Documentos de Académico
Documentos de Profesional
Documentos de Cultura
Abstract: Builders of passenger rail vehicles need methods for predicting vibrational behaviour
so that they can meet ride quality specifications. Conventional finite-element (FE) models used
for stress analysis may not be accurate enough for this because of imponderables such as the stiffness of spot-welded joints. One solution is to adjust FE models based on vibration measurements.
This model updating is now possible using commercially available software. Usually, the updating
is based on vibration tests carried out under laboratory conditions, which may produce different
results from normal operation. In this paper, two simple FE models of a railcar are updated using
vibration data obtained from a newly designed railcar under in-service conditions. Measurements were made on a railcar loaded to represent crush-loaded conditions and running down
track, with the only excitation that naturally occurring due to track roughness, engine excitation,
aerodynamic loading etc. Unlike standard laboratory vibration testing, this input excitation could
not be measured, so advanced forms of output-only analysis were required. By using these data
as input to a commercial updating package, the simple FE models have been updated and the
effectiveness of the technique assessed.
Keywords: operational modal analysis, finite-element model updating, in-service dynamic
model
1
INTRODUCTION
314
Fig. 1
Hunter railcar
The modal properties of a railcar, the natural frequencies, modal damping, and mode shapes, have
traditionally been obtained using experimental modal
analysis (EMA), in which the responses of the structure
to a contrived and measured input, such as impact
hammer excitation, are recorded. Techniques for EMA
have been well documented [1], but face a fundamental problem, which is that the excitation rarely represents the force distribution, magnitude, and type of
forcing the railcar will experience while in operation.
Moreover, the properties of the railcar might also be
different in service, for example because of the effects
of aerodynamic or passenger loading. One example
related to the rail vehicle described in this paper is that
the bogie suspension of the train is known to be nonlinear and so will behave differently running down the
track than when stationary in the workshop. The result
is that the modal properties estimated from EMA may
be poor estimates of the trains in-service behaviour.
Ideally, to obtain properties that apply when the
railcar is in service, EMA would be performed under
operational conditions. Unfortunately, it is not possible in this situation to measure the excitations, only the
responses, so traditional inputoutput modal analysis
techniques cannot be applied. Instead, a blind identification technique is required, so-called because it
uses only the responses. When applied to measurements obtained in service this is termed operational
modal analysis (OMA).
OMA, blind identification applied to mechanical
systems, has become increasingly popular in recent
years. New techniques have become available, which
allow the properties of a system to be identified for
the case when the system is in service, thereby ensuring that the properties identified correspond with the
working environment of the system. Some popular
techniques which are available in commercial software packages include frequency domain decomposition (FDD) [2, 3], stochastic subspace identification
(SSI) [4, 5], and Operational PolyMAX [6]. This paper
presents an application of the FDD algorithm to the
OMA of the Hunter railcar and the subsequent use of
the modal information identified by this technique.
Proc. IMechE Vol. 222 Part F: J. Rail and Rapid Transit
T
T
d k lk mk
dk lk mk
+
j k
j k
k = sub()
(1)
(2)
(3)
315
Experiment layout
3 TEST METHODOLOGY
3.1
On track test
Response measurements were recorded while the railcar was running along fully welded track at an approximately constant speed of 100 km/h. At this speed, the
JRRT105 IMechE 2008
316
4.1
OMA results
The resonance frequencies and modal damping identified using the EFDD technique are contained in
Table 1. The mode shapes corresponding to the
in-service modes in Table 1 are represented in Fig. 5.
It is worth emphasizing that these modal properties were obtained with no measurements of the input
excitation. One consequence of this is that the mode
shapes have arbitrary scaling, but this scaling can
be recovered either experimentally [11] or from a FE
model.
4.2
Fig. 3
The resonance frequencies and modal damping identified using the EFDD technique are compared in
Table 1 with those from a workshop modal test performed using impact hammer excitation on the same
vehicle but without the simulated passenger loading.
Note that the in-service resonance frequencies are
not simply obtained by lowering the resonance frequencies from the workshop test to account for the
added passenger mass. Although some modes have
decreased in frequency, others have increased and
the mode order has also changed. This interesting
phenomenon, which appears counter-intuitive, is not
without precedent in the literature [12]. In other applications, it has arisen because the passengers, and
in this case the water bottles, do not simply represent added mass, but form a springmassdamper,
system which adds DOFs, and hence modes, to the
system. Consequently, a structural mode can appear
to increase in frequency with the addition of the
passenger mass.
The rigid body modes were not identified in the
workshop modal test, possibly because they were too
low in frequency or too heavily damped to be adequately excited or measured by the impact hammer
modal test.
Table 1
3.3
EMA
Mode
Frequency
(Hz)
Damping
ratio (%)
Frequency
(Hz)
Damping
ratio (%)
Bounce
Pitch
First bending
First torsion
Second bending
Second torsion
1.3
1.5
6.0
13
14
18
8.2
16
13
1.2
3.2
1.8
8.0
3.7
3.6
3.4
6.7
9.8
18
15
Fig. 5
317
4.3
4.3.1
Table 2
FE model updating
Plate model
The carbody was modelled using simple plate elements and the suspension was represented by vertical
one DOF spring elements, as depicted in Fig. 6.
The car body was divided into nine sections and the
plate thickness of these sections, their elastic moduli, and Poissons ratio were first updated manually,
JRRT105 IMechE 2008
Mode
FEA
OMA
% Difference
MAC (%)
Bounce
Pitch
First bending
First torsion
Second bending
Second torsion
1.3
1.5
5.5
12
13
17
1.3
1.5
6.0
13
14
18
2.4
3.2
7.1
6.4
5.1
3.8
99.0
99.0
82.0
73.0
37.0
59.0
318
Fig. 7
Comparison of the modes shapes of the updated FE model (solid) and the Hunter railcar
(line): (L to R, T to B) bounce, pitch, first bending, first torsion, second torsion, and second
bending
Three-dimensional model
Fig. 8
Table 3
Mode
FEA
OMA
Bounce
Pitch
First bending
First torsion
Second bending
Second torsion
1.0
1.6
6.0
12
13
13
1.3
1.5
6.0
13
14
18
% Difference
21
5.5
0
7.2
3.7
32
MAC (%)
94
74
79
55
56
61
The model was divided into three sections, representing the floor, walls, and roof of the railcar. The
density and elastic moduli of these sections, and the
stiffness of the springs, were updated as for the plate
model. The modal properties of the updated model are
compared with those of the railcar in Table 3.
The percentage error between the resonances of
the updated FE model and the railcar are all relatively small, with the exception of the bounce and
second bending modes. The absolute error of the predicted bounce mode frequency is very small and would
Fig. 9
319
Comparison of the modes shapes of the second updated FE model (solid) and the Hunter
railcar (line): (L to R, T to B) bounce, pitch, first bending, first torsion, second torsion, and
second bending
320
CONCLUSIONS
10
11
12
APPENDIX
Notation
dk
G
Si
S
u 1i
Ui
W
zj
zm
k
k
j