Вы находитесь на странице: 1из 9

COLORADO SCHOOL OF MINES Numerical Solutions of the Linear SDOF - Issue 1

Introduction Numerical solutions of the SDOF system can be obtained through discretization of the time t, displacement u, velocity ! u, acceleration " u, and loading function p(t). In the discretization process time is defined only at specific instances t0, t1, t2, . . ., ti, . . ., etc. The discretized values of time generate intervals !ti = ti+1 - ti. In most cases the discrete values of t are equidistant (all !ts are equal), while in some cases strategies of varying time intervals are employed based on need for accuracy. The dependent variables u(t), ! u(t), u " (t), and p(t), are also defined at the discrete times ti. In general any function F(ti) is commonly described as Fi. The numerical schemes that we shall consider can be divided in two major categories: a) the ones that are based on derivative approximations, commonly called finite differences and b) the ones that implicitly are based on integration. Finite difference approximations

At a time ti, the first derivative of a function F(t) can be approximated by numerous relations. Some of the most common ones are listed below:

First Derivatives

Forward Difference: Backward Difference: Central Difference: Second Derivative

! Fi+1 Fi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (1) F i t ! Fi Fi1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (2) F i t ! Fi+1 Fi1 F . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (3) i 2 t

At a time ti, the second derivative of a function F(t) can also be approximated by numerous expressions. However, the most common one is the second order central difference relations:
!! = Fi+1 2Fi + Fi1 F i t2

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (4)

Page 1

January 24, 2002

Implementation - Central Difference Method

The dynamic equation of equilibrium is valid at any time. Therefore it is valid at time ti: !! i + cu ! i + ku i = p i . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (5) mu We can now use Equations (4) and (3) to approximate the second and first derivatives of u:
m u i+1 2u i + u i1 t
2

+c

u i+1 u i1 + ku i = p i . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (6) 2 t

Collecting terms in Equation (6) we find:


c $ c * 2m * ! m ' m ' u i 1 ) k ui # 2 + & u i+ 1 = p i ) 2 , " t t2 , 2 t % 2 t + ( t ( +

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (7)

The above equation can be written in a shorter form as

" " ku i+ 1 = p i . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (8)


where
"= m + c k t 2 2 t

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (9)

and

m c * 2m * ' " i = pi ' p u i 1 ) k u i . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (10) 2 , ) 2 t + t2 , + ( ( t

Calculating ui+1 using Equation (8) is now straight forward as long as ui and ui-1 are known. The first point to evaluate is u1, which requires the knowledge of u0 (known from initial conditions) and u-1, which is not known. To calculate u-1 so that well start the solution process, the following procedure is followed: Using central difference, we write: u u 1 !o = 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (11) u 2 t Note, in Equation (11) that ! uo is known from initial conditions. Using central difference second derivative we write: u 2u o + u 1 !! . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (12) uo = 1 t2 Note, that in Equation (12) " uo is known from the equation of motion at time to: ! p o cu o ku o !! . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (13) uo = m Solving Equation (11) for u1 and substituting into Equation (12) we find:

!o + u 1 = u o tu

t2 !! u . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (14) 2 o

The process for solving a SDOF system using the central difference method, when the loading function is known in a discretized form is now straight forward: Page 2 January 24, 2002

Initial Calculations:
! o ku o p o cu m t2 !o + !! u 1 = u o tu u 2 o "= m + c k t 2 2 t !! uo =

m c 2 2 t t 2m b= k t2 a=

Calculations for Step i " i = p i au i1 bu i p

" 1p "i u i+ 1 = k

If needed:
!i = u

u i+ 1 u i1 2 t u i+1 2u i + u i1 !! ui = t2

or

p i cu i ku i !! ui = m

" 1p " i involves an equivalent stiffness matrix k # which is explicit solutions is that the solution u i+ 1 = k
not a function of the system stiffness k. The importance of this observation will become obvious when we shall examine systems with multiple degrees of freedom. Some of the disadvantages of the central difference method include: 1. ui+1 is found based on two earlier time step values ui and ui-1. Dependence of ui+1 on the previous value ui alone is preferable. 2. This method, as all explicit methods is conditionally stable. This means that if !t is larger than a limiting value the solution becomes unstable and diverges. This will be demonstrated in the example that follows. The conditional stability of this method is satisfied if t <

We have developed a technique that calculates the displacement ui+1 at time ti+1 by satisfying the differential equation of motion (Eq. 5) at time ti. That is, we calculate ui+1 without satisfying the equation of motion at time ti+1! This type of solution approaches are called explicit. A feature of

Tn

Page 3

January 24, 2002

Example

! + 4u = 5sin3t with initial conditions of rest, i.e. u + 0.2u Consider the dynamic SDOF system: !!
u(0) = 0 and u ! (0) = 0. We would like to obtain a solution for the first 12 seconds. The problem can be solved analytically since it describes a damped vibration with harmonic excitation. However, we shall develop a numerical solution based on the central difference equations.

n =

4 = 2, or Tn = 2"/#n = ". Thus !t < 1 to satisfy stability.

We shall first try a time increment !t = 0.1 seconds. That means that our times ti are: t0 = 0, t1 = 0.1, t2 = 0.2, t3 = 0.3, etc. We know that at time t=0 we have: u0 = 0, and ! u0 = 0, (initial conditions). Initial calculations:

!! u o = 0 0.2 0 4 0 = 0

Step i =0

0.12 u 1 = 0 0.1 0 + 0=0 2 " = 1 + 0.2 = 101 k 0.12 2 0.1 1 0.2 a= = 99 0.12 2 0.1 2 1 b= 4 = 196 0.12

" 0 = 5sin0 99 0 + 196 0 = 0 . Thus u1=0. p


" 1 = 5sin0.3 99 0 + 196 0 = 1.4776 . Thus u2 = 1.4776/101 = 0.01463. p

Step i = 1 (t = 0.1)

Step i = 2 (t = 0.2) Step i = 3 (t = 0.3)

" 2 = 5sin0.6 99 0 + 196 0.01463 = 5.6907 . Thus u3 = 5.6709/101 = 0.05634. p " 3 = 5sin0.9 99 0.01463 + 196 0.05634 = 13.5109 . Thus u4 = 13.5109/101 = 0.1338. p

The solution proceeds in the same way until t = 12 seconds. The results are plotted in Figure 2. To understand the issues of accuracy, we solve the problem again with !t = 0.5 seconds. In this case, we find k # = 4.2, a = 3.8, and b = -4. The solution process is identical to the previous case, and the answers are plotted in Figure 3. Despite the cruder shape of the solution, the results are fairly similar to those of !t = 0.1. The instability that results if !t = 1 (violation of limit) is demonstrated in Figure 4. Page 4 January 24, 2002

2 1 Displacement 0 -1 -2 -3 0 2 4 6 Time (sec) 8 10 12

Figure 2: Solution for !t = 0.1 seconds

3 2 Displacement 1 0 -1 -2 -3 0 2 4 6 Time (sec) 8 10 12

Figure 3: Solution for !t = 0.5 seconds

150 100 Displacement 50 0 -50 -100 -150 0 2 4 6 Time (sec) 8 10 12

Figure 4: Solution for !t = 1 second Page 5 January 24, 2002

Constant Acceleration Method The following method is used frequently. In stead of generating solutions based on displacements, which then require numerical differentiation to obtain velocities and accelerations, we can set the acceleration as the base for calculating velocities and displacements through integration. This produces more accurate, and in certain cases unconditionally stable solutions. The method of constant acceleration is based on the following simple concept: In two consecutive ! i, and u " i and ui+1 , ! ui+1, and u " i+1 respectively. We shall assume time instances ti and ti+1 we have ui , u that in the time interval !ti between times ti ti+1 the acceleration is constant and equal to the ui+1: average value of " ui and " 1 !! u(t) = 2 (!! ui + !! u i+1 ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (15) The assumption of constant acceleration within a time interval leads in linear velocity and quadratic displacement, as demonstrated in Figure 5. Integrating Equation (15) and applying the initial condition that for t = ti ! u=! ui, we find that

ti $ ! i+1 = u !i + ! u u i + !! u i+1 ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (16) # & (!! " 2 %

Integration of the displacement results in:


! t2 $ ! i t i + # i & (!! u + !! u i+1 ) u i+1 = u i + u # 4 & i " %
4

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (17)

We can solve Equation (17) for " ui+1:


!! u i+1 = t2 i

(u i+1 ui ) 4 t

! i !! u ui

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (18)

Combining Equations (16) and (17) we find:


! i+1 = 2 (u i+1 u i ) u !i u ti

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (19)

ai =

1 2

u i + !! u i+1 ) (!!

Figure 5: Geometric assumptions of the constant acceleration method. Substituting Equations (18) and (19) into the dynamic equation of Equilibrium at ti+1 results in Page 6 January 24, 2002

* K* i+1 u i+1 = p i+1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (20)

where
K* i+1 = 4m t2 i + 2c + k . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (21) ti

and
! 4m 2c $ $ 4m &u + ! # ! i + mu !! i + c& u pi+* = p + + # 1 i+1 # i 2 & t t " % t i% i " i

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (22)

The solution of a dynamic problem with initial conditions u(t0) = u0 and ! u(t0) = v0 is now straight forward: 1. For i = 0 we know u0 and ! u0 from the initial conditions and " u0 from the dynamic equilibrium equation. ! 1 from Equation (19), and u " 1 from Equation (18), or 2. Thus we can find u1 from Equation (20), u from the dynamic equilibrium at t1 3. Now that we know u1, ! u1, and u " 1, we can set i = 1 and repeat step 2 to find u2, ! u2, and u " 2. This process continues for all subsequent time steps. However, the most common implementation of the constant acceleration approach is in the incremental time form: Since the dynamic equilibrium is valid at all times, it is valid at times ti and ti+1:
!! i + cu ! i + ku i = p1 mu

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (23) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (24)
! i + k u i = p i or m !! u i + c u

and
!! i+1 + cu ! i+1 + ku i+1 = p i+1 mu

Subtracting (23) from (24) we have


! i+1 u ! i ) + k(u i+1 u i ) = p i+1 p i m(!! u i+1 !! u i ) + c(u

. . . . . . . . . . . . . . . . . . . . . . . . . . . . (25)

Now, we can solve Equation (17) for !u " $:


! 4 $ & ( u u !! i . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (26) !! ui = # i ! i t i ) 2u # t2 & " i %

Equations (16) and (17) can also be combined to give:


! $ ! i = # 2 & u i 2u ! i . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (27) u " ti %

Substituting Equations (26) and (27) in (25) results in: * k* i u i = p i . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (28) where
$ ! 2c $ ! # 4m & k* = k + + # & i 2& " ti % # " ti %

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (29)

and

Page 7

January 24, 2002

The solution of a dynamic problem with initial conditions u(t0) = u0 and ! u(t0) = v0 is now straight forward: u0 from the initial conditions and " u0 from the dynamic equilibrium 1. For i = 0 we know u0 and ! equation. 2. Thus we can find !u0 from Equation (28), !u ! 0 from Equation (27), and !u " 0 from Equation (26). u1 = ! u0 + ! ! u0, and u " 1 from the dynamic equilibrium at time 3. Now that we can find u1 = u0 + !u0, !
" m%

' ! 4m $ * * ! i + 2mu !! i . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (30) pi = pi + ) # & + 2c, u t " % ) , i ( +

! 1$ ! 1 ku 1 ) or from u "1 = " uo + ! u " 0. Then we can set i = 1 and repeat step 2 to find t1: !! u i = # & ( p1 cu

u2 , ! u2, and u " 2. This process continues for all subsequent time steps.

Example
Consider the same dynamic SDOF system we considered earlier: !! ! + 4u = 5sin3t with initial conditions of rest, i.e. u(0) = 0 and u u + 0.2u ! (0) = 0. We would like to obtain a solution for the first 12 seconds. Let us establish a time increment !t = 0.1 seconds. That means that our times ti are: t0 = 0, t1 = 0.1, t2 = 0.2, t3 = 0.3, etc. We know that at time t=0 we have: u0 = 0, and ! u0 = 0, (initial conditions) and " u0 = 0 (from the * * dynamic equilibrium equation). Note that for i = 0, k0 = 408 (Eq. 29) and !p0 = 1.478 (Eq. 30). Now we can find !u0 = !p0*/k0* = 0.00362. Thus u1 = u0 + !u0 = 0.00362. From Eq. 27, !! u0 = 0.0724 ! ! u1 = 0.0724. Finally, from the dynamic equilibrium at t1 we find u " 1 = 1.449. This process continues until t = 12 seconds. The results for displacement are plotted in Figure 6 and are for all practical purposes identical to the theoretical answer. The solution is repeated for !t = 0.5 seconds (Figure 7), and !t = 1 second (Figure 8). Note that the !t = 0.5 seconds solution is still fairly accurate, and as opposed to the earlier finite difference approach, the !t = 1 sec. solution is stable even though it is not very accurate.

Page 8

January 24, 2002

2 1 Displacement 0 -1 -2 -3 2 4 6 Time (sec) 8 10 12

Figure 6: Solution for t = 0.1 sec.

1 0.5 Displacement 0 -0.5 -1 -1.5 2 4 6 Time (sec) 8 10 12

Figure 7: Solution for t = 0.5 sec.

0.1 0.05 0 -0.05 -0.1 2 4 6 Time (sec) 8 10 12

Figure 8: Solution for t = 1 sec.

Displacement

Page 9

January 24, 2002

Вам также может понравиться