You are on page 1of 16

~ ~ '~ "~, ?

, :-,
FINITE ELEMENTS IN ANALYSIS A N D DESIGN

ELSEVIER

Finite Elements in Analysis and Design 21 (1996) 129-144

Discretization considerations in moving load finite element beam models


J o s e p h R. R i e k e r a, Y i h - H w a n g L i n b, M a r t i n W . T r e t h e w e y a' *
aDepartment of Mechanical Engineering, The Pennsylvania State University, University Park, PA 16802, USA bDepartment of Mec,~anical and Marine Engineering, National Taiwan Ocean University, Keelung City, Taiwan, ROC

Abstract

This paper investigates continuum discretization for finite element models analyzing a moving load on an elastic beam. Moving load analysis :is shown to require accurate evaluation of beam deformations over the entire length of an element, and not only the nodes. Model accuracy is shown to be related to element interpolation which in turn directly affects three aspects of the moving load finite element model; (1) calculation of equivalent nodal reactions for the moving load; (2) calculation of transmitted forces from a moving sprung mass, and; (3) system responses for a moving load initially positioned within the support beam span. The analysis indicates that the model accuracy can be maintained at an acceptable level provided that, in general, the number of elements used to discretize the support structure continuum is at least two to eight times greater than the number used in static analysis.

1. Introduction

The discretized g e o m e t r y used to f o r m u l a t e a finite element model of a n y structure can greatly affect the accuracy a n d validity. A balance m u s t be m a i n t a i n e d between having sufficient resolution to define the p r o b l e m a n d having a m a n a g e a b l e model size. The proper discretization of a cont i n u u m remains in the area where significant engineering j u d g e m e n t is usually required to develop a useful a n d accurate model. Typically, a m i x t u r e of small a n d large finite element dimensions can p r o d u c e a feasible model. In areas where high strain gradients are expected the m e s h density is increased to provide the necessary detail. Tlae discretization problem can become further c o m p o u n d e d when an area requiring fine resolution changes t h r o u g h o u t the analysis. F o r example, finite element analysis m o d e l i n g acoustic p r o p a g a t i o n p h e n o m e n a are c o n f r o n t e d by a high gradient region which moves t h r o u g h the entire m e s h in a time variant fashion. It is infeasible to finely discretize the entire region because * Corresponding author. 0168-874X/96/$15.00 1996 Elsevier Science B.V. All rights reserved SSDI 0 1 6 8 - 8 7 4 X ( 9 5 ) 0 0 0 2 9 - 1

130

J.R. Rieker et al./Finite Elements in Analysis and Design 21 (1996) 129-144

computationally the problem quickly becomes too unwieldy. Problems of this nature are often handled by using an adaptive mesh generation scheme whereby the region of fine mesh resolution continually changes in parallel to the propagating high gradient region. Although adaptive mesh generation has been successfully utilized, it requires significantly greater programming and computational effort [-1]. Another class of problems where the region requiring a finer mesh resolution changes throughout the analysis is found in moving load type problems. Engineering problems of this nature are encountered in the study of bridge [-2, 3] and machine tool [-4] dynamics. In these problems, the applied load moves its geometric point of application on the structure in a time-dependent manner. To produce accurate results, a finer mesh is required in the area surrounding the application point of the force. This implies that, either a fine mesh be used over the entire discretized domain of the moving load or an adaptive mesh scheme implemented. The objective of this work is to investigate discretization errors induced in the finite element analysis of moving load type problems to help guide the development of suitable model meshes. Most of the previously reported moving load formulation associated with the finite element method (FEM) have been restricted to beam type structures because of the overall analysis difficulty [5]. Therefore, the work presented here will focus on only beam type structures. The analysis will investigate the relationship between model accuracy and the number of elements used to discretize a structure for a moving load analysis. The work will be presented in the following three sections. First the analysis of moving load problems on beam structures will be reviewed focusing on aspects of the resulting FEM which can be most affected by discretization. An analysis is then performed to examine the magnitudes and locations of the errors for a statically loaded beam structure. These results are then applied to examine the errors induced in the moving load type problem and examine the overall effect on a dynamic model.

2. Moving load finite element models Closed-form solutions for the governing partial differential equations of moving load problems are limited to rather simple cases. If a beam is not simply supported, Ritz type solutions using sinusoidal functions cannot be used as assumed mode shapes and the analytical solutions become very cumbersome [-6, 7]. Therefore, most work on problems of this type have applied the finite element method [-3, 5]. To enable a discussion of potential errors induced by the discretization process it is necessary to form a finite element model of a continuum; the development of this class of models will first be briefly reviewed. The review will concentrate on areas of the model which are most sensitive to the discretization process. 2.1. Concentrated moving load finite element models A simple model with all the basic characteristics of moving load problems is shown in Fig. 1. A beam structure is simply supported while a concentrated force traverses across the beam in a time-dependent manner. Alternative models of this type utilize fixed, fixed boundary conditions, intermediate supports and a sprung mass instead of a concentrated force. These more advanced

J R . Rieker et al./Finite Elements in Analysis and Design 21 (1996) 129-144

131

L
I-

x
'

-,

J
q

Fig. 1. A simply supported beam subjected to a moving concentrated force.

capabilities are necessary to analyze actual engineering systems I-4]. However, since the model in Fig. 1 possesses all the fundamental characteristics of moving load problems the development will focus on it for the sake of simplicity. The results from the model serve as a foundation which can be extended to more complicated structures. The application of the finite element method to the problem in Fig. 1 yields the governing structural equation [5] [M] {~/'} + [C]{d} + [K]{d} = { f } ,

(1)

where [M] is the structural mass matrix, I-C] is the structural damping matrix, [ K ] is the structural stiffness matrix, { f } is the external load vector and {d} is the displacement vector and the dot notation i,; used to denote differentiation with respect to time. The concentrated load traversing the beam is handled by modification of the external load vector {f} in Eq. (1) by {f} = LN JTao, (2)

wherefo is the magnitude of the applied concentrated load, and LN ja- is the transposition of the shape functions evaluated at the position where the force is applied, i.e. the location indicated by x in Fig. 1. LNJ a" is a vector with zero entries except those corresponding to the beam nodal displacements on which the concentrated load is acting. Thus, for a beam element with four degrees of freedom, the number of nonzero entries within the n by 1 vector will be four, where n is the total number of degrees of freedom in the finite element model of the support structure. The values of this 4 by 1 "subvector" are dependent on the application point of the load within an element. Since external loads can only be applied at nodes, the concentrated moving load must be reduced to equivalent nodal forces and moments. This process is illustrated in Fig. 2. Basically, the concentrated load is transformed to appropriate nodal moments and forces. The equivalent nodal loads are expressed as a, 4 by 1 vector obtained by multiplying the transpose of the shape functions, evaluated at the position of the applied load, with the concentrated force. When the concentrated force moves to another position, the numerical values of the shape functions change and consequently the equivalent nodal loads vary accordingly. It should be apparent that when the concentrated force, moves to an adjacent element the shape functions of the new element must be used. This results in a positional shift of the equivalent nodal loads within the structural load column vector.

132

J.R. Rieker et al./Finite Elements in Analysis and Design 21 (1996) 129-144

fl,,

fR
Equivalent Loading
>

1
t,D:) : t,<j,o
J evaluated at x Actual Loading

fo

Fig. 2. Equivalent beam element nodal loads for a concentrated force located between the nodes.

2.2. Moving mass finite element analysis

This approach can be extended to the case where the concentrated moving load is replaced by a sprung mass. This model is represented pictorially in Fig. 3 and its implementation is more complicated. The moving system now consists of two masses ml, the sprung mass and m2, the unsprung mass which is assumed to always remain in contact with the support beam structure. Following the convention used in Ref. [5-1 the equations governing this dynamic system can be obtained using the finite element method:
m l y + c(.~ -- Wo) + k ( y - Wo) = O,

(3) [ M ] {d'} + [C]{d} + [ K ] { d } = { f }


where fo = (ml -I- m 2 ) g -- m2wo + FT, (4)
FT = k ( y -- Wo) + c(~ -- fro).

=LNJafo,

The variable Wo denotes the vertical dynamic displacement of m2 which is equal to the deflection of the support beam at the point of contact along the beam. Fa- is the transmitted force to the beam from the sprung mass and can be written alternatively as [8-1
FT = f - - m~ y , (5)

J'.R. Rieker et al./Finite Elements in Analysis and Design21 (1996) 129-144

- Ty
k c

133

=
X

i,d vI L

w0

v!

Fig. 3. A fixed-fixedbeam subjectedto a movingsprung mass system. where F, in general, is an applied external force. For this case, since there is no external forcing function acting on ml, F is zero. Therefore, a simplified relationship for fo is obtained by substituting Eq. (5) into Eq. (4) yielding

fo = ml(g -- j)) + m2(g -- f4o).

(6)

The function Wo(X,t), representing the displacement at any position, is obtained from the shape functions and nodal displacements of the support beam:

Wo = LN J{a}.
The respective time derivatives of Wo required in Eqs. (3) and (4) are

(7)

Ow aw Wo(x,t) = ~ ~ + 0-7'
(8)
O2W "2 O2W OW 02W

#o(X, t) = -~x:':x + 2 ~ - ~ 2 + ~x y + -atT " Applying the time derivatives in Eq. (8) to the general beam displacement expression in Eq. (7) yields
~2 W 02W

ax~ aw = a~

LNJ..{d},

axat-LNJ={d},
(9)

L N J~{d},

a~W=LNj{~.}, ~3t2

where the subscript, Ix, denotes differentiation with respect to x. Substituting Eqs. (6), with mz = 0, (7), (8), and (9) into Eq. (3) yields I-8]
LOJ + m, /

-cLN_I
(10)

[g] -cYcL N J= - kL N J {:} ) ({dy} ) = (L N ~m~o)

134

J.R. Rieker et al./Finite Elements in Analysis and Design 21 (1996) 129-144

The system matrices in Eq. (10) are time variant, with the entries being directly dependent on the position of the sprung mass along the beam. The submatrices LN JTml, - c LN J, -c2[N_]x, - k E N _]and k N JTmlg are time dependent and move within the structural matrices as the sprung mass travels from one element to another. Similarly, a point mass moving at a constant velocity may be analysed by setting ml = 0.0, which yields

[ [ M ] + m2LNPLN]] {2"} + [[C] + 2m=aLNPLNJx]{d} + [ [ K ] + m2a2L PLN Ixx]{d} = L N N

]Tm2g-

(11)

The finite element methods developed in these sections have the capabilities necessary to study a moving load on an elastic beam. Eqs. (10) and (11) are capable of evaluating the system's dynamic response as the mass traverses the beam, provided the movement is initiated from outside the beam span, i.e. the sprung mass enters the beam from one end. Extensions readily follow for more complicated systems. However, examination of these models will be sufficient to examine the fundamental accuracy questions related to the discretization of the beam continuum into finite elements.

3. S e n s i t i v i t y to e l e m e n t d i s e r e t i z a t i o n

To solve Eqs. (10) and Eq. (11), the beam structure continuum must be discretized into m nodes connected by m - 1 beam elements. The element mass and stiffness matrices can be obtained using shape functions derived from a cubic Hermitian interpolation function. The shape functions are [9]

LNd=LN,

N3

N4J ,

N1 = 1 - - 3

(7) 2 + 2
- ,

(/)'

N2 =x

N3=3

(# (#
--2 7 '

(12)
N4=x

The F E M formulation applied to Eqs. (1) and (11) yields a set of n second-order differential equations where n is the number of discretized degrees of freedom. Eq. (1) can be solved by either mode superposition or direct step by step integration. However, direct integration should be used for Eqs. (10) and (11) because of the time variant nature of the modal characteristics. The number of elements used to represent the support beam plays an important role in establishing the accuracy and required computation time in the developed model. Beyond standard F E M discretization modeling practices, there are additional demands induced because of the moving nature of the load. All are related to how well the interpolation function can represent characteristics of the entire beam using only the nodal values. In moving load problems it is necessary to evaluate accurately the interaction between the support beam and the moving sprung mass at any instant of time or location along the beam. The interaction requires the displacement profile be known accurately over the entire beam span and not just at the nodes.

J.R. Rieker et al./Finite Elements in Analysis and Design 21 (1996) 129-144

135

Three aspects of the moving load F E M s are most affected by the discretization-interpolation process: 1. The computation of the equivalent external nodal moments and forces from the moving load (e.g. Fig. 2). Any interpolation error can affect the magnitude of the equivalent nodal loads. 2. The computation of the transmitted force created by the vertical motion of the sprung mass requires an accurate value of the beam deformation directly below the mass at the point of contact with the beam. Any induced beam displacement error will directly affect the magnitude of the transmitted load applied to the beam. 3. When the sprung mass is initially positioned within the beam span, an accurate evaluation of the static deformation from the preload is required. An interpolation error can affect the resulting dynamic displacements which in turn affect the transmitted load from the sprung mass. All of these potential errors stem from the need to know accurately the characteristics over the entire beam span at any spatial location and not at only the node. It is also apparent that an error induced at one integration step will readily propagate into and affect subsequent time step computations. Therefore, it is crucial to understand how the number of elements used to discretize the support beam effects the interpolation of beam characteristics over the entire span. To examine the interpolation difficulties, two related studies were performed. First the discretization errors under static conditions were evaluated with simply supported and fixed-fixed boundary conditions. These results served as baseline results. The second study examined the discretization-induced difficulties under a moving load and moving mass, respectively.

3.1. Static deflection discretization errors

Consider the application of a concentrated force at different locations along a beam supported at the ends. The beam continuum is discretized to form three different models, each with an increasing number of equal-,~ized elements (i.e. four, eight, and sixteen). A load is applied at a position x, and the static deformation at that point calculated by each respective FEM. The deformation from each model can then be compared to the theoretical values to form a quantitative measure of the error. The load position can then be moved across the beam and the comparison repeated. Therefore, two separate analyses were performed, each with different boundary conditions; (1) two simple supports and; (2) two fixed-fixed supports. These two cases represent an upper and lower bound, with respect to possible beam deformation magnitudes. The results for the analysis applied to a simply supported beam is shown in Fig. 4. The abscissa axis represents the position x, of a static load normalized by the beam length L. The ordinate axis denotes the static displacement error between the finite element model YVEM and theoretical values Ytheory at the respective load application point. The results indicate that the displacements computed by the three finite element models are essentially identical to the theoretical results at the nodes of each model. The finite element formulation strives to attain this behavior and indicates that all subsequent computations based on only the nodal deformations will be correct. However, this will not be true for calculations requiring accurate displacement values between the nodal locations, as is the case for the moving load class of problems. The error is greatest at a position midway between the element nodes, irrespective of the number of elements in the model. The greatest error value, for the cases evaluated here, is approximately 3 %. The error decreases rapidly

136

J.R. Rieker et al./Finite Elements in Analysis and Design 21 (1996) 129-144


4

3-

/ / /
x

\ \ \ \ \
\

1-

\ \
\
l I l I
~ ~ ~i ,~

I I I I I

It-

I l m l l l l l l l

0.0

0.i

0.2

0.3

0.4

0.5

NORMALIZED LOCATION

X "~)

Fig. 4. Static displacement error at the point of force application for three finite element models with simply supported boundary conditons. ( ) 16 elements; ( - - - ) 8 elements; ( - - - - - ) 4 elements.

with respect to the number of elements, a doubling of elements decreases the normalized displacement error significantly. For the case with sixteen elements the only appreciable error (0.6%) occurs near the boundary where the values are small. Fig. 5 illustrates the results when this analysis is repeated with fixed-fixed beam boundary conditions. The same trends are apparent for this case with the differences occurring near the fixed-fixed boundary. The errors become large (greater than 10%) near the boundary because of the inherently small displacements. Excellent agreement is obtained at the nodal locations with the errors becoming larger between the nodes. As the number of elements increase the high error region near the boundary is reduced. F r o m this analysis it is apparent that the displacements computed at the nodal positions of F E M are accurate and are relatively insensitive to the number of elements used to discretize the continuum. However, for a position between nodal points an error exists and is dependent on the number of elements used to form the model and the boundary conditions. In general, the error tends to be greater for more constraining boundary conditions and the error decreases as the number of elements increases. Stress computation involves higher derivatives of the displacement. To calculate the bending stress requires the computation of bending moment, which involves the second derivative of displacement, while calculation of shear stress needs the knowledge of shear force, which is a function of the third derivative of the displacement. Therefore, it is expected that

,/.R. Rieker et al./Finite Elements in Analysis and Design 21 (1996) 129-144 20


I I I I I I

137

].5

I i I |

J0

/ / \
I

\ \

0.0

'

'

'

'

'

'

'

'

0.i

0.2

0.3

0.4

0.5

NORMALIZED LOCATION

('~)

Fig. 5. Static displaca,~ment error at the point of force application for three finite element models with fixed-fixed boundary conditons. ( - - ) 16 elements; ( - - - ) 8 elements; ( - - - - - ) 4 elements.

error in the interpolation process for displacement will greatly affect the accuracy in stress analysis. The degree of error in stress analysis is expected to be higher than that in displacement analysis due to the use of insul~cient number of elements.

3.2. Dynamic deflection discretization errors


The discretizatJion errors in the moving load/mass class of problems are more difficult to evaluate because of the limited conditions for which analytical solutions exist. Analytical solutions for only the moving force problem with simple supports have been developed [6, 10]. Initially, this case will be used to evaluate potential discretization errors by comparing FEM results to the analytical results. Next, an evaluation for a moving force with fixed-fixed boundary conditions and moving mass with simple supports and fixed-fixed boundary conditions will be presented. Since analytical solutions do not exist for these conditions, the rate of convergence versus the number of elements will be used as an evaluation criterion.

Moving point force


Consider a moving point force on an elastic simply supported beam shown in Fig. 1. The analytical beam response [10] can be compared to numerical results, with varying number of

138

J.R. Rieker et al./Finite Elements in Analysis and Design 21 (1996) 129-144

elements, by implementation of Eq. (1). The impact factor, defined as the ratio of the dynamic deflection to the deflection with an equivalent static load, Ydynamic/Ystatie, a t the beam's center, is typically cited as the parameter describing the dynamic performance of this problem class [-5, 8]. Therefore, the following evaluation will be based on the impact factor at the beam center. Fig. 6 shows the impact factor when a force which traverses the beam span in a time equal to the period of the beam's first mode. The analytical [10] and F E M results show excellent agreement. Fig. 7 shows percent error functions, based on the analytical solutions [10] for a finite element analysis with 4, 8 and 16 elements for the same velocity in Fig. 6. The largest errors occur at the beginning and end times and are a result of computations with relatively small impact factors. The discretization can best be evaluated by examining the variation of the error functions with respect to the number of elements. The only notable differences occur with the four element model. The impact factor errors for the eight and sixteen-element models are essentially the same. An alternate manner to evaluate discretization-induced errors is to examine the convergence of the maximum impact factors for various traveling speeds and elements. Table 1 presents the impact factor at the beam center for normalized four speeds, T f / z , in relation to six different F E M discretizations. The analytical impact factor based on Ref. 1-10] is also included for comparison. Observation of the impact factor convergence indicates no changes are apparent beyond the beam

2.0

I- ~ /,/

1.5
//

//
/"

"~ o I< F<


~w

1.0

0.5
/
/ / / //" / /

,/

0.0 0.0

"i ~

0.2

0.4

0.6

0.8

1.0

NORMALIZEDTIME

(t)

Fig. 6. Beam center impact factor for a traveling point force with Te/z = 1 with simply supported boundary conditions: ( - - - ) Analytical solution from Ref. 1-10];( - - - - - ) Finite element solution from Eq. (1) with 8 elements.

.t.R. Rieker et al./Finite Elements in Analysis and Design 21 (1996) 129-144

139

o
0.0

I
0.2 0.4 0.6 0.8
i .0

NORMALIZED TIME ()
Fig. 7. Beam center impact factor error for a traveling point force with Tf/~ = 1 with simply supported boundary conditions for three finite element models based on the analytical solution [10]: ( ) 16 elements; ( - - - ) 8 elements; ( - - - - - ) 4 elements. Table 1 Impact factors at the center of an elastic beam for a moving point force with simply supported boundary conditions

Tf/z

No. of elements 3 4 1.258 1.706 1.730 1.703 5 1.257 1.706 1.731 1.701 8 1.258 1.706 1.732 1.702 9 1.258 1.707 1.732 1.702 16 1.258 1.707 1.732 1.702

Analytical

[9]
0.5 1.0 1.234 1.5 1.253 1.700 1.732 1.695 1.2577 1.7053 1.7316 1.7016

discretized into eight elements. M o d e l s with even n u m b e r s of elements t e n d to c o n v e r g e faster t h a n o d d n u m b e r models. This is n o t surprising, since e v e n - n u m b e r e d element m o d e l s will h a v e a n o d e at b e a m c e n t e r w h e r e the i m p a c t f a c t o r is m e a s u r e d a n d is i n h e r e n t l y m o r e accurate. C o n s i d e r a m o v i n g force o n a b e a m with f i x e d - f i x e d b o u n d a r y conditions. Since analytical solutions d o n o t exist for this c o n d i t i o n , c o n v e r g e n c e of the i m p a c t f a c t o r can be assessed to e v a l u a t e the m o d e l a c c u r a c y as the n u m b e r o f elements increases. Results are s h o w n in T a b l e 2. T h e results s h o w t h a t the impact factor basically c o n v e r g e s to its final values with a n eight-element

140

J.R. Rieker et al./Finite Elements in Analysis and Desion 21 (1996) 129-144


Table 2 Impact factors at the center of an elastic beam for a moving point force with fixed-fixed boundary conditions

Tr/T

No. of elements 3 4 1.307 1.640 1.557 5 1.303 1.637 1.553 8 1.311 1.637 1.552 9 1.310 1.637 1.551 16 1.311 1.638 1.551

0.5 1.0 1.5

1.281 1.623 1.526

model for the traveling speeds evaluated. The rate of convergence is slower with the fixed-fixed boundary conditions, with some small differences apparent in the four-element model. Again, even-numbered element models reach the final value more rapidly.

Moving point mass The moving point mass problem may be analyzed by the implementation of Eq. (1). The rate convergence of the impact factor with different boundary conditions, number of elements, traveling speeds and point mass values is used to determine the effectiveness of the discretization. Table 3 presents impact factors for a simply supported beam with similar operational conditions to those used in Table 2. In addition, this study has included three different point mass values. The mass values are normalized with respect to the beam mass and presented in terms of a mass ratio mmoving/mbeam. For a mass ratio of 0.5, similar results to the moving point force are observed, impact factor convergence is reached with eight elements. As the mass ratio increases the impact factor convergence requires more elements. Differences can be observed between the eight and sixteen element models. Convergence delineation between even and odd numbers of elements is not apparent as before. For all cases presented the impact factor has converged with sixteen elements. Table 4 presents impact factors for a fixed-fixed beam for similar conditions as those in Table 3. Similar convergence trends, with slightly larger deviations, are observed. As the mass ratio increases for a given velocity ratio, the number of elements required for convergence increases. A similar trend is observed for a constant mass ratio as the velocity ratio increases. The data also suggest that the rate of convergence is more sensitive to the velocity ratio (traveling speed) than the mass ratio. The phenomena affecting impact factor convergence for the traveling point mass can be traced back to the equation of motion (11). It can be seen that the system's mass, damping and stiffness matrices are continuously changing due to the point mass, its location and velocity. This is a result of the inertial coupling that exists between the point mass and the beam. A comparison between the equations of motion for a point mass (11) and a point force (1) illustrate these differences. It is the inertial coupling that affects number of elements required for convergence. Since the system matrices are continuously changing, the modal properties of the system are also changing. Furthermore, the sensitivity to traveling speed may also be explained by the presence of the velocity squared term in the system's stiffness matrix. The coupling term in the system's stiffness matrix is given as m2~2LN]TLNAx x. If the value of the velocity is increased by a factor 'b', this coupling term increases by a factor of b 2. For example, a doubling of the velocity will increase the coupling term by a factor of four. Similarly, an increase of "b" in the mass, only results in an

J.R. Rieker et al./Finite Elements in Analysis and Design 21 (1996) 129-144

141

Table 3 Impact factors at the center of an elastic beam for a moving point mass with simply supported boundary conditions Mass ratio
0.5

Tf/z

No. of elements 3 4 1.417 2.048 2.253 1.574 2.506 2.688 1.891 3.401 3.716 5 1.417 2.046 2.260 1.570 2.479 2.691 1.873 3.384 3.692 8 1.418 2.049 2.270 1.571 2.454 2.697 1.864 3.408 3.702 9 1.418 2.049 2.270 1.571 2.452 2.694 1.862 3.408 3.716 16 1.418 2.048 2.271 1.571 2.452 2.693 1.857 3.406 3.707 24 ----2.453 2.690 1.857 3.404 3.709 32 ----2.453 2.690 1.857 3.404 3.709

0.5 1.0 1.5 0.5 1.0 1.5 0.5 1.0 1.5

1.414 2.043 2.246 1.565 2.522 2.757 1.900 3.357 3.654

1.0

2.0

Table 4 Impact factors at the center of an elastic beam for a moving point mass with fixed-fixed boundary conditions Mass ratio
0.5

Tf/'r

No. of elements 3 4 1.533 2.212 2.155 1.701 2.588 2.365 2.390 3.008 2.550 5 1.522 2.199 2.137 1.696 2.580 2.329 2.408 2.968 2.377 8 1.531 2.184 2.160 1.684 2.490 2.375 2.414 2.990 2.438 9 1.533 2.190 2.155 1.684 2.523 2.378 2.396 2.936 2.415 16 1.531 2.192 2.172 1.686 2.523 2.351 2.391 2.897 2.447 24 1.531 2.197 2.171 1.686 2.551 2.289 2.413 2.896 2.433 32 1.531 2.198 2.172 1.685 2.542 2.293 2.413 2.896 2.360

0.5 1.0 1.5 0.5 1.0 1.5 0.5 1.0 1.5

1.514 2.169 2.076 1.656 2.533 2.317 2.454 2.963 2.460

1.0

2.0

increase of"b" in the coupling term. Again if the mass m 2 is doubled, then the coupling term is also doubled. Therefore, any change in the velocity will have a greater affect on the system's dynamics than a corresponding change in the mass.
3.3.

Summary

The results indicate that stiffer boundary conditions (i.e. fixed-fixed ) require a great number of elements in the model. For the traveling mass class of problems: (1) the greater the ratio of the traveling mass to the beam mass; and/or (2) the greater the traveling velocity to the beam's fundamental frequency, the finer the required discretization. It is difficult to determine exactly to what extent the be,am displacement error will affect the overall accuracy of a finite element model, because so many .other system variables must be known (e.g. stiffness of a moving sprung mass). However, it is quite conceivable that displacement errors obtained would be suffcient to induce

142

J.R. Rieker et al./Finite Elements in Analysis and Design 21 (1996) 129-144

significant errors in subsequent computations requiring the beam displacement. Furthermore, the effect on the overall model results may be significant and not directly traceable due to the interaction and dependent nature on the other system parameters and computations. Therefore, it suffices to indicate that the beam deformation errors must be maintained as low as possible.

4. Concluding remarks
The work presented in this paper has investigated the discretization of beam structures for the moving load class of finite element models. These type of problems have more demanding requirements because of the need to characterize accurately the support beam behaviour at any location along the beam span and not just the nodes. The interpolation error potentially affects three aspects of the moving load model: (1) calculation of equivalent nodal reactions for the moving load; (2) calculation of the transmitted force from a moving sprung mass, and; (3) overall system response when the moving load is initially positioned within the support beam span. Although the normalized displacement errors can be significant (e.g. greater than 10%), they can usually be reduced to a negligible level by increasing the number of elements. For a stationary static deformation analysis with the loading and boundary conditions used in this work, a model with four elements would usually be deemed acceptable. Furthermore, since a static F E M analysis usually focusses around the model nodes, the results would indeed be accurate and usable. However, for a moving load type problem the mesh density must be increased because of the need to have accurate deformations between the model nodes. The results shown here indicate that the number of nodes and elements needed to discretize a continuum for a moving load analysis must be greater than that used for a similar static analysis. Realizing that in general discretization of a continuum is an inexact science and somewhat of an art, a guideline specifying the required number of nodes and elements is difficult to produce. The analysis indicates that when holding all other parameters constant, the discretization errors tend to increase when (1) the traveling speed Tf/'c increases; and (2) the mass ratio mmoving/mbeam increases. The analysis has demonstrated that for the moving load class of problems the number of nodes and elements should be at least two to eight times greater, and even more if possible, than that which would be used for a similar static analysis. In the situation where the movement of the traveling load is confined to a certain area within the beam, it would be a good practice to use a finer mesh in the motion area. Placement of nodes at critical locations is a good practice since computed values are more accurate. The results have shown that increasing the mesh density is sufficient to maintain the induced errors to a manageable level. Furthermore, the increase in the number of elements is well within reason and can readily be handled computationally. Therefore, there does not appear to be a need to consider the use of adaptive mesh generating method for moving loads/sprung mass on beam type structures. However, for larger more complicated structures which inherently have greater nodal degrees of freedom (e.g. two-dimensional plate or three-dimensional solid elements) this question still remains to be studied. It has been demonstrated in [Ref. 11] that for dynamic analysis of beam structures traversed by a moving sprung dynamic system initially positioned within the beam span, as applicable for analysis of a high-speed precision drilling machine [4], the error can be drastic if insufficient number of elements is used. It was observed that accurate results can be obtained if a finer mesh is

3.R. Rieker et al./Finite Elements in Analysis and Design 21 (1996) 129-144

143

used in the motion area since only one motion cycle was analyzed. However, in the case of multiple motion cycle analysis, the finer mesh will have to be updated according to where the contact positions are between the moving sprung dynamic system and the support beam. Adaptive meshing, in which the area requiring the most definition changes with time [13, 14], may be needed in such analysis to ensure high accuracy with a reasonable computation cost.

Nomenclature

[M] [C] [g]

{f} {a} LNJ LN_I


fo
n ml, m 2 Wo

Y 9

FT
F k
C X

X I L

re
"C

removing
mbeam

fL,fR
ML, MR

structural mass matrix of a beam structural damping matrix of a beam structural stiffness matrix of a beam external load vector nodal displacement vector shape fimction vector first derivative of the shape functions with respect to x second derivative of the shape functions with respect to x magnitude of an applied concentrated force number of degrees of freedom (d.o.f.) the sprung and unsprung masses vertical displacement of rnz vertical displacement of m l gravitational acceleration force transmitted to the beam applied external force stiffness of a moving sprung mass system damping of moving sprung mass system local position of a moving load within a beam element global position of a moving load length of a beam element length of a beam structure fundamental period of a support beam time required for a moving system to travel across a beam mass of the moving point mass total beam mass equivalent nodal forces equivalent nodal loads

References
[1] P.L. George, Automatic Mesh Generation, Application to Finite Element Methods, Wiley, New York, 1991. [2] R.K. Wen and T. Toridis, "Dynamic behavior of cantilever bridges," ASCE J. Eng. Mech. Div. 88-EM-4, pp. 27-43, 1962.

144

J.R. Rieker et al./Finite Elements in Analysis and Design 21 (1996) 129-144

[3] J. Hino, T. Yoshimura, and K. Konishi, "A finite element method prediction of the vibration of a bridge subjected to a moving vehicle Load," J. Sound Vib. 96, pp.45-53, 1984. [4] Y.-H. Lin, M.W. Trethewey, H.M. Reed, Shawley, J.D. and S.J. Sager, "Dynamic modeling and analysis of a high speed precision drilling machine," ASME J. Acoust. Vib. 112, pp. 355-365, 1990. [5] F.V. Filho, "Finite element analysis of structures under moving loads," Shock Vib. Digest 10, pp. 27-35, 1978. [6] L. Fryba, Vibration of solids and Structures, Noordhoff, Leiden, 1972. [7] L. Meirovitch, Analytical Methods in Vibrations, Macmillan, New York, 1967. [8] Y.-H. Lin and M.W. Trethewey, "Finite element analysis of elastic beams subjected to moving dynamic loads," J. Sound Vib. 136, pp. 323-342, 1990. [9] H. Honda, T. Kobori, and Y. Yamaha, "Dynamic factors of highway steel girder bridges", Int. Assoc. Bridge Struct. Eng. P-98, pp. 57-75, 1986. [10] M. Olsson, "On the fundamental moving load problem," J. Sound Vib. 145, pp. 299-307, 1991. [11] Y.-H. Lin and M.W. Trethewey, "Finite element analysis and experimental model verification for structures subjected to moving dynamic loads", Proc. lOth Int. Modal Analysis Conf., San Diego, CA, 1992. [12] K. Miller, "Alternate modes to control the nodes in the moving finite element method," in: I. Babuska, J. Chandra, and J.E. Flaherty (eds.), Adaptive Computational Methods for Partial Differential Equations. SIAM, Philadelphia, 1983. [13] R.D. Cook, Concepts and Applications of Finite Element Analysis, Wiley, New York, 1976. [14] H.A. Dwyer, "The use of adaptive gridding", in: I. Babuska, J. Chandra and J.E. Flaherty (eds.), Adaptive Computational Methods for Partial Differential Equations, SIAM, Philadelphia 1983.

You might also like