-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathModalAnalysis.html
269 lines (197 loc) · 13.7 KB
/
ModalAnalysis.html
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
<!DOCTYPE html>
<html>
<head>
<script src="javascripts/flot/jquery.js"></script>
<script src="javascripts/jqmath/jqmath-etc-0.2.0.min.js"></script>
<link rel="stylesheet" href="javascripts/jqmath/jqmath-0.2.0.css">
<link rel="stylesheet" type="text/css" media="screen" href="stylesheets/style.css">
<title>Modal Analysis</title>
</head>
<body>
<div id="main">
<div id="menu">
<a href="https://github.com/llarsen71/GMA1D"> Git Repository </a>
</div><!--menu-->
<div id="title">
<h1>Generalizing Modal Analysis</h1>
</div><!--title-->
<p><a href="index.html" style="text-decoration: none;">← Previous Topic: Why Mathematical Modeling?</a>. </p>
<h2>Classical Modal Analysis</h2>
<p>For linear systems, modal analysis is an elegant blend of quantitative and qualitative techniques. It accurately predicts
linear system response, but also characterizes the general behavior of the linear system. Modal analysis is built on
simple idea - find a new coordinate system where the original system has a simpler form that is easier to understand.
It turns out that for linear systems, where the linear coefficients are constant, there is always a coordinate system
where the system of ODEs has a simple form called the Jordan cannonical form. In many cases, when coordinates are
selected such that the system is in Jordan cannonical form, all of the ODEs in the system are decoupled, which means
they can be solved independently. In some cases, referred to as degenerate cases, there is a simple coupling between
some of the ODEs in the system. Only the more common nondegenerate cases are discussed below.</p>
<h4>Example System</h4>
<div class="box-right" style="margin-top: 0em;">
<img src="images/Mass-Spring-Damper-2-body.png" width="440 px">
</div><!--box-right-->
<p>As an example consider a linear two mass system with springs connecting the masses to each other an to the walls, and a
damper between each mass and the wall. To model the system, Newtons law ($F=ma$) can be applied to a free body diagram of
each of the masses. The equations for the masses are: </p>
<p>$m{d^2x_1}/{dt^2} = -c{dx_1}/{dt} - k x_1 + k(x_2-x_1)$<br>
$m{d^2x_2}/{dt^2} = -c{dx_2}/{dt} - k x_2 + k(x_1-x_2)$</p>
<p>If this is simplified by assuming that the spring constant (k) and the damper constant (c) each have a value of 1, and that
the mass (m) is small enough to be ignored, then the following system of equations predicts the motion of the two masses:</p>
<p>${dx_1}/{dt} = -2 x_1 + x_2$<br>
${dx_2}/{dt} = x_1 - 2 x_2$</p>
<p> For the system, both $x_1$ and $x_2$ change with $t$ (which in this system represents time). This is a coupled system
because the change in $x_1$ with respect to $t$ depends on the value of both $x_1$ and $x_2$. Similarly, the change in
$x_2$ with respect to $t$ depends on both $x_1$ and $x_2$. Since the equations depend on one another, both must be solved
simultaneously.</p>
<p>If the terms $a_{ij}$ are constants, a linear change of coordinates has the form:</p>
<p>$q_1 = a_{11} x_1 + a_{12} x_2$<br>
$q_2 = a_{21} x_1 + a_{22} x_2$</p>
<p>With the right choice of coordinates, the system above can be represented in the uncoupled form:</p>
<p>${dq_1}/{dt} = -q_1$<br>
${dq_2}/{dt} = -3 q_2$</p>
<p>In this form, the change in $q_1$ with respect to $t$ only depends on $q_1$ so this can be solved independent of $q_2$.
Similarly, the change in $q_2$ with respect to $t$ only depend on $q_2$ so this can be solved independent of $q_1$.</p>
<h2>What is Model Analysis</h2>
<p>Modal Analysis the process of finding the linear choice of coordinates where the ODEs of the system are decoupled
from one another. The new coordinates are called modal coordinates.</p>
<div class="box-right" style="margin-top: 0em;">
<h1>Definition</h1>
<b>Modal Coordinate</b> - A linear function $q(x_1,x_2,...,x_n)$ that when used as a coordinate for a system of ODEs
results in a decoupled ODE. The ODE will have the linear form ${dq}/{dt} = λ q$.
</div><!--box-right-->
<p>For linear systems, the quantitative accuracy of modal analysis comes from the fact that the new modal system is
the same as the original system, or in other words no approximation is required. The only difference is the
coordinate system that is used. The qualitative understanding comes from the fact that modal analysis converts
coupled ODEs that are typically too complex to understand as a whole, because the solutions are intertwined, to a
decoupled set of ODEs which are simpler to understand because the modal ODEs can be solved and explored independently.
The modal solutions can be added (or superimposed) to get the total system response. The power of modal analysis
comes from the fact that you can understand the total behavior of the system in terms of the individual behavior of the modes.<p>
<p>For nonlinear systems, there is often a region around an equilibrium or steady state solution were a linear system
of ODEs provides a good approximation and modal analysis gives reasonably accurate results. However, there are many
cases where a linear approximation misses important trends, and cases where the response grows to the point where
nonlinear terms are important, and modal analysis gives poor results. So while modal analysis will remain an
important tool for the study of linear and nonlinear systems, there is a need to generalize the approach so that the
elegant blend of quantitative and qualitative analysis are preserved.</p>
<h2>Calculating the Modal Equations</h2>
<h3>First Order Form</h3>
<p>Before generalizing modal analysis to nonlinear systems, lets take a look at how modal coordinates, which
decouple a linear system of ODEs, are calculated for linear systems. We will start with a system of ODEs expressed
as a matrix in first order form. This is a way of representing the system of ODEs where only first order derivatives
are part of the system of ODEs. <a href="FirstOrderForm.html">Click here</a> for a discussion of how to convert a
system with higher order derivates to first order form. A linear system in first order form, with no forcing terms
can be written like:</p>
<p>(1) ${dx_i}/{dt} = A_{ij} x_j$</p>
<p>As noted above, in order for modal analysis to work, the value of the coefficients in $A_{ij}$ must not change with $t$.
In other words, the coeffients can be constant values, or constant parameters such as a mass or stiffness quantity, but can't
depend on $t$ (or $x_1, x_2, ..., x_n$, since these values change with $t$). We will explain why the coefficients
can't be dependent on $t$ in the discussion below. The linear modal coordinate function $q$ can be written in the form:</p>
<p>(2) $q = c_i x_i = c_1 x_1 + c_2 x_2 + ... + c_n x_n$ </p>
<p>where $c_i$ is a vector with values that do not depend on $t$.</p>
<h4>Example System</h4>
<p>Consider the example system shown above:</p>
<p>${dx_1}/{dt} = -2 x_1 + x_2$<br>
${dx_2}/{dt} = x_1 - 2 x_2$</p>
<p>In Matrix Form this is represented as:</p>
<p>$(\table {dx_1}/{dt}; {dx_2}/{dt}) = (\table -2, 1; 1, -2) (\table x_1; x_2)$</p>
<p> The modal coordinate function in matrix form is written as:</p>
<p>(3) $q = (\table c_1, c_2) (\table x_1; x_2) = c_1 x_1 + c_2 x_2$</p>
<p>To calculate a modal coordinate of a system of ODEs, we do the following:
<h3>Step 1 - Calculate an Eigenvector</h3>
<p>The first step for determining a modal coordinate is to calculate a left eigenvector $c_i$. A left eigenvector
a vector that solves the equation:</p>
<p>(4) $c_i A_{ij} = λ c_j$</p>
<p>This is called a left eigenvector because it is an eigenvector calculated by multiplying the matrix $A_{ij}$ by a
row vector $c_i$ on the left side.</p>
<h4>Step 1 Example</h4>
<p>For the example system above the left eigenvector equation is:</p>
<p>$(\table c_1, c_2) (\table -2, 1; 1, -2) = λ (\table c_1, c_2)$</p>
<p>The associated algebraic equations are:</p>
<p><table style="margin-left: 1em"><tr><td align="right">$-2 c_1 + c_2 =$</td><td> $λ c_1$</td></tr>
<tr><td align="right">$c_1 - 2 c_2 =$</td><td>$λ c_2$</td></tr></table></p>
<h4><a name="modes">Modes of the Example System</a></h4>
<p>The above algebraic equations have two (modal coordinate) solutions which can be found with a little algebra:</p>
<hr>
<table>
<tr><td><div class="box"><b>Mode 1</b></div> <div class="pad"></td><td>$λ = -1$ and $(\table c_1, c_2) = (\table s, s)$ where $s$ is a real number. From this we get $q_1 = s x_1 + s x_2$</td></tr>
</table>
<hr>
<table>
<tr><td><div class="box"><b>Mode 2</b></div> <div class="pad"></td><td>$λ = -3$ and $(\table c_1, c_2) = (\table s, -s)$ where $s$ is a real number. From this we get we get $q_2 = s x_1 - s x_2$</td></tr>
</table>
<hr>
</p>
<p>For modal analysis, the arbitrary constant $s$ is typically set of some specific value such as 1. However, we will just
keep the constant $s$ below since it helps identify where our modal coordiantes have been substituted.</p>
<h3>Step 2 - Left Multiply the System of ODEs by the Eigenvector</h3>
<p>Left multiplying the system of ODEs from equation (1) by the eigenvector $c_i$ and applying the eigenvector
equation from (4) gives:</p>
<p>(5) $c_i {dx_i}/{dt} = c_i A_{ij} x_j = λ c_i x_i$ </p>
<h4>Step 2 Example</h4>
<hr>
<table>
<tr><td><div class="box"><b>Mode 1</b></div> </td><td>$(\table s, s) (\table {dx_1}/{dt}; {dx_2}/{dt}) = -1 (\table s, s) (\table x_1; x_2)$</td></tr>
</table>
<hr>
<table>
<tr><td><div class="box"><b>Mode 2</b></div> </td><td>$(\table s, -s) (\table {dx_1}/{dt}; {dx_2}/{dt}) = -3 (\table s, -s) (\table x_1; x_2)$</td></tr>
</table>
<hr>
<h3>Step 3 - Bring the Left Hand Side Eigenvector inside the Derivative</h3>
<p>Since the coefficient in $c_i$ are contant relative to $t$, we can insert $c_i$ into the differential on the
left hand side of the equation. This is important because it results in the produce $c_i x_i$ inside the differential,
which equals $q$ [see equation (2)].</p>
<p>(6) ${d(c_i x_i)}/{dt} = λ c_i x_i$ </p>
<h4>Step 3 Example</h4>
<hr>
<table>
<tr><td><div class="box"><b>Mode 1</b></div></td><td> ${d}/{dt} ((\table s, s) (\table x_1; x_2)) = -1 (\table s, s) (\table x_1; x_2)$ ⇒ ${d}/{dt} (s x_1 + s x_2) = -1 (s x_1 + s x_2)$</td></tr>
</table>
<hr>
<table>
<tr><td><div class="box"><b>Mode 2</b></div></td><td> ${d}/{dt} ((\table s, -s) (\table x_1; x_2)) = -3 (\table s, -s) (\table x_1; x_2)$ ⇒ ${d}/{dt} (s x_1 - s x_2) = -3 (s x_1 - s x_2)$</td></tr>
</table>
<hr>
<h3>Step 4 - Substitute in the Modal Coordinate</h3>
<p>Both the left and right hand side of equation (6) now contain the product $c_i x_i$. From equation (2) this can
be replaced by $q$ to get the decoupled modal ODE:</p>
<p>(7) ${dq}/{dt} = λ q$</p>
<h4><a name="eqn">Step 4 Example - Modal Equations</a></h4>
<hr>
<table>
<tr><td><div class="box"><b>Mode 1</b></div></td><td>${dq_1}/{dt} = -q_1$</td></tr>
</table>
<hr>
<table>
<tr><td><div class="box"><b>Mode 2</b></div></td><td> ${dq_2}/{dt} = -3 q$</td></tr>
</table>
<hr>
<h3>The Solution</h3>
<p>The solution of the modal equation is $q = C e^{λ t}$ where $C$ is an arbitrary constant determined by
initial conditions.</p>
<h4>Solution for the Example Problem</h4>
<hr>
<table>
<tr><td><div class="box"><b>Mode 1</b></div></td><td> $q_1 = C_1 e^{-t}$</td></tr>
</table>
<hr>
<table>
<tr><td><div class="box"><b>Mode 2</b></div></td><td> $q_2 = C_2 e^{-3t}$</td></tr>
</table>
<hr>
<h2>Nonlinear Systems</h2>
<p>The modal analysis steps outlined above do not result in a decoupled ODE for nonlinear systems. The problem is
that some of the coefficients in the matrix $A_{ij}$ will be functions of $x_1, x_2, ..., x_n$. This means the
eigenvectors will also depend on these variables. Since $x_1, x_2, ..., x_n$ change with $t$, the eigenvectors
$c_i$ have a dependence on $t$, and the process fails at step 3 because a terms that changes with $t$ can't be
brought inside the derivative term. Since we can't bring $c_i$ inside the derivative, we don't have the product
$c_i x_i$ on the left hand side of the equation, so we can't replace the $c_i$ and $x_i$ with $q$, and the process
for decoupling the ODE fails.</p>
<p>So how do we fix this issue for nonlinear systems? Well, it turns out that for linear systems, the step shown above
are a combination of the <b class="emphasis">chain rule</b> along with a decoupling condition, although this isn't obvious. The chain rule is
the general rule for applying a change of coordinates to any system of ODEs - linear or nonlinear. <i>The problem is that
the steps above are not consistent with the chain rule when we are dealing with a nonlinear system.</i> In order to generalize
modal analysis to nonlinear systems, the change of coordinates and the decoupling condition should be expressed in a form
that is consistent with the chain rule.</p>
<p>Next Topic - <a href="extendModalAnalysis.html">Extending Modal Analysis to Nonlinear Systems</a></p>
</div><!--main-->
</body>
</html>