FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

MATLAB
for Engineering Applications
Fourth Edition

Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 1

 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

Chapter 06

Model Building and Regression

Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 2

 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

Model Building and Regression

An important application of the plotting techniques is function discovery, the technique

for using data plots to obtain a mathematical function or “mathematical model” that
describes the process that generated the data.

Function Discovery: A systematic way of finding an equation that best fits the data is
regression (also called the least-squares method).

Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 3

 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

Using the Linear, Power, and Exponential

Functions to Describe data

Each function gives a straight line when plotted using a specific set of axes:
1. The linear function y = mx + b gives a straight line when plotted on
rectilinear axes. Its slope is m and its intercept is b.
2. The power function y = bxm gives a straight line when plotted on log-log
axes.
3. The exponential function y = bemx and its equivalent y = b(10)mx form give
a straight line when plotted on a semilog plot whose y-axis is logarithmic.

Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 4

 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

Function Discovery
The power function y = 2x −05 and the exponential function y = 101−x plotted on
linear, semi-log, and log-log axes.

Linear Semi-log

Log-Log

Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 5

 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

Steps for Function Discovery

1. Examine the data near the origin. The exponential function y = b(10)mx or y = b(e)mx
can never pass through the origin (unless of course b = 0, which is a trivial case).

Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 6

 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

Steps for Function Discovery

The linear function y = mx + b can pass through the origin only if b = 0. The power
function y = bxm can pass through the origin but only if m > 0.

Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 7

 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

Steps for Function Discovery 1

2. Plot the data using rectilinear scales. If it forms a straight line, then it can be
represented by the linear function and you are finished. Otherwise, if you have
data at x = 0, then

a. If y(0) = 0, try the power function.

b. If y(0) ≠ 0, try the exponential function.
If data is not given for x = 0, proceed to step 3.

Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 8

 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

Steps for Function Discovery 2

3. If you suspect a power function, plot the data using log-log scales. Only a power
function will form a straight line on a log-log plot. If you suspect an exponential
function, plot the data using the semilog scales. Only an exponential function will
form a straight line on a semilog plot.

4. In function discovery applications, we use the log-log and semilog plots only to
identify the function type, but not to find the coefficients b and m. The reason is that
it is difficult to interpolate on log scales.

Power function y = bxm

Exponential function y = b(10)mx

Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 9

 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

The polyfit Function 1

Command Description
p = polyfit(x,y,n) Fits a polynomial of degree n to data described by
the vectors x and y, where x is the independent
variable. Returns a row vector p of length n + 1
that contains the polynomial coefficients in order of
descending powers.

𝑝 𝑥 = 𝑝1 𝑥 𝑛 + 𝑝2 𝑥 𝑛−1 + ⋯ + 𝑝𝑛 𝑥 + 𝑝𝑛+1

Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 10

 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

Using polyfit to Fit Equations to Data

Syntax: p = polyfit(x,y,n)

where x and y contain the data, n is the order of the polynomial to be fitted, and p is
the vector of polynomial coefficients.

The linear function: y = mx + b.

We can find the linear function that fits the data by typing p = polyfit(x,y,1). The
first element p1 of the vector p will be m, and the second element p2 will be b.

Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 11

 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

Example: Fitting a Linear Function

Speed Estimation from Sonar Range Measurements

Sonar measurements of the range of an approaching underwater vehicle are given in the
following table, where the distance is measured in nautical miles (𝑛𝑚𝑖). Assuming the relative
speed v is constant, the range as function of time is given by 𝑟 = −𝑣𝑡 + 𝑟0 where is 𝑟0 the
initial range at 𝑡 = 0. Estimate (a) the speed v and (b) when the range will be zero.

Time, t (min) 0 2 4 6 8 10
Range, r (nmi) 3.8 3.5 2.7 2.1 1.2 0.7

Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 12

 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

Example :Fitting a Linear Function

Time , t (min) 0 2 4 6 8 10
Range ,r (nmi) 3.8 3.5 2.7 2.1 1.2 0.7

t = 0:2:10;
r = [3.8,3.5,2.7,2.1,1.2,0.7];
% First-order curve fit.
p = polyfit(t,r,1)
% Create plotting variable.
y = p(1)*t+p(2);
plot(t,r,'o',t,y),xlabel('t (min)'),ylabel('y (nmi)')
% Speed calculation.
v = -p(1)*60 % speed in knots (nmi/hr)
------------------------

p =-0.3286 3.9762 y = −0.3286t + 3.9762.

v = 19.7143 y=0 at t = 12.1 minutes

Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 13

 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

Example: Fitting an Exponential Function1

Temperature of a Cooling Cup of Coffee

The temperature of coffee cooling in a mug at room temperature (68°F) was measured at various
times. The data follow.

Time t (sec) Temperature T (°F)

0 145
620 130
2266 103
3482 90

Develop a model of the coffee’s temperature as a function of time, and use the model to estimate
how long it took the temperature to reach 120°F.

Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 14

 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

Example: Fitting an Exponential Function2

Temperature of a Cooling Cup of Coffee

Time t (sec) Temperature T (°F)

0 145
620 130
2266 103
3482 90

1. Because T(0) is finite but nonzero, the power function cannot describe these data. So we do
not bother to plot the data on log-log axes.

2. Common sense tells us that the coffee will cool, and its temperature will eventually equal the
room temperature. So, we subtract the room temperature from the data and plot the relative
temperature, T − 68, versus time.
• If the relative temperature is a linear function of time, the model is 𝑇 − 68 = 𝑚𝑡 + 𝑏.
• If the relative temperature is an exponential function of time, the model is 𝑇 − 68 =
𝑏(10)𝑚𝑡 .
Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 15
 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

Example: Fitting an Exponential Function3

Temperature of a Cooling Cup of Coffee

time = [0, 620, 2266, 3482]; Linear

temp = [145, 130, 103, 90];
subplot(2,1,1);
temp = temp -68; % Subtract the room temperature.
plot(time, temp);
hold on
plot(time, temp,'o');
subplot(2,1,2);
semilogy(time, temp); Semilog
hold on
semilogy(time, temp,'o');

𝑇 = 68 + 𝑏(10)𝒎𝑡
Temperature of a cooling cup of coffee,
plotted on various coordinates
Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 16
 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

Example: Fitting an Exponential Function4

Temperature of a Cooling Cup of Coffee

% Enter the data.

Time t (sec) Temperature T (°F)
time = [0,620,2266,3482]; 0 145
temp = [145,130,103,90]; 620 130
% Subtract the room temperature. 2266 103
temp = temp - 68; 3482 90
p = polyfit(time,log10(temp),1);
m = p(1)
b = 10^p(2) % logb=p(2)
𝑇 = 68 + 𝑏(10)𝑚𝑡
---------------
m = -1.5557e-04 To estimate how long it will take for the coffee to cool to 120°F, we
b = 77.4469 must solve the equation 120 = 68 + (−1.56e−04(10)77.4t ) for t.

The time to reach 120°𝐹 is 𝑡 = 1112 sec

Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 17

 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

Q1-The distance a spring stretches from its “free length” is a function of how much tension force
is applied to it. The following table gives the spring length y that the given applied force f
produced in a particular spring. The spring's free length is 4.7 in. Find a functional relation
between f and x, the extension from the free length, 𝑓 = 𝑘𝑥 + 𝑏. The spring constant k is:

Force f (lb) Spring length y (in.)

0 4.7
0.94 7.2
2.30 10.6
3.28 12.9

1. 0.1
2. 0.2
3. 0.3 Password: polynomial
4. 0.4

Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 18

 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

Q2. The population data for a certain country are as follows:

Year 2012 2013 2014 2015 2016 2017

Population (millions) 10 10.9 11.7 12.6 13.8 14.9

Obtain a function that describes these data. How may years after 2012 the population will be
double its 2012 size.

1. 5.9 years
2. 8.7 years
3. 12.3 years
4. 14.8 years

Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 19

 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

Q3-A 15-cup coffee pot was placed under a water faucet and filled to the 15-cup
line. With the outlet valve open, the faucet’s flow rate was adjusted until the
water level remained constant at 15 cups, and the time for one cup to flow out
of the pot was measured. This experiment was repeated with the pot filled to
the various levels shown in the following table:

Liquid volume V (cups) Time to fill 1 cup t (sec)

6 9
9 8
12 7
15 6

(a) Use a power function, log10 V = m log10(1/t) + log10b , to obtain a linear relation
between the flow rate (1/t) and the number of cups in the pot (V). M and b are:

1. m= 0.4331, b= 0.0499
2. m= 1.3254, b= 1.1545
3. m= 4.9873, b= 6.5343
4. m= 5.4754, b= 2.3451
Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 20
 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

The Least Squares Criterion

The Least Squares Criterion: used to fit a function f (x). It minimizes the
sum of the squares of the residuals, J. J is defined as

m
J =   f ( xi ) − yi 
2

i =1

We can use this criterion to compare the quality of the curve fit for two or
more functions used to describe the same data. The function that gives the
smallest J value gives the best fit.

Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 21

 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

Illustration of the least squares criterion

Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 22

 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

The polyfit function is based on the least-squares

method

p = polyfit(x,y,n) Fits a polynomial of degree n to data

described by the vectors x and y, where x is
the independent variable. Returns a row
vector p of length n + 1 that contains the
polynomial coefficients in order of
descending powers.

𝑝 𝑥 = 𝑝1 𝑥 𝑛 + 𝑝2 𝑥 𝑛−1 + ⋯ + 𝑝𝑛 𝑥 + 𝑝𝑛+1

Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 23

 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

Example: Regression using Polynomials

Consider the data set where 𝑥 = 1, 2, 3, . . . , 9 and 𝑦 = 5, 6, 10, 20, 28, 33, 34, 36, 42. Fit
polynomials of first through fourth degree to this data and compare the results.

x = 1:9;
y = [5,6,10,20,28,33,34,36,42];
for k = 1:4
coeff = polyfit(x,y,k);
subplot (2,2,k);
plot(x,y, 'O');
hold on
plot(x,polyval(coeff,x));
J(k) = sum ((polyval(coeff,x)-y).^2)
end

J = 71.5389 56.6727 41.8838 4.6566

Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 24
 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

Beware of using Polynomials of High Degree

This fifth-degree polynomial passes through all six data points but exhibits large
excursions between points.

Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 25

 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

Q4 -The plot of a fourth-degree polynomial for the following data is:

𝑥 = 0, 1, . . . , 5 and 𝑦 = 0, 1, 44, 40, 41, 47

1- 3-

4-
2-

Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 26

 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

Assessing the Quality of a Curve Fit2

The coefficient of determination, also known as the r-squared value, is
defined as

m m
J
J =   f ( xi ) − yi  S =  ( yi − y )
2
r = 1−
2 2

i =1 i =1 S

The value of S indicates how much the data is spread around the mean 𝑦, ത and
the value of J indicates how much of the data spread is unaccounted for by the
model.
Thus the ratio J/S indicates the fractional variation unaccounted for by the
model.
Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 27
 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

Q5 -The J value for the second-degree polynomial representing the following data is 𝑥 =
0, 1, . . . , 5 and 𝑦 = 0, 1, 44, 40, 41, 47

m first-degree second-degree
J =   f ( xi ) − yi 
2

i =1

1. 1485.7
2. 1151.7
3. 73.161 fourth-degree
third-degree
4. 1333.7

Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 28

 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

Effect of Coefficient Accuracy

14 decimal-place accuracy

8 decimal-place accuracy

• high-degree polynomials can produce large errors if their coefficients are not represented
with a large number of significant figures.
• The effect of computational errors in computing the coefficients can be lessened by properly
scaling the x values.
Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 29
 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

Using Residuals
High Residuals Low Residuals No Clear Pattern

Low Residuals with a Pattern High Residuals

In general, if you see a pattern in the plot of the residuals, it indicates that another function can
be found to describe the data better.
Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 30
 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

Assessing the Quality of a Curve Fit3

For a perfect fit, J = 0 and thus r 2 = 1. Thus the closer r 2 is to 1, the better the fit. The
largest r 2 is 1.

It is possible for J to be larger than S, and thus it is possible for r 2 to be negative. Such
cases, however, are indicative of a very poor model that should not be used.

As a rule of thumb, a good fit accounts for at least 99 percent of the data variation. This
value corresponds to r 2 ≥ 0.99.

Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 31

 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

Scaling the Data

The effect of computational errors in computing the coefficients can be lessened by

properly scaling the 𝑥 values. You can scale the data yourself before using polyfit.
Some common scaling methods are

1. Subtract the minimum 𝑥 value or the mean 𝑥 value from the 𝑥 data, if the range of 𝑥
is small, or

2. Divide the 𝑥 values by the maximum value or the mean value, if the range is large.

Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 32

 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

Example: Scaling the Data

The following data give the number of vehicles (in millions) crossing a bridge each year for 10
years. Fit a cubic polynomial to the data and use the fit to estimate the flow in the year 2010.

Year 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009
Vehicle flow(millions) 2.1 3.4 4.5 5.3 6.2 6.6 6.8 7 7.4 7.8

Year = 2000:2009;
Veh_Flow = [2.1,3.4,4.5,5.3,6.2,6.6,6.8,7,7.4,7.8];
p = polyfit(Year,Veh_Flow,3);
Warning: Polynomial is badly conditioned. Add points with distinct X values,
reduce the degree of the polynomial, or try centering and scaling as described
in HELP POLYFIT.
> In polyfit (line 79)

The problem is caused by the large values of the independent variable Year

Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 33

 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

Example: Scaling the Data

The following data give the number of vehicles (in millions) crossing a bridge each year for 10
years. Fit a cubic polynomial to the data and use the fit to estimate the flow in the year 2010.
Year 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009
Vehicle flow(millions) 2.1 3.4 4.5 5.3 6.2 6.6 6.8 7 7.4 7.8

Year = 2000:2009;
Veh_Flow = [2.1,3.4,4.5,5.3,6.2,6.6,6.8,7,7.4,7.8];
x = Year-2000; y = Veh_Flow;
Veh_Flow = [2.1,3.4,4.5,5.3,6.2,6.6,6.8,7,7.4,7.8];
p = polyfit(x,y,3);
J = sum((polyval(p,x)-y).^2);
S = sum((y-mean(y)).^2);
r2 = 1 - J/S
plot(x, y, 'O', x, polyval(p,x))
r2 = 0.9972
𝑓 = 0.0087(𝑡 − 2000)3 − 0.1851(𝑡 − 2000)2 + 1.5991(𝑡 − 2000) + 2.0362
Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 34
 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

Q6- The U.S. census data from 1790 to 1990 is stored in the file census. dat, which is supplied
with MATLAB. Type load census to load this file. The first column, cdate, contains the years, and
the second column, pop, contains the population in millions. First try to fit a cubic polynomial to
the data. If you get a warning message, scale the data by subtracting 1790 from the years, and
fit a cubic. The coefficient of determination (r-squared value) is:

load census load census

x = cdate; x = cdate-1790;
y = pop;
a) 2.8945
p = polyfit(x,y,3)
b) 1.1324 Warning: Polynomial is badly conditioned. Add
points with distinct X values,
c) 0.9988
reduce the degree of the polynomial, or try
d) 0.7837 centering and scaling as described
in HELP POLYFIT.
> In polyfit (line 79)

Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 35

 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

Q7- The U.S. census data from 1790 to 1990 is stored in the file census. dat, which is supplied
with MATLAB. Type load census to load this file. The first column, cdate, contains the years, and
the second column, pop, contains the population in millions. First try to fit a cubic polynomial to
the data. If you get a warning message, scale the data by subtracting 1790 from the years, and fit
a cubic. The estimate population in 1965 is.

a) 154.1 millions
b) 162.2 millions
c) 178.3 millions
d) 189.4 millions

Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 36

 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

Q8- Load the census data. Use the scaled data and try three linear, quadratic, cubic, and an
exponential fits. Then plot the residuals. Which one is the better fit.

Exponential fit: polyfit(x,log10(y),1)

a) Linear
b) Quadratic
c) Cubic
d) Exponential

Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 37

 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

Q9-. The following function is linear in the parameters 𝑎1 and 𝑎2 .

𝑦( 𝑥 ) = 𝑎1 + 𝑎2 ln(𝑥) .
Use least-squares regression with the following data to estimate the values of 𝑎1 and 𝑎2 .

x 1 2 3 4 5 6 7 8 9 10
y 10 14 16 18 19 20 21 22 23 23

a) 𝑎1 = 14.4353 𝑎2 = 2.4543
b) 𝑎1 = 7.34764 𝑎2 = 6.6895
c) 𝑎1 = 5.7518 𝑎2 = 9.9123
d) 𝑎1 = 3.4344 𝑎2 = 11.472

Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 38

 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

Multiple Linear Regression

Suppose that y is a linear function of two or more variables 𝑥1, 𝑥2, . . . ,

for example, 𝑦 = 𝑎0 + 𝑎1 𝑥1 + 𝑎2 𝑥2 .

How to find the coefficient values 𝑎0 , 𝑎1 , and 𝑎2 to fit a set of data (𝑦, 𝑥1 , 𝑥2 ) in the least-
squares sense?

Xa = y

The solution for the coefficients is given by a = X\y.

Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 39

 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

Example: Multiple Linear Regression

Obtain a linear model 𝑦 = 𝑎0 + 𝑎1 𝑥1 + 𝑎2 𝑥2 to describe the relationship.

Breaking strength (kN) % of element 1 % of element 2

y x1 x2
1 0 5 7.1 0 5
1 1 7 19.2 1 7
1 2 8 31 2 8
1 3 11 45 3 11

1 0 5 𝑎0 7.1
1 1 7 𝑎1 = 19.2
Xa = y 1 2 8 31
𝑎2
1 3 11 45

Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 40

 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

Example: Multiple Linear Regression

Obtain a linear model 𝑦 = 𝑎0 + 𝑎1 𝑥1 + 𝑎2 𝑥2 to describe the relationship.
Breaking strength (kN) % of element 1 % of element 2
x1 = (0:3)'; x2 = [5,7,8,11]'; y x1 x2
y = [7.1,19.2,31,45]'; 7.1 0 5
X = [ones(size(x1)), x1, x2]; 19.2 1 7
a = X\y 31 2 8
yp = X*a; 45 3 11
Max_Percent_Error = 100*max(abs((yp-y)./y))
plot3(x1,x2,yp, 'O')
hold on
plot3(x1,x2,y)
grid on

a = 0.8000
10.2429
1.2143
Max_Percent_Error = 3.2193

Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 41

 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

Linear-in-Parameters Regression

Sometimes we want to fit an expression that is neither a polynomial nor a

function that can be converted to linear form by a logarithmic or other
transformation.
In some cases we can still do a least-squares fit if the function is a linear
expression in terms of its parameters.

Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 42

 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

Example: Linear-in-Parameters Regression

The following data give the output voltage of a certain device as a function of time.
Obtain a function that describes these data.

𝑣(𝑡) = 𝑎1 + 𝑎2 𝑒 −3𝑡/𝑇 (first-order model)

The first-order model written for each of the n data points results in n equations, which can be
expressed as follows:

Xa = y′

Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 43

 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

Linear-in-Parameters Regression

𝑦(𝑡) = 𝑎1 + 𝑎2 𝑒 −3𝑡/𝑇 (first-order model)

t = [0,0.3,0.8,1.1,1.6,2.3,3];
y = [0,0.6,1.28,1.5,1.7,1.75,1.8];
X = [ones(size(t));exp(-t)]';
a = X\y'
plot(t, y, 'O')
hold on
yp=a(1)+a(2).*exp(-t);
plot(t,yp);

Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 44

 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

Q10- The following function is linear in the parameters 𝑎1 and 𝑎2 .

𝑦( 𝑥 ) = 𝑎1 + 𝑎2 ln(𝑥) .
Use least-squares regression with the following data then use the curve fit to estimate
the values of y at x = 3.5.
x 1 2 3 4 5 6 7 8 9 10
y 10 14 16 18 19 20 21 22 23 23

a) y = 12.4734
b) y = 17.1179
c) y = 23.4754
d) y = 34.7321

Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 45

 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

Constraining the Curve to Pass Through

a Specified Point

Suppose the model is required to pass through a point not at the origin, say the point
(𝑥0, 𝑦0) and that point is known to be an exact solution to the equation, so that

𝑦0 = 𝑚𝑥0 + 𝑏

In that case, simply subtract 𝑥0 from all the 𝑥 values, subtract 𝑦0 from all the 𝑦 values.

𝑢 = 𝑥 − 𝑥0 and 𝑤 = 𝑦 − 𝑦0.

The resulting equation will be of the form 𝑤 = 𝑚𝑢, and the coefficient 𝑚 can be calculated
using right division. In MATLAB we would write 𝑚 = 𝑤’\u’ where 𝑢 and 𝑤 are row vectors
containing the transformed data.

Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 46

 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

Basic Fitting Interface

MATLAB supports curve fitting through the Basic Fitting interface. Using this
interface, you can quickly perform basic curve fitting tasks within the same easy-to-
use environment. The interface is designed so that you can:

• Fit data using a cubic spline or a polynomial up to degree 10.

• Plot multiple fits simultaneously for a given data set.
• Plot the residuals.
• Examine the numerical results of a fit.
• Interpolate or extrapolate a fit.
• Annotate the plot with the numerical fit results and the norm of residuals.
• Save the fit and evaluated results to the MATLAB workspace.

Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 47

 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

A figure produced by the Basic Fitting interface

Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 48

 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

The Basic Fitting Interface

x 1 2 3 4 5 6 7 8 9 10
y 10 14 16 18 19 20 21 22 23 23

load census
x = cdate;
y = pop;
plot(x,y,'O')

Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 49

 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

Q11- A mass attached to a spring and damper is displaced a distance x0 (cm) while being given
an initial velocity v0 (cm/s). We know from physics and mathematics (see Chapter 8) that the
displacement x as a function of time is given by
5𝑥0 𝑣0 −2𝑡 2𝑥0 + 𝑣0 −5𝑡
𝑥 𝑡 = + 𝑒 − 𝑒
3 3 3
The displacement is measured every 0.2 s. The measured displacement versus time is given by

t (s) 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 2

x (cm) 1.4 2.6 1.6 1.2 1 0.6 0.4 0.3 0.2 0.1 0.1

The estimated the initial displacement is:

a) x0 = 1.5268
b) x0 = 3.4378
c) x0 = 5.3298
d) x0 = 7.1983
Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 50
 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

Q12- The U.S. census data from 1790 to 1990 is stored in the file census. dat, which is supplied
with MATLAB. Type load(census) to load this file. The first column, cdate, contains the years,
and the second column, pop, contains the population in millions. Use the Basic Fitting interface
to solve this problem. First try to fit a cubic polynomial to the data. If you get a warning
message, center and scale the data by checking Center and scale x data in the Interface, and fit
a fourth degree polynomial 𝑝1 𝑧 4 + 𝑝2 𝑧 3 +𝑝3 𝑧 2 +𝑝4 𝑧 + 𝑝5 . The coefficient, 𝑝1 is

1. 0.6
2. 0.7
3. 0.8
4. 0.9

Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 51

 FACULTY OF ENGINEERING - UNIVERSITY OF WINDSOR

Q13- Consider the following data. Find the best-fit line, 𝑦𝑜 = 𝑚𝑥𝑜 + 𝑛, that passes through
the point 𝑥𝑜 = 10 and 𝑦𝑜 = 11. 𝑚 and 𝑛 are:

x 0 5 10
y 2 6 11

1. m=1.1673, n=0.4234
2. m=1.2567, n=0.6575
3. m=1.0849, n=0.1509
4. m=1.3443, n=0.1172

Numerical Methods and Modeling for Engineering Applications R. Rashidzadeh 52