0.0 .0000 .0040 .0080 .0120 .0160 .0199 .0239 .0279 .0319 .0359
row. 0.9 .3159 .3186 .3212 .3238 .3264 .3289 .3315 .3340 .3365 .3389
1.0
1.1
.3413
.3643
.3438
.3665
.3461
.3686
.3485
.3708
.3508
.3729
.3531
.3749
.3554
.3770
.3577
.3790
.3599
.3810
.3621
.3830
z 0.04 0.05 P (0 ≤ Z ≤ 2.24)
• The top row serves as the 1.2
1.3
.3849
.4032
.3869
.4049
.3888
.4066
.3907
.4082
.3925
.4099
.3944
.4115
.3962
.4131
.3980
.4147
.3997
.4162
.4015
.4177
Table Provide:
3.6 .4998 .4998 .4999 .4999 .4999 .4999 .4999 .4999 .4999 .4999
P (0 ≤ Z ≤ a )
3.9 .5000 .5000 .5000 .5000 .5000 .5000 .5000 .5000 .5000 .5000
4.0 .5000 .5000 .5000 .5000 .5000 .5000 .5000 .5000 .5000 .5000
0.4 P( z1 ≤ Z ≤ 0) P(0 ≤ Z ≤ z1 )
P( z1 ≤ Z ≤ z 2 )
f(z)
0.2
z2 z1 z1 z1
z1 z z
0 z -2 -1 0 1 2 3 -2 -1 0 1 2 3
-3 -2 -1 0 1 2 3
P( z1 ≤ Z ≤ 0) = P(0 ≤ Z ≤ z1 )
P (0 ≤ Z ≤ z2 ) P( z1 ≤ Z ≤ 0)
• Standard Normal Distribution Table:
– It is practical to perform the integration numerically and tabulate
z2 z2 the results as a function of z.
z1 z1
z z
-2 -1 0 1 2 3 -2 -1 0 1 2 3
P( z1 ≤ Z ≤ z2 ) = P(0 ≤ Z ≤ z2 ) + P( z1 ≤ Z ≤ 0)
© Dr, J. Chen, School of Engineering and Design, Brunel University 31 ME1302 Analytical Methods and Skills
Statistical Analysis of Experimental Data
f(z)
0.2
x1 − μ
z2 z1 =
pdf 0
z1
σ
-3 -2 -1 0 1 2 3 P ( x1 ≤ X ≤ x2 ) P ( z1 ≤ Z ≤ z2 )
x1 x2
x
Random variable, z x2 − μ
x1 − μ z2 =
z1 = σ
σ
P ( x1 ≤ X ≤ x2 ) P ( z1 ≤ Z ≤ z2 )
x2 − μ • Convert Normal Distribution into Standard Normal
Convert Normal Distribution z2 = Standard Normal Distribution
σ Distribution
Example
• The random variable x follows normal distribution
μ = 10.0; σ = 1
z1 =
x1 − μ
z2 =
x2 − μ Mean to be found: μ
• Find probability: P(9 ≤ X ≤ 12); σ σ
9 − 10 12 − 10 Mean found from x1 + x2 + xn
z1 =
1
= −1; z2 == 2; experiment: x=
1 f(z)
n
P (9 ≤ X ≤ 12) = P ( −1 ≤ Z ≤ 2)
=
μ = x ±δ
z2
P(−1 ≤ Z ≤ 2)
z1 z
-3 -2 -1 0 1 2 3 Mean estimation:
= P(−1 ≤ Z ≤ 0) + P(0 ≤ Z ≤ 2)
f(z) f(z)
= P(0 ≤ Z ≤ 1) + P(0 ≤ Z ≤ 2) +
δ
z2
z z1 z
= 0.8185 ; 81.85%
Table |z1| z
-3 -2 -1 0 1 2 3
© Dr, J. Chen, School of Engineering and Design, Brunel University 32 ME1302 Analytical Methods and Skills
Statistical Analysis of Experimental Data
Population Sample
• The population • A sample is a
• A medical laboratory analyses specimens of a
comprises the entire representative subset of a
pharmaceutical product to determine the concentration of
collection of objects, population on which an
the active ingredient.
measurements, experiment is performed
– The chemical analysis procedure is not precise so that repeated
observations, and so on. and numerical data are
measurements on the same specimen will give different results.
– All of ball bearings obtained.
– The standard deviation for the procedure is σ=0.02 grams per litre.
produced. – 100 (or two dozen) ball
– The laboratory analyzes each specimen 36 times and reports the
– Wind speeds at every bearings selected.
mean result.
instant in 30 minutes. – wind speed sampled once
– The laboratory has been asked to evaluate the claim that the
each minutes over the 30
concentration of the active ingredient is 0.86 grams per litre.
minutes period.
36 test results, Accept/Reject claim of active
ingredient 0.86 with 99% confidence • Different samples of the same population may be
mean value is
0.8404. level. chosen for experimentation.
© Dr, J. Chen, School of Engineering and Design, Brunel University 33 ME1302 Analytical Methods and Skills
Statistical Analysis of Experimental Data
Estimation of Population Mean and Std Estimation of Population Mean and Std
• The sample mean can be used as an estimation of the • Population: 1000 ball bearings μ = 0.5021cm σ = 0.2892cm
population mean.
Sample Mean Standard 90%
x ( x − δ ,x + δ )
of 100 x deviation S interval δ
Sample 1 0.5286 0.2818 0.0464 (0.4822, 0.5750)
• The sample standard deviation can be used as an Sample 2 0.4652 0.2790 0.0459 (0.4193, 0.5111)
estimation of the population standard deviation. Sample 3 0.5312 0.2715 0.0447 (0.4865, 0.5759)
Sample 4 0.5150 0.2633 0.0433 (0.4717, 0.5583)
Sample 5 0.4933 0.3153 0.0519 (0.4414, 0.5452)
S
• Different samples from the same population will yield
different values.
• However, x − δ ≤ μ ≤ x + δ
• Simple estimation of mean is not sufficient.
© Dr, J. Chen, School of Engineering and Design, Brunel University 34 ME1302 Analytical Methods and Skills
Statistical Analysis of Experimental Data
Confidence Level and Level of Significance Mean and Standard Deviation of Sample Mean
• A population of the random variable x:
• The confidence level is normally expressed in terms of a μ
– Population mean
variable α called the level of significance. – Population standard deviation σ
confidence level = P (| x − μ |≤ δ ) = 1 − α • Take a sample of size n:
– Sample mean x
• α is then the probability that the mean will fall outside S
– Sample standard deviation
the confidence interval.
• Different sample yields different sample mean (There are
P (| x − μ |> δ ) = α
random variations in the sample mean).
• Examples: • The sample mean is also a random variable! x
α = 2%; P (| μ − x |≤ δ ) = 0.98 = 98%
• Mean and standard deviation of the sample mean.
α = 5%; P (| μ − x |≤ δ ) = 0.95 = 95% σ
μx = μ σx =
α = 10%; P (| μ − x |≤ δ ) = 0.9 = 90% n
© Dr, J. Chen, School of Engineering and Design, Brunel University 35 ME1302 Analytical Methods and Skills
Statistical Analysis of Experimental Data
Sample Mean Probability Distribution Sample Mean Distribution Standard Normal Distribution
⎛ σ ⎞ z ~ N (0,1)
• The sample mean approaches to a normal distribution if x ~ N⎜μ, ⎟
⎝ n⎠ x −μ
the sample size n is sufficiently large. z=
⎛ σ ⎞ σx σ
x ~ N⎜μ, ⎟ σx =
⎝ n⎠ n
f(z)
α • Confidence Level: 1 − α = 90% α = 0.1
P( 0 < z < zc ) = 0.5 − α
Area = α/2
confidence 2 P( 0 < z < zc ) = 0.5 − = 0.45
interval Area = α/2 2
1-α Use Standard Normal Table: P(0 < z < 1.64 ) = 0.4495 P (0 < z < 1.65) = 0.4505
Distribution Table to find:
-zc z=0 zc zc zc ≈ 1.645
© Dr, J. Chen, School of Engineering and Design, Brunel University 36 ME1302 Analytical Methods and Skills
Statistical Analysis of Experimental Data
Example
• We would like to determine the confidence interval of
the mean of a batch of resistors made in a certain
Confidence Confidence Confidence Confidence process.
level coefficient level coefficient
99% 2.58 68.26% 1 • Based on 36 samples, the average resistance is 25Ω. and
98% 2.33 95.44% 2 the sample standard deviation is 0.5Ω.
97% 2.17 99.74% 3
96% 2.05 99.9% 3.3 • Determine the 90% confidence interval of the mean
95% 1.96 99.96% 3.5 resistance of the batch.
90% 1.645
80% 1.28 70% 1.036 μ = x ±δ σ
• Solution: δ = zc
50% 0.6745 60% 0.842 n
x = 25Ω S = 0.5Ω
S
μ = x ± zc • The desired confidence level is 90%:
n
1-α=0.90 and α=0.1
© Dr, J. Chen, School of Engineering and Design, Brunel University 37 ME1302 Analytical Methods and Skills
Statistical Analysis of Experimental Data
f(v,t)
Standard Normal
Use Standard Normal Distribution Table
Student's t 0.2
Distribution Table to P( 0 < z < zc ) distribution: f(t, ν).
find confidence
ν = n −1
coefficient: 0
zc • Degree of freedom -3 -2 -1 0 t 1 2 3
zc z
© Dr, J. Chen, School of Engineering and Design, Brunel University 38 ME1302 Analytical Methods and Skills
Statistical Analysis of Experimental Data
1-α=90% ; n=15
21 1.323 1.721 2.080 2.518 2.831
22 1.321 1.717 2.074 2.508 2.819
23 1.319 1.714 2.069 2.500 2.807
α=0.1; v=n-1=14 24
25
26
1.318
1.316
1.315
1.711
1.708
1.706
2.064
2.060
2.056
2.492
2.485
2.479
2.797
2.787
2.779
27 1.314 1.703 2.052 2.473 2.771
α
tc = 1.761
28 1.313 1.701 2.048 2.467 2.763
= 0.05 29 1.311 1.699 2.045 2.462 2.756
2 30 1.310 1.697 2.042 2.457 2.750
∞ 1.283 1.645 1.960 2.326 2.576
S=
• Using the t-distribution table:
α
(1250-1372) 2 +(1320-1372) 2 +(1542-1372) 2 +(1464-1372) 2 +(1275-1372) 2 +(1383-1372)2 ν = n −1= 5 ; = 0.025 ; tc = 2.571
6-1 2
64482 S 114
=
5
= 114 h δ = tc = 2.571 × = 120 μ = x ± δ = 1372 ± 120h (95%)
n 6
© Dr, J. Chen, School of Engineering and Design, Brunel University 39 ME1302 Analytical Methods and Skills
Statistical Analysis of Experimental Data
© Dr, J. Chen, School of Engineering and Design, Brunel University 40 ME1302 Analytical Methods and Skills
Statistical Analysis of Experimental Data
Uncertainty
• As the uncertainty is an estimate, the confidence level is
Uncertainty and Percentage Uncertainty
normally associated with the uncertainty. • Uncertainty: V = 10 (V ) wV = 1 (V )
• General, a variable to be measured can be described as:
Voltage V 10 1 V 95
x x wx P
• In some cases, the uncertainty is quoted as
x Measured (or estimated) value. percentage or fractional uncertainty:
wx Uncertainty. P Confidence level.
wx
fx
x
Resistance is: 25 ± 0.14 Ω (90%)
Voltage example: wV 1
VCR average life: 1372 ± 120 h (95%) fV = = = 10%
V 10
© Dr, J. Chen, School of Engineering and Design, Brunel University 41 ME1302 Analytical Methods and Skills
Statistical Analysis of Experimental Data
© Dr, J. Chen, School of Engineering and Design, Brunel University 42 ME1302 Analytical Methods and Skills
Statistical Analysis of Experimental Data
© Dr, J. Chen, School of Engineering and Design, Brunel University 43 ME1302 Analytical Methods and Skills
Statistical Analysis of Experimental Data
© Dr, J. Chen, School of Engineering and Design, Brunel University 44 ME1302 Analytical Methods and Skills
Statistical Analysis of Experimental Data
wm = ⎜ wx ⎟ + ⎜⎜ w y ⎟⎟ + ⎜ wz ⎟
⎝ ∂x ⎠ ⎝ ∂y ⎠ ⎝ ∂z ⎠ • Uncertainty: wy = 2 + 0.5
2 2
= 2.062
– Uncertainty.
• The fractional (percentage) uncertainty. • Result: y = 210 ± 2 062
wm
fm = ; m = f ( x , y, z )
m
wP = ⎜ wv ⎟ + ⎜ wI ⎟
• The maximum possible error in P? ⎝ ∂V ⎠ ⎝ ∂I ⎠
• The best estimation of uncertainty in P?
= 10 × 2 + 100 × 0.2 = 28.3W
2 2
• Requirement:
P = P ± w p = 1000 ± 28 3 W – to achieve an accuracy of 0.1% of the maximum reading, 10000Pa.
© Dr, J. Chen, School of Engineering and Design, Brunel University 46 ME1302 Analytical Methods and Skills
Statistical Analysis of Experimental Data
P 10000
h= = = 0.408 m
ρ g 2500 × 9.81 ∂P ∂ d
= {ghρ} = gh {ρ} = gh = 9.81 × 0.408
The maximum uncertainty in ∂ρ ∂ρ dρ
wP = 10 kPa × 0.1% = 10.0Pa
pressure is 0.1% of 10 kPa = 4.00 m 2 / s 2
wh = 0.1 mm = 0.0001 m
∂P ∂ d
= {gρh} = gρ {h} = gρ = 9.81 × 2500.0
• Problem to solve: find wρ wρ ?? ∂h ∂h dh
= 24525.0 kg / m 2 s 2
wρ = 2.4 kg / m 3
• Problem Type 1: R = R1 + R2
m = ax + by + cz
• The maximum allowable uncertainty in the gage fluid
density is f mnD
• Problem Type 2: K=
fρ =
wρ
=
2.4
≈ 0.1%
d2 G
± 2.4 kg / m 3 or
ρ 2500
m = kx a y b z c
© Dr, J. Chen, School of Engineering and Design, Brunel University 47 ME1302 Analytical Methods and Skills
Statistical Analysis of Experimental Data
© Dr, J. Chen, School of Engineering and Design, Brunel University 48 ME1302 Analytical Methods and Skills
Statistical Analysis of Experimental Data
wm = ⎜ wx ⎟ + ⎜⎜ w y ⎟⎟ + ⎜ wz ⎟ m = kx a y b z c
⎝ ∂x ⎠ ⎝ ∂y ⎠ ⎝ ∂z ⎠ P = 1000; wP = 10
fm = + bf y + cf z
2 2 2
2 h = 0.408; wh = 0.0001 af x
m⎞ ⎛ m⎞ ⎛
2 2
⎛ m⎞
= ⎜ awx ⎟ + ⎜⎜ bw y ⎟⎟ + ⎜ cwz ⎟ g = 9.81; ρ = 2500
⎝ x⎠ ⎝ y⎠ ⎝ z⎠
wρ = ? fP = 1× fρ
2
+ 1 × fh
2
2
⎛ w ⎞ ⎛ w ⎞ ⎛ w ⎞
2 2
= m ⎜ a x ⎟ + ⎜⎜ b y ⎟⎟ + ⎜ c z ⎟ 2 f P2 = f ρ2 + f h2
⎝ x ⎠ ⎝ y ⎠ ⎝ z ⎠ ⎛ wP ⎞ ⎛ wρ ⎞ ⎛ wh ⎞
2 2
⎜ ⎟ = ⎜⎜ ⎟⎟ + ⎜ ⎟
⎝ P ⎠ ⎝ ρ ⎠ ⎝ h ⎠
fm = af x
2
+ bf y + cf z
2 2
wm = mf m wρ = 2.4 kg / m 3
© Dr, J. Chen, School of Engineering and Design, Brunel University 49 ME1302 Analytical Methods and Skills
Statistical Analysis of Experimental Data
Example Example
• A certain obstruction-type flowmeter (orifice, venturi, nozzle) is C = 0.92 ± 0.005 p1 = 25 ± 0.5 Δp = 1.4 ± 0.005
used to measure the flow of air at low velocities. The relation
describing the flow rate is:
T1 = 530 ± 15 A = 1.0 ± 0.001 gc = 32.17 R = 53.35
2 g c p1 p
v CA 2 g c p1 p
RT1 v CA
RT1
C: empirical-discharge coefficient A: flow area
p1: upstream pressure R: gas constant for air
2 32.17 25 1.4
0.92 1 0.2596
Δp: the difference between upstream and downstream pressures 53.35 530
1 1 1
• Experiment results: C = 0.92 ± 0.005 2 g c p1 p 2 gc −
p1 = 25 ± 0.5 v CA CA( p1 ) ( p ) (T1 ) 2
2 2
Example Example
2 2 1 2 1 2 1 2 2 21 2 1 2 1 2
fv fC fA fp f Δp fT fv fC fA fp f Δp fT
4 1 4 4 1 4 1 4 4 1
C = 0.92 ± 0.005 p1 = 25 ± 0.5 Δp = 1.4 ± 0.005 0.01827 1.827%
Practice Questions
• Question 1: Suppose you measure three numbers as
x=200±2, y=50 ± 2, z=20 ± 1. The three variables are
independent and random. What you would give for the
values of q=x+y-z and r=xy/z with their uncertainties?
© Dr, J. Chen, School of Engineering and Design, Brunel University 51 ME1302 Analytical Methods and Skills