Sie sind auf Seite 1von 42

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/228673297

Modal Testing and Finite-Element Model Updating of a Lively Open-Plan


Composite Building Floor

Article  in  Journal of Structural Engineering · April 2007


DOI: 10.1061/(ASCE)0733-9445(2007)133:4(550)

CITATIONS READS
30 253

3 authors:

Zoran Milosav Miskovic Paul Reynolds


University of Belgrade University of Exeter
24 PUBLICATIONS   101 CITATIONS    168 PUBLICATIONS   2,130 CITATIONS   

SEE PROFILE SEE PROFILE

Aleksandar Pavic
University of Exeter
165 PUBLICATIONS   2,482 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Realistic approaches to vibration serviceability assessment of office floors. View project

Utilization of by-products and recycled waste materials in concrete composites in the scope of sustainable construction development in Serbia: investigation and
environmental assessment of possible applications View project

All content following this page was uploaded by Paul Reynolds on 31 May 2014.

The user has requested enhancement of the downloaded file.


Modal testing and FE model updating of a lively open-plan composite

building floor

A. Pavic
Professor of Vibration Engineering, Department of Civil & Structural Engineering, University of
Sheffield, UK

Z. Miskovic
Research Associate, Department of Civil & Structural Engineering, University of Sheffield,
UK

P. Reynolds
Lecturer, Department of Civil & Structural Engineering, University of Sheffield, UK

Contact author: Professor Aleksandar Pavic


Department of Civil & Structural Engineering
University of Sheffield
Sir Frederick Mappin Building
Sheffield S1 3JD
E-mail: A.Pavic@Sheffield.ac.uk
Tel: 0114-2225721
Fax: 0114-2225791

1
Abstract

This paper presents results of a combined experimental and analytical approach to investigate modal

properties of a lively open plan office floor. It is based on state-of-the-art FE modelling, FRF-based

shaker modal testing, FE model correlation, manual model tuning and sensitivity-based automatic

model updating of a detailed FE model of this composite floor structure. The floor studied

accommodates a fully-furnished office. Such environments can be problematic regarding their

vibration serviceability. However, there is a lack of reliable information about their as built modal

properties and the ability of designers to predict them. Therefore this paper has two aims: (1) to

assess the ability to both predict and measure as accurately as possible the fundamental and higher

modes of floor vibration, and (2) to correlate and update the initially developed FE model of the

floor, so that its modes match as accurately as possible their measured counterparts.

It was found that even a very detailed FE model, the development of which was based on best

engineering judgement, missed the natural frequencies by as much as 10-20% in some of the first

four modes of vibration which possibly could be excited by walking. The key reasons for this were

both over- and under-estimation of the stiffness of the main composite beams, depending on the

beam location. This was probably caused by uncertainties due to visible cracking of the lightweight

concrete in the zone above the beams, effects of non-structural elements, such as false flooring, and

the inevitable uneven distribution of mass and stiffness in the real-life floor in operation. All of

these factors are difficult to model explicitly in the floor, so their aggregate effect was taken into

account via changes in the beam stiffness. This was found by performing a sensitivity-based FE

model updating in which the first four vertical bending modes of the floor were successfully

updated. Such updating was possible only after all perimeter walls were explicitly modelled.

2
The obtained updated properties are by no means a unique solution which minimises the difference

between natural frequencies and maximises the MAC values between the experimental and

analytical mode shapes. Rather, it is a reasonable set of modelling parameters which quantifies the

possible uncertainty when specifying FE modelling parameters for an open plan floor structure like

the one described in this paper.

Keywords: modal analysis, modal testing, model updating, correlation, sensitivity analysis,

composite floor, steel-concrete, floor structure

3
1 Introduction

Vibration serviceability has become a governing design criterion for many new building structures.

This is typically the case for modern light and slender structures occupied and dynamically excited

by human walking, such as long-span office floors. Vibration performance of open-plan steel-

concrete composite floors accommodating offices is becoming particularly problematic. When high

strength and lightweight steel-concrete composite floor construction is employed in open-plan

configurations, long-span and slender floor structures emerge. Although this type of floor tends to

satisfy the ultimate limit state and allowable static deflection criteria, excessive vibration may easily

occur due to human walking and create a vibration serviceability problem.

Not surprisingly, historically most of the vibration problems in building floors have been observed

on lightweight open plan and long span steel-concrete composite floors (Pavic and Reynolds, 2003).

Consequently, most of the available design guidance dealing with floor vibrations in the UK and

USA (Wyatt, 1989; Murray et al., 1997) was developed bearing in mind that particular type of

construction. These methods assume that a single floor vibration mode is excited by walking.

However, floor vibration modes tend to be closely spaced and when this happens a single mode

approach is inadequate. Therefore, a recent method for checking floor vibration proposed by the UK

Concrete Society (Pavic and Willford, 2005) makes full use of a multi-mode superposition

methodology to calculate floor response.

A common feature in all these methods is that they require an accurate prediction of one or more

modal properties of the floor - its natural frequencies, modal damping ratios and modal masses.

This prediction has to reflect the fully finished state of the floor, which is the relevant state when

checking its day-to-day operation with regards to vibration serviceability. The prediction requires

4
some form of mathematical modelling which is commonly performed by some form of finite

element (FE) analysis. However, there is a considerable lack of information on the reliability of the

FE prediction of multi-modal properties of fully furnished building floors. This is due to the fact

that measurement of more than one mode of vibration is notoriously difficult in floors which are in

operation. The effects of furniture, non-structural elements and rattling introduce non-linear effects

which make dynamic testing and its correlation with FE modelling difficult. Therefore, there is a

lack of information on the correlation between the measurements and modelling of more than the

fundamental mode of floor vibration (Pavic and Reynolds, 2003). Consequently, there is a

considerable uncertainty as to the ability of the state-of-the-art FE modelling, which employ best

engineering judgement, to predict multi-modal properties of fully furnished open plan office floors.

Bearing all this in mind, the aim of this paper is to present and compare modal results from high-

quality initial FE modelling and modal testing of a fully-furnished open plan office floor. This floor

is of composite steel-concrete construction. It is interesting to note that this floor is quite lively and

its vibrations due to walking have caused concern amongst the floor occupants. The comparison

between two sets of modal data will be achieved via FE model correlation. The FE model will then

be updated using a sensitivity-based updating, a technology well developed in the mechanical and

aerospace engineering disciplines (Friswell and Mottershead, 1995). Although powerful, due to the

nature of civil engineering design where there is no mass production of identical components and

every design is usually a unique structure, this technology is used mainly as a research tool in civil

structural engineering. Examples where the technology has been used are high-rise buildings (Lord

et al, 2004) and bridges (Brownjohn and Xia, 2000). To the best knowledge of the writers, this is

the first journal paper which describes results of a formal sensitivity-based FE model updating of a

building floor.

5
The paper firstly describes the floor structure and its initial FE modelling. It then describes modal

testing results and their correlation with the initial FE modelling. The need for the ‘manual’ tuning

of the FE model to reconcile it as much as possible with measurements is then described and carried

out. The manually tuned model is then updated automatically using the specialised FEMtools

software (DDS, 2004) and the outcomes of this exercise are discussed at the end of the paper.

2 Floor description

The test structure is the second floor in a three storey office building. It is an open plan composite

steel-concrete floor spanning approximately 11m between gridlines 1-2 and 2-3 (Figure 1). As usual

in this type of construction, steel decking supporting the in-situ cast concrete slab spans in the

direction orthogonal to the primary beams. At the time of testing, the floor was fully furnished as an

open-plan office (Figure 2). As a result, there were services and ceilings mounted beneath the floor,

there were additional services and access flooring mounted on top of the floor and there were office

furnishings (desks, filing cabinets, etc.). It is interesting to note that occupants had reported that the

floor was vibrating perceptibly under normal office usage conditions, typically during normal

walking of a single person across the floor. In addition to feeling the vibrations directly, they had

reported problems with computer monitors shaking. The vibration was perceived to be most

disturbing at the location between gridlines 2-3 and F-G, as shown on the floor plan (Figure 1).

3 Preliminary FE modelling

The first FE model in this analysis was developed based on best engineering judgement. This means

that only information typically available in design was used, such as construction and architectural

drawings and specifications for various construction materials used. The effects of uneven mass

6
distribution, stiffening effects of non-structural elements, such as false flooring and partitions

around the slab edge (Figure 1) and non-uniform structural properties, such as slab depth, have been

neglected. All beam-beam and beam-column connections have been assumed to be rigid due to

small deformations occurring in floor vibrations when even friction in connections is not overcome.

These assumptions are usual when developing FE models for checking vibration serviceability

(Pavic and Reynolds, 2003).

The main cellular beams (Hart and Peacock, 2003) were modelled using ANSYS BEAM4 elements.

The effects of openings in the beam web were neglected and a constant second moment of area of

the cross section was assumed. Following the parallel axis theorem, the second moment of area was

calculated with regard to the mid-plane of the composite slab supported by the beam (see inset in

Figure 1).

The composite floor spanning typically 4.8 m between the main beams (see inset in Figure 1) was

modelled using ANSYS SHELL63 orthotropic elements. This is a four-node element capable of

resisting both in-plane and out-of-plane loads. The contribution of 1.2 mm corrugated steel

sheeting, acting as a part of the composite slab, was neglected. This is normal practice when

analysing this type of slab. The slab was 140 mm thick and made of lightweight in-situ cast

concrete. Considering the existence of 140 mm deep ribs supporting 60 mm deep slab every

300 mm (Figure 1), the bending stiffness of this orthotropic element in the less stiff direction

perpendicular to the ribs, was assumed to be 10% of the main stiffness. This was done following a

procedure proposed by Szilard (1974). It was assumed that the SHELL63 elements had constant

thickness. This thickness was determined from the condition that 1 m width of the element in the

direction of the ribs has the same bending stiffness as 1 m width of the real ribbed section (see inset

in Figure 1). This led to an equivalent constant thickness of 122 mm assuming lightweight concrete

7
properties for the SHELL63 element. The orthogonal bending stiffness was modelled by adjusting

the corresponding modulus of elasticity and reducing it to 10% of the dynamic modulus of

lightweight concrete. The concrete density was also adjusted to ensure proper distribution of mass

in shell elements having ‘artificially’ established thickness as previously described.

The main beams in the floor were supported by steel columns modelled using ANSYS BEAM4

elements. The column height was assumed to be 3.2 m above and below the floor levels which is

the same as the corresponding storey heights. The columns were fixed at their ends. The whole of

the FE model was restrained horizontally at two opposite corners of the floor plan to prevent

horizontal motion of the whole floor.

The modulus of elasticity assumed for lightweight concrete was 20.5 GPa, as appropriate for short

term loads and lightweight concrete (Lawson and Wickens, 1992). The steel dynamic modulus was

assumed as 205 GPa. The lightweight concrete density was assumed to be 1,800 kg/m3. Steel

density was assumed to be 7,850 kg/m3. Imposed dead loading due to services, ceilings, false floor

and furniture (Figure 2) was assumed to be 120 kg per square meter of the floor. Poisson ratio for

lightweight concrete was assumed to be 0.2 and for steel 0.3.

The developed FE model was then used to calculate modal properties of the floor. There were 53

modes of floor vibration between 5.6 and 20 Hz. Such high modal density is normal for building

floors having strong orthotropic properties caused by the existence of main load-bearing beams

running in one direction and a relatively weak composite slab spanning orthogonal to the beams.

Closely spaced modes of vibration of building floors also occur due to repetitive geometry and

overall symmetries existing in the floor structure. This feature is the main reason why a mode

8
superposition is a prudent way forward when calculating floor response, rather than a SDOF single-

mode analysis.

4 Modal testing

The modal tests were performed using a single APS Dynamics Model 113 electrodynamic shaker as

an excitation source (Figure 3). This was operated in reaction mode, meaning that the shaker was

placed on the structure and the dynamic force was generated by accelerating a reaction mass

assembly attached to the shaker armature. The excitation force was measured by attaching an

accelerometer to the shaker armature and by multiplying the measured acceleration by the mass of

the armature and reaction mass assembly.

The structural responses were measured using Endevco Model 7754-1000 piezoelectric

accelerometers, which were mounted on base plates that could be levelled to ensure proper

alignment (Figure 3). The base plates, in turn, were positioned on the sub-floor, as shown in Figure

3, to avoid them picking up local vibrations of the access flooring (Reynolds & Pavic, 2003). These

accelerometers have a nominal sensitivity of 1000 mV/g, a noise floor of less than 10 μg and a low

frequency limit of approximately 0.1 Hz.

Digital data acquisition was performed using a Data Physics DP730 portable spectrum analyser,

which has 24 24-bit input channels and four output channels. One of the output channels was used

to provide the shaker excitation signal and five input channels were used to acquire the shaker

excitation and four floor acceleration response time histories. The analyser provided immediate

calculation of FRFs so that the quality of the measurements could be assessed as the testing

progressed.

9
4.1.1 Test procedure

To build a complete ‘picture’ of the vibration properties of the floor structure, including mode

shapes, it is necessary to develop a grid of test points at which FRF measurements are made. The

size of this grid is dictated by the number of channels of instrumentation and the amount of time

available for the testing.

For these tests, it was decided that a grid of 79 test points should be utilised as illustrated in Figure

4. The slightly irregular nature of the grid was required to avoid obstacles on the floor, such as

desks, cupboards, etc.

This grid was deemed to be sufficient to measure the modes of vibration of interest. Four reference

accelerometer locations were selected at test points 7, 15, 39 and 50.

The number of reference accelerometers (four) was dictated by their availability. The fifth

accelerometer was on the shaker, so, as previously mentioned, in total five of the accelerometer

channels were sampled using the DP730 spectrum analyser.

After a few trial measurements, the modal testing commenced by placing the shaker at the first test

point and measuring the FRFs between that point and all four response locations. Once that had

been completed, the shaker was then moved to the next point and the next set of FRFs was

measured. The measurements were completed when FRFs had been measured with the excitation at

all 79 test points.

Each FRF was measured by applying a chirp excitation through the shaker and by measuring

simultaneously both the excitation and the corresponding response signals. A chirp signal is simply

10
a fast sine sweep through a predefined frequency range. For this round of tests, a range of 3 to 19

Hz was selected. Figure 5 shows typical excitation and response signals measured when applying a

chirp excitation and Figure 6 shows their auto-spectral densities (ASDs). It can be seen that there is

energy in the excitation at all frequencies of interest (approx 3-19 Hz) and that the response ASD

has peaks that probably correspond to modes of vibration.

By performing several averages and using the H1 frequency response function estimator (Ewins,

2003), it is possible to reduce the effects of extraneous unmeasured excitation. By performing some

exploratory measurements, it was decided that 5 averages provided FRFs of sufficient quality. A

typical averaged FRF is presented in Figure 7.

4.1.2 Modal parameter estimation

After the acquisition of FRF data had been completed, the modal parameter estimation was

performed to estimate modes of vibration of the floor structure. This was performed very quickly on

site to ensure that the acquired data were of sufficient quality and that there were no points that

required re-measurement. More detailed and comprehensive modal parameter estimation was

performed following return from site.

In principle, curve fitting of FRFs corresponding to each of the four reference points may yield

modal parameters of, say, the first mode of vibration which could be slightly different. This happens

as different reference points produce FRF data quality which varies between modes. For example, a

reference which is close to a nodal line for a particular mode will yield less reliable modal data for

that mode due to the fact that the mode is not well measured at that reference.

11
The results from the most reliable modal parameter estimations are presented in Figure 8. The data

in this figure demonstrate that the lowest four measured modes of vibration engage predominantly

the floor area between axes 2 and 3.

5 Manual model updating

By comparing the previously mentioned results from the initial FE modelling and modal testing, it

became apparent that, despite best engineering judgement and quite detailed FE modelling, the

numerical results in general overestimated the measured natural frequencies. Also, some of the

modes appeared to be out of sequence with the measured modes.

To reconcile these differences the initial FE model required updating with an aim to understand

better the shortcomings of the FE modelling based on best engineering judgement. The flowchart

outlining the process of updating is shown in Figure 9. This process had two main parts: the

‘manual’ model tuning and ‘automatic’ model updating.

The aim of the ‘manual’ tuning was to reduce as much as possible the initially large differences

between the analysis and testing results, which cannot be reconciled via sensitivity-based automatic

updating. Also, the result of the manual tuning should be an FE model which resembles as much as

possible the real structure. All this is done to produce an FE model suitable for automatic updating.

Without this manual tuning, due to large differences between the two sets of modal results (i.e.

large differences between physical properties of the model and real-life structure) automatic

sensitivity-based updating would have been impossible. This is a common problem in the FE model

updating pertinent to civil engineering structures, as already noticed by Brownjohn and Xia (2000).

12
In principle, there are two stages in manual tuning: model refinement and parameter adjustment. In

the former, the preliminary model is changed in terms of adding new features, such as elements

(including springs), improved and more detailed geometry as well as changed boundary conditions.

In the latter, various uncertain modelling parameters in the already refined model are varied by trial

and error within reasonable limits to improve the matching with the test data.

In this particular case, the main model refinement step was to include brick walls running at the

perimeter of the floor area and continuous from the foundations of the building. Although not part

of the main load-bearing steel frame, walls were included as they may have significant effect on the

floor dynamic behaviour. In particular, their stiffness may affect the frequency, shape and sequence

of modes of vibration (Pavic and Reynolds, 2003). Similar to columns, the top and bottom edges of

the walls were assumed to be fixed.

As to the material properties, only one change was made to account for low-strain dynamic effects

in the lightweight concrete: dynamic modulus was increased from 20.5 to 30 GPa, which is a more

reasonable assumption. The FE model developed at the end of the manual tuning is shown in Figure

10 and this was used in the automatic updating as described in the next section.

The developed model had 29,161 elements, 27,185 nodes (with six degrees of freedom at each

node) and 153,256 active degrees of freedom. Modal analysis was performed using Lanczos

eigenvalue extraction algorithm. This led to 36 modes of vertical floor vibration having frequencies

between 5.78 Hz and 20 Hz. The first four modes are shown in Figure 11. Note that in this figure

walls are omitted for clarity.

13
Table 1 shows a comparison between the test and FE results using the above described manually

tuned model. It is interesting to note that adding the perimeter and other masonry walls and

modelling them as SHELL63 elements and using best engineering judgement apparently further

increased the difference between the FE model and the measurements pertinent to the fundamental

mode of vibration. It increased from 5.6 Hz in the initial FE model to 5.78 Hz in the manually

adjusted FE model, compared with the 5.2 Hz measured. However, the inclusion of the walls was

done as it became obvious that only a model featuring walls was leading to the right sequence of FE

modes and a meaningful set of updated parameters produced during automatic updating. This

updating will be described later and it suffices to say here that first four modes, rather than only the

fundamental mode, were aimed to be reconciled with their measured counterparts. This kind of

reconciliation is difficult to achieve, but is also the key reason why walls had to feature in the

finally adjusted FE model ready for automatic updating.

5.1 Correlation analysis

Using pairing shown in Table 1, the aim of the FE model updating was to minimise the difference

between the first four measured and FE calculated natural frequencies and maximise the Modal

Assurance Criterion (MAC) value (Friswell and Mottershead, 1995) between the experimental and

FE calculated mode shapes.

5.2 Sensitivity analysis

Having selected the natural frequencies and mode shapes of the first four modes as ‘targets’ (DDS,

2004) for matching, a number of uncertain modelling parameters was selected for sensitivity

analysis with regard to the given natural frequency targets. In the initial selection of these

parameters the following were included: bending stiffness of all beams, stiffness of slab elements

14
directly above the main beams, bending stiffness of columns, thickness and orthotropic properties

of slab shell elements away from the main beams (via dynamic modulus of lightweight concrete),

column stiffness and effect of dead mass. At the end, 15 most significant ‘global’ parameters were

selected for updating. These parameters are called ‘global’ because they feature in all finite

elements having the same properties (DDS, 2004). Therefore, a change of any of the selected

parameters will affect not just one, but a range of elements in the model. The selected parameters

are listed in Table 2.

In Table 2, RHO is concrete density including the effects of steel sheathing and additional non-

structural mass, IY beam or column second moment of area about the main axis, H is the

orthotropic shell thickness, and EX and EY are dynamic moduli of elasticity in the main (i.e. rib)

and lateral slab direction, respectively. Element IDs from Table 2 are shown in Figure 1. Finally,

lower and upper bounds are variations of the parameters allowed during automatic model updating,

given as a percentage of the starting value. These limits have been set to account for the modelling

uncertainty due to uneven mass and stiffness distributions. These are caused by low-strain

behaviour and effects of friction which is not overcome, as well by the effects of the false flooring,

visible cracking in the lightweight concrete slab, presence of furniture and other features which

exist in real-life structures but are difficult to incorporate explicitly in the FE model.

Figure 12 shows the sensitivity of the four natural frequencies (‘responses’ in FEMtools) to a 1%

variation of the selected 15 updating parameters listed in Table 2. It can be seen that all natural

frequencies are, in general, most sensitive to changes in parameters 1 (RHO), 13 (H), 14 (EX) and

15 (EY).

15
6 Automatic model updating

Automatic updating was carried out using the FEMtools software through 11 iterations. After each

iteration a new sensitivity matrix was calculated using the properties of the interim FE model and a

new iteration was carried out using this model as the starting point. Figure 13 shows how the

calculated natural frequencies approached the measured ones (i.e. how the difference between them

reduced) through the iterations. The initial differences between the test and analysis results shown

in Table 1 generally reduced through the process of automatic updating, the final results of which

are shown in Table 3. It can be seen that the maximum difference in the natural frequency reduced

from 11.2% to 5.8% (for the first mode of vibration) and that the lowest MAC value increased from

54% to 65% (for the third mode of vibration).

The required percentage change of parameters to achieve this improved matching of natural

frequencies is shown in Figure 14. The modes of the finally updated FE model are shown in and

these shapes visually correlate well with their analytical counterparts in Figure 8. This is not

surprising considering that the lowest MAC value is 65% (Table 3) which is quite acceptable for a

civil structural engineering application.

7 Interpretation of parameters in updated FE model

One of the problems with the current limited literature on FE model updating in civil structural

engineering is that a physical interpretation of the updated results is seldom provided. However, this

interpretation is crucial for the updating exercise to have a purpose. Therefore, an attempt to provide

such an interpretation is made here. Table 4 contains a comparison of the initial and updated values

of the 15 selected updating parameters. It can be seen that eight out of 11 uncertain beam

parameters required maximum allowed change of ±20% indicating a considerable level of

16
uncertainty in the stiffness of the cellular beam. This should not be surprising considering the

irregularities in the composite slab above the main beams caused by a number of reasons. These

include significant amount of cracking of the lightweight concrete slab which existed above some

beams, the presence of the ends of the corrugated sheeting and difficulties with concreting around

dowels. Also, other previously mentioned factors, such as the effects of the false floor and uneven

distribution of mass, could be the reason for the changes of the beam stiffness parameters required

by the automatic FE model updating. All these factors are difficult to model explicitly but

undoubtedly have the ability to affect modal properties of the floor. Considering the overall good

quality of the updated model with the first four modes matching well their experimental

counterparts, the ±20% change could be used as an upper and lower bounds when performing

sensitivity study of the FE model, which should be done in any dynamic analysis in order to assure

its quality (NAFEMS, 1993).

Changes of all orthotropic slab properties, RHO, H, EX and EY (Table 4) are within reasonable

boundaries. Updated values of the density and thickness are within 15% from the original values,

which means that they were well estimated before the updating procedure. The most uncertain slab

parameters, the modulus of elasticity in the main (EX) direction was increased by almost a quarter,

probably taking into account the uncertainties related to the low-strain behaviour and friction effects

related to the composite slab. On the other hand, the lateral (EY) modulus of the slab required

reduction of 8.4% due to similar, difficult to model explicitly, reasons.

8 Conclusions

This paper has presented a relatively rare exercise comprising FE modelling, modal testing and FE

model correlation and updating of a real-life fully-furnished open plan office floor, which was

17
reportedly lively under walking excitation. This was done as there is a growing need to consider

more than a single mode of floor vibration when checking floor vibration serviceability. This is

particularly the case in strongly orthotropic composite floors, like the one which was investigated in

this paper, where closely spaced modes of vibration may occur. The key issue here is to assess

ability to predict as accurately as possible not only the fundamental but also the higher modes of

vibration.

It was demonstrated that a fairly complex FE model of the floor developed on the basis of best

engineering judgement could easily yield natural frequencies which were 10-20% above their

measured counterparts. This is not conservative as lower natural frequencies in reality may lead to

more frequent noticeable floor vibrations.

It was demonstrated that the two-phase FE model updating, which consisted of manual tuning and

sensitivity-based automatic updating, was absolutely essential for the successful updating. The

manually updated FE model had to feature perimeter and other walls to be able to match not only

the fundamental, but the lowest four modes of vibration of this floor which may be excitable by

walking.

After reviewing the updated parameters, it became clear that major adjustments were required for

the stiffness (via second moment of area) of the main cellular beams supporting the composite slab.

Adjustments varying from -20% to +20% were required indicating the level of uncertainty when

modelling these beams using best engineering judgement. This is probably caused by the

irregularities caused by the visible but difficult to model explicitly cracking of lightweight concrete

in the composite slab above the main beams. Knowing this, a sensitivity study of the FE models

used in future to predict vibration behaviour of similar floors is prudent. When doing this, vibration

18
performance could be checked for various scenarios when stiffness of one or more of the main

beams varies plus or minus 20%. Modern FE codes like ANSYS have a programmable facility

which would allow a quick evaluation of the possible outcomes of this sensitivity study.

Finally, it should be stressed that the obtained updated properties are by no means a unique solution

which minimises the difference between natural frequencies and maximises the MAC values

between the experimental and analytical mode shapes. Rather, it is a reasonable solution which

quantifies possible uncertainty when specifying FE modelling parameters for a floor structure like

the one described in this paper.

9 Acknowledgements

This paper was prepared with the financial support from the UK Engineering and Physical Sciences

(EPSRC) research grant GR/S14924/01 entitled Investigation of the As-Built Vibration Performance

of System Built Floors for which the writers are grateful.

19
List of tables

Table 1: Comparison of numerical and experimental natural frequencies after manual model

adjustment and tuning.

Table 2: Parameters selected for model updating (modified)

Table 3: Results of automatic FE model updating.

Table 4: Change of parameters selected for model updating (modified)

20
List of figures

Figure 1: Floor Plan. The problematic area in the bottom right corner of the plan is shaded

Figure 2: General view of the ‘lively’ area of the office floor between axes 2-3 and F-G

Figure 3: APS Dynamics Model 113 shaker and Endevco model 7754-1000 accelerometer

Figure 4: Floor test grid.

Figure 5: Typical chirp excitation and response signals, measured at test point 7.

Figure 6: ASDs of typical excitation and response signals

Figure 7: Typical FRF (point mobility at TP7)

Figure 8: Summary of measured natural frequencies and mode shapes.

Figure 9: Flowchart describing key steps in manual and automatic model updating

Figure 10: FE model after adjustment

Figure 11: Initially calculated modes using the preliminary FE model.

Figure 12: Sensitivity of the selected target responses to 1% increase in the values of each of the 15

selected updating parameters(modified)

21
Figure 13: Target response (natural frequencies) tracking through iterations(modified)

Figure 14: Change of parameters through nine iterations(modified).

22
Tables

Table 1: Comparison of numerical and experimental natural frequencies after manual model

adjustment and tuning.

FE modes Test modes MAC Difference of


No. Freq. No. Freq. [%] frequencies
[Hz] [Hz] [%]
1 5.78 1 5.20 78 11.2
2 6.10 3 6.40 60 -4.7
3 6.13 2 5.60 54 10.4
4 6.72 4 7.00 56 -3.7

23
Table 2: Parameters selected for model updating (modified)

Parameter Element Boundary [%] Starting Unit


No. Type ID Lower Upper Value
1 RHO 201 -15 15 3.28E+03 kg/m3
2 IY 17 -20 20 5.91E-07 m4
3 IY 24 -20 20 4.98E-05 m4
4 IY 25 -20 20 4.98E-05 m4
5 IY 52 -20 20 8.30E-04 m4
6 IY 54 -20 20 1.33E-03 m4
7 IY 56 -20 20 3.04E-04 m4
8 IY 57 -20 20 8.09E-04 m4
9 IY 58 -20 20 9.47E-04 m4
10 IY 59 -20 20 7.04E-04 m4
11 IY 61 -20 20 3.56E-04 m4
12 IY 101 -20 20 1.14E-04 m4
13 H 201 -15 15 1.22E-01 m
14 EX 201 -30 30 3.00E+10 N/m2
15 EY 201 -30 30 3.00E+09 N/m2

24
Table 3: Results of automatic FE model updating.

Mode shape pair FEA EMA Difference MAC


[Hz] [Hz] [%] [%]
1 5.52 5.22 5.8 96
2 5.87 5.65 3.9 68
3 6.32 6.35 -0.5 65
4 7.04 7.04 0.0 75

25
Table 4: Change of parameters selected for model updating (modified)

Parameter Element Value Change Unit


No. Type ID Starting Final [%]
1 RHO 201 3.28E+03 3.07E+03 -6.5 kg/m3
2 IY 17 5.91E-07 7.09E-07 20.0 m4
3 IY 24 4.98E-05 5.97E-05 20.0 m4
4 IY 25 4.98E-05 5.14E-05 3.3 m4
5 IY 52 8.30E-04 9.42E-04 13.5 m4
6 IY 54 1.33E-03 1.06E-03 -20.0 m4
7 IY 56 3.04E-04 2.43E-04 -20.0 m4
8 IY 57 8.09E-04 6.47E-04 -20.0 m4
9 IY 58 9.47E-04 1.14E-03 20.0 m4
10 IY 59 7.04E-04 8.11E-04 15.1 m4
11 IY 61 3.56E-04 2.84E-04 -20.0 m4
12 IY 101 1.14E-04 9.10E-05 -20.0 m4
13 H 201 1.22E-01 1.40E-01 15.0 m
14 EX 201 3.00E+10 3.69E+10 22.9 N/m2
15 EY 201 3.00E+09 2.75E+09 -8.4 N/m2

26
Figures

Figure 1: Floor Plan. The problematic area in the bottom right corner of the plan is shaded

27
Figure 2: General view of the ‘lively’ area of the office floor between axes 2-3 and F-G

28
Access floor

Accelerometer
to measure force

Shaker
Accelerometer
mounted
on baseplate
on sub-floor

Sub-floor

Figure 3: APS Dynamics Model 113 shaker and Endevco model 7754-1000 accelerometer

29
Figure 4: Floor test grid.

30
Excitation at Test Point 07
200
100
Force [N]

0
-100

0 5 10 15 20

Response at Test Point 07


Acceleration [m/s 2]

0.05
0
-0.05

0 5 10 15 20
Time [s]

Figure 5: Typical chirp excitation and response signals, measured at test point 7.

31
Excitation ASD
600
ASD [N2/Hz]

400

200

0
0 5 10 15 20
-4
x 10 Response ASD
ASD [(m/s 2)2/Hz]

1.5
1
0.5
0
0 5 10 15 20
Frequency [Hz]

Figure 6: ASDs of typical excitation and response signals

32
-4

FRF Modulus [m/s 2/N]


x 10
8
6
4
2
0
2 3 4 5 6 7 8 9 10 11 12
Frequency [Hz]
FRF Phase [degrees]

180
90
0
-90
-180
2 3 4 5 6 7 8 9 10 11 12
Frequency [Hz]

Figure 7: Typical FRF (point mobility at TP7)

33
Mode 1: f1 =5.20 Hz; ζ1 = 1.8% Mode 2: f 2 =5.60 Hz; ζ 2 = 2.8%

Mode 3: f 3 =6.40 Hz; ζ 3 = 3.0% Mode 4: f 4 =7.00 Hz; ζ 4 = 1.9%

Figure 8: Summary of measured natural frequencies and mode shapes.

34
Figure 9: Flowchart describing key steps in manual and automatic model updating

35
Figure 10: FE model after adjustment

36
Mode 1: 5.78 Hz Mode 2: 6.10 Hz

Mode 3: 6.13 Hz Mode 4: 6.72 Hz

Figure 11: Initially calculated modes using the preliminary FE model.

37
0.2

0.1

0 [%]

-0.1

4 -0.2
3 -0.3
Response
2
-0.4
1 14
4 6 8 10 12
1 2 Parameter

Figure 12: Sensitivity of the selected target responses to 1% increase in the values of each of the 15

selected updating parameters(modified)

38
12.5
10
7.5
5
[%]
2.5
0
-2.5
4
-5
3 11
9
Response 2 7
5
1 3 Iteration
1

Figure 13: Target response (natural frequencies) tracking through iterations(modified)

39
30
20
10
[%]
0
14
12 -10
10
8 -20
Parameter 6 11
4 5 7 9
2 1 3 Iteration

Figure 14: Change of parameters through nine iterations(modified).

40
References

ANSYS Inc. (2002). ANSYS Documentation Manuals, Version 6.1. Author, Canonsburg, USA.
Brownjohn, J. M. W. and Xia, P. (2000). “Dynamic Assessment of a Curved Cable-Stayed Bridge
by Model Updating.“ ASCE Journal of Structural Engineering, 126(2), 252–260.
DDS. (2000). FEMtools User’s Guide, Version 3.0.03, Dynamic Design Solutions NV, Leuven,
Belgium.
Ewins, D. J. (2003). Modal Testing: Theory, Practice and Application. Research Studies Press.
Baldock, England
Friswell, M. I., and Mottershead J. E. (1995). Finite Element Model Updating in Structural
Dynamics, The Netherlands Kluwer Academic Publishers, Dordrecht.
Hart, A and Peacock, P. (2003). Multi-storey buildings. Chapter 2 in The Steel Designers’ Manual,
6th Edition. Blackwell Science, Malden, USA.
Lawson, M. and Wickens, P. (1992). Composite Beams. Steel Designers Manual, 5th Ed., Steel
Construction Institute, Blackwell, London.
Lord, J-F., Ventura, C. E. and Dascotte, E. (2004). “Automated Model Updating using Ambient
Vibration Data from a 48-Storey Building in Vancouver. “ Proc. of the 22nd International
Modal Analysis Conference (IMAC XXII), Detroit, USA.
Murray, T. M., Allen, D. E. and Ungar, E. E. (1997). Floor Vibrations due to Human Activity,
American Institute of Steel Construction, Chicago.
NAFEMS. (1993). A Finite Element Dynamics Primer. Edited by D. Hitchings. National Agency
for Finite Element Methods and Standards. Glasgow, UK.
Pavic, A. and Reynolds, P. (2003). “Vibration Serviceability of Long-Span Concrete Building
Floors. Part 1: Review of Background Information. “, The Shock and Vibration Digest,
34(3), 187–207.
Pavic, A. and Willford, M. (2005). Vibration Serviceability of Post-Tensioned Concrete Floors.
Appendix G of Post-Tensioned Concrete Floors, Design Handbook, 2nd Ed.,. Technical
Report 43, Concrete Society, Slough, UK. pp. 99-107. ISBN 1-904482-16-3
Szilard, R. (1974). Theory and Analysis of Plates: Classical and Numerical Methods, Prentice-Hall,
Upper Saddle River, New Jersey.
Wyatt, T. (1989). Design Guide on the Vibration of Floors, SCI design guidance 076, Steel
Construction Institute, Ascot, UK.

41

View publication stats

Das könnte Ihnen auch gefallen