You are on page 1of 5

CLEANROOM

Improved methodologies for the prediction of footfall-induced vibration


Michael Willford, Arup, London, UK, & Caroline Field, Arup, San Francisco, USA
Floors for semiconductor facilities must generally have high natural frequencies so as to avoid the possibility of resonant response. The following sections describe the basis, application and validation of the new Ar up prediction methods, with particular emphasis on higher-frequency floors.
ABSTRACT

The development of sophisticated semiconductor technologies requires ultralow vibration research and production environments. Human footfall is a significant source of vibration, and if its effects are not assessed accurately during the design of a facility, the workspaces may be rendered unusable for sensitive equipment. In view of the importance of achieving an adequately low vibration environment in modern facilities it is perhaps surprising that the design methods employed by most structural and vibration engineers comprise very simple and semiempirical hand calculations based on research available in the 1970s. In order to provide clients with greater confidence in the performance of their structures, Arup has developed new and more accurate performance-based prediction procedures incorporating recent, comprehensive, experimental research into footfall forces, and taking advantage of the capabilities of modern design-office computer software. The new methods are not limited by the approximations and inaccuracies inherent in the old empirical approaches, and extensive calibration against field measurements has shown them to be far more reliable than other methods currently in use. Use of these methods can also bring substantial cost benefits in some cases.

Background
We d e ve l o p e d o u r p r e d i c t i o n m e t h o d o l og y t o ove rc o m e t h e inconsistencies we have perceived in the commonly used methods [24]. In particular, our experience does not concur with the indication of formulae of the form [2,3]: Vrms = 1.0 x 1010 , kf0N

Introduction
In order to meet specified vibration criteria under footfall loading, it is necessary to be able to predict, as accurately as possible, how a particular floor structure will vibrate when persons walk on it. Therefore, the first essential ingredient to the design process must be a clear understanding of the dynamic forces imparted by persons walking across a floor, and the second component is an accurate method of predicting how a floor structure will respond to those forces. Our methods are based on a large number of direct measurements of forces applied by pedestrians to an instrumented floor. Typical footfall rates are between 1.5 and 2.5 steps per second (Hertz). As expected, contact forces vary widely between different persons, and with walking speed, stride length and footwear. A statistical approach is adopted to account for this variability. The vibration of structures subjected to time-varying forces can be predicted by well-established pr inciples of structural dynamics. Any structure has a number of natural modes, which are favored patterns of vibration. Each mode has a shape (defining the relative amplitudes of motion at different points on the structure) and a natural frequency of vibration (cycles per second, Hertz) at which it occurs. The ratio between the frequency of the applied forces and the natural frequencies of the modes of a structure
64

is a key factor in determining the extent to which the structure vibrates. If the natural frequencies of a floor are high compared to the footfall rate (10 Hertz or higher) then the vibration generated is essentially a transient response to each separate footfall impact. However, if the floor has natural frequencies close to the footfall rate, then the vibration generated by one footfall is reinforced by the response to following footfalls, leading to a build-up of vibration known as resonance. Resonance can also be generated in floors having natural frequencies of two, three and four times the footfall rate, known as the first four harmonics of the force. The magnitude of resonant response is sensitive to the exact footfall rate such that when the footfall rate (or the frequency of any of its first four harmonics) is very close to a natural frequency of the structure, then the response will be much higher than that occurring at slightly lower or higher footfall rates. The transient response of higher-frequency floors (those having natural frequencies more than four times the footfall rate) is not very sensitive to the close synchronization of a footfall rate to a natural frequency. The distinction between low-frequency (resonant) and high-frequency (transient, nonresonant) floor responses leads to the need for two different characterizations of dynamic footfall forces and two different vibration calculation procedures.

where Vrms is the root mean square vibration velocity, k is the stiffness and f0 the natural frequency, since this implies that for a structure of given stiffness, the mass should be minimized (since this would then maximize the natural frequency). We have found mass to be beneficial. We also considered the reduction factors N (1, 2, or 4 to allow for various degrees of fit-out) to be crude corrections and to some extent subjective. We were, therefore, not confident that methods of this type could be applied with confidence to the optimization of floor designs of new structural forms to achieve everreducing vibration targets. For this reason we developed a method based on first principles, using the best data available on footfall forces, and well-established responsecalculation methods of general applicability to the prediction of the dynamic response of structures. With this basis we believed that our method would be applicable to the majority of structures, simple or complex, of conventional or novel structural form, and constructed of any material. The method presented in this article has evolved and been validated over 15 years and has been used extensively world-wide within Arups design practice for the past five years.

Measured footfall forces


We derived the input forcing function from an extensive survey of measured
WWW.FABTECH.ORG

SEMICON D U C T O R F A B T E C H 28 TH E D I T I O N

CLEANROOM

footfall forces. The principal source of data is the research of Kerr [1], who measured 880 single footfall force time histories from over 40 individuals walking on an instrumented force plate set into a floor at a range of walking speeds. Kerr provided his raw digitized data to the authors for further processing. We also took footfall forcetime history data from other published sources [58]. Since Kerr had measured only single footstep forces the first stage of our analysis was to synthesize forcetime curves for continuous walking by overlapping single footfall measurements to simulate a number of successive footsteps. This was done for all of the 880 sample single-footstep measurements. The continuous records were then analyzed to obtain parameters suitable for predicting the response of low-frequency and highfrequency floors. For low-frequency floors, we derived the harmonic components of the force functions to permit the response of the floor to many successive footfalls to be calculated by steady-state resonance calculations. For higher-frequency floors, we were concerned with identifying the peak velocity of a floor due to each separate footfall of the continuous record.

Harmonic components of footfall force


Figures 1(a)(c) show the dynamic load factors (DLFs defined as harmonic force amplitude/weight of person) for the first three harmonics of footfall forces derived from all the measured data. The mean trend is indicated by a line through the individual data points. It can be seen that the first harmonic force increases strongly with footfall rate, and that there is modest variability between individual measurements. For the second third harmonics the increase of DLF with walking rate is slight, but there is a very significant scatter in the data.

Effective footfall impulse


The measured vibrationtime history of a higher-frequency floor subjected to walking excitation may typically look like that of Figure 2. Here the vibration of the floor can generally be characterized by an initial peak, associated with the impact of one foot, followed by vibration at the natural frequency of the floor decaying at a rate determined by the damping present. As the next impact is made another decaying transient is generated. The degree of decay and the phasing of the second impact are assumed to be such that, on average, there is no build up of vibration due to successive footfalls. This
Figure 1(a)(c) Dynamic load factors from measured footfall force.

provides a conceptual model for the prediction of the vibration of higherfrequency floors. If the magnitude of the initial velocity peak can be determined, the remainder of the idealized transient response can be synthesized using the analytical expression for the decay of vibration caused by an impulse. The initial velocity of a floor mode under an impulsive action can be calculated by dividing the numerical value of the impulse by the numerical value of the modal mass. If the modal

mass is 1.0 then the initial velocity is numerically equal to the applied impulse. This gives a method to determine the effective impulse of a particular footfall forcetime history. We applied each of the 880 continuous walking forcetime histories to an analytical representation of a series of single floor modes of different natural frequency (but each with a mass value of 1.0) and calculated the velocity response of the mode in the time domain. The peak velocities of the vibration that was induced in each floor
65

S E M I C O N D U C T O R F A B T E C H 28 TH EDITION

CLEANROOM

This procedure produces a simulated vibration trace such as shown in Figure 4, which can then be processed in exactly the same way as a digital vibration record physically measured on a floor to determine the peak and root mean square (rms) velocity, displacement or acceleration. It can also be filtered and processed in third-octave or other frequency-domain components. Since the prediction is based on the typical effective impulse the vibration simulated is identical for each footfall. In reality there is substantial variability that leads to less uniform traces as in Figure 2. However, the simulated trace replicates well the overall vibration level.
Figure 2. Typical measured vibration trace due to walking on high-frequency floor.

mode were then extracted, and these are numerically the values of the effective impulse for that footfall record as the floor frequency varies. This analysis approach is identical to the derivation of a response spectrum for a large series of earthquake ground motion records. The mean results of this analysis are presented in Figure 3, with results for different walking speed ranges (in 0.1 Hertz increments) plotted separately. It can be seen that the effective impulse reduces as floor natural frequency increases and as footfall rate decreases, both of which are intuitively correct. The coefficient of variation of the effective impulse value between the different force time histories was found to be about 0.4. For design we propose a formula for the effective impulse giving values higher than the mean, having a 25% chance of exceedance. Ieff = 54 f 1.43 / fn1.30, (1) where: Ieff = effective impulse in N s, f = footfall rate in Hertz fn = floor natural frequency (Hertz)

Calculation of vibration using the effective impulse


The response of a higher-frequency floor to footfall excitation is predicted using the concept of the effective impulse in the following steps: 1) All the natural modes of vibration of the floor with frequencies up to twice the lowest natural frequency are calculated. 2) The effective footfall impulse ( Ieff) for each mode is calculated from equation (1) for the selected footfall rate. 3) The peak velocity vn in each mode n at point i on the floor due to walking at point j is calculated using Ieff,n , (2) v n = in jn Mn where in and jn are the modeshape deflections at points i and j, and Mn is the mass of the mode. 4) The total vibration response to each footfall is found by summing the velocity responses in each mode (which decay with time) in the time domain.

Prediction of modal properties of a floor


Standard finite element analysis software packages provide a well-established and accessible way in which to predict the natural frequencies, mode shapes and modal masses of any structure, and this method is recommended for use with these vibration-prediction procedures. The ease with which suitable numerical structural models can be built with modern desktop software means that this can now be done as a routine design procedure, and there is no need to resort to simple but much less reliable static hand calculation methods. Structures of any degree of complexity can be modeled, and known fit-out features such as partitions can be included in the assessment. Design alternatives can be evaluated and optimized with confidence. For very simple structural forms, particularly single-bay structures, the necessary modal properties can be obtained from standard analytical formulae for uniform plates and beams.

Implementation and visualization


Our methodologies for both resonant

Figure 3. Effective impulse derived from Kerr footfall traces. 66 SEMICON D U C T O R F A B T E C H 28 TH E D I T I O N

Figure 4. Simulated vibration trace.


WWW.FABTECH.ORG

CLEANROOM

Figure 5. Arup vibration prediction program.

Figure 6. Contour plot of predicted vibration level for a building floor.

and transient calculation have been implemented in spreadsheet format (Figure 5).Various methods of increasing complexity are available for calculating a floors modal properties from analytical solutions for simple beams and plates to finite element analysis of an entire floor. Finite element software is seamlessly connected to the spreadsheet, enabling the predicted modal properties of structures of any size or degree of complexity to be read into the response calculation routines automatically. After the vibration responses have been calculated, the results can be read back into the finite element software from which contour plots of vibration level across a floor may be produced. The software allows different parts of the structure to be subjected to different levels of vibration input for example the excitation due to people walking quickly in a corridor would be higher than that due to people circulating slowly within a laboratory. The program calculates the worst vibration at any point arising from walking at any other point on the structure. Figure 6 shows a typical contour plot of the predicted vibration level across a laboratory floor plate. This

informs the optimum positioning for vibration-sensitive equipment. If desired the structural design can be optimized to provide high-level performance only where needed, thereby realizing considerable cost savings while providing a more sustainable building. This is useful to the structural/vibration design engineer and is an effective means for discussion with clients seeking to optimize the layout of sensitive equipment and the structural design.

Predictions

Validation and comparison with measurements


Measurements
Measurements of vibration caused by people walking on floors of various recently constructed buildings have been made for compar ison with predictions. Measurements in four buildings are compared with predictions by various methods below. Buildings A C are laboratories with high-frequency floors and Building D is a hospital with a low-frequency floor; key structural details are given in Table 1. For the Arup prediction method, the modal properties were derived from simple finite element models of each floor.

The vibration levels in micro-inches/s for the four floors are compared to the measurements in Figure 7 below. The results for floors A to C show that predictions from AISC [2] and BBN [3] are inconsistent, with the BBN generally underpredicting vibration levels. It should be noted that the AISC method calculates peak velocity, and we have adjusted these to rms to provide a direct comparison. The SCI Guide [4] predictions are extremely conservative, and we attribute this to a crude and very conservative assessment of effective mass, together with the absence of any benefit from a higher natural frequency. Building D is slightly different in that the vibration was measured at the location of sensitive equipment (not the center of the panel). This illustrates the inflexibility of the SCI and AISC methods, which only calculate peak panel response. It was also a low-frequency floor, so the low-frequency method of AISC was used for calculation. Whilst some of the current methods descr ibed above are effective in estimating the dynamic response of a particular type of structure, none is generally applicable, and this makes it
Slab 3.5 nwt concrete on 2 metal decking 3.25 lwt concrete on 3 metal decking 5 nwt concrete 3.25 lightweight slab 2 on metal deck

TABLE 1: FLOOR GEOMETRY AND CONSTRUCTION

Column Spacing A B C D Steel/RC (fully fitted out) Steel/RC (core & shell) RC (fully fitted out) Steel/RC (ductwork & partitions) 21 4 32 10 @85.3c/c 21 0 28 1 21 0 27 0 28.5 28.5

Beam Size W27 84 W21 44/W21 73 @ 84 c/c 12W 14D @42c/c W16 31 @9.5 oc

Girder Size W30 116 W24 76/W24 84 30W 24D 21W 73D

S E M I C O N D U C T O R F A B T E C H 28 TH EDITION

67

CLEANROOM

REFERENCES

Figure 7. Comparison of predicted vibration levels.

difficult to compare the performance of designs of different structural forms. These simple procedures all tend to disguise the theory behind the equations, making it difficult for the engineer to fully understand. The difference between peak and RMS vibration predictions, and the use of either total or 1/3 octave RMS criteria adds to the confusion. It is also difficult for the engineer to decide how to optimize designs for vibration performance particularly for higherfrequency floors because the different methods give completely conflicting indications on the effect of changing mass or stiffness.

induced vibration for such floors provide a basic level of design they are not generally reliable and may be very conservative or unconservative depending on the circumstances. The new method Arup has developed predicts vibration levels consistently within +/20% of the measured values, and is reliable over a wide range of floor types.This consistently good performance is due to: The use of fundamental dynamic theory and finite element analysis to predict dynamic properties of floors. Footfall force loading effects based on extensive experimental research, including statistical quantification of variability. The use of this more advanced method enables designs to be optimized with more confidence, reduces risks associated with the achievement of increasingly stringent vibration requirements, and can lead to more economical deign solutions.
ABOUT THE COMPANY
Arup is one of the worlds largest Consulting Engineering companies, with some 7,000 employees in over 70 offices across 40 countries. Arup provides services in design, planning and management and in many specialist disciplines including vibration and acoustics, fire, security and risk, business and economic consulting, environmental planning and sustainability.

1. S.C. Kerr, Human Induced Loading on Staircases, PhD Thesis, University of London, 1998. 2. T.M. Murray, D.E. Allen, and E.E. Ungar, AISC Floor Vibration due to Human Activity. Design Guide. Series No. 11, 1997. 3. BBN Laboratories, The Control of Vibration in Buildings. 4. T.A.Wyatt, Design Guide on the Vibration of Floors. Steel Construction Institute Publication 076, London, 1989. 5. J.H. Rainer, G. Pernica and D.E. Allen, dynamic loading and response of footbridges; Canadian Journal of Civil Engineering, 15(1), 66, 1988. 6. S.V. Ohlsson, Floor vibration and Human discomfor t, Chalmer s University of Technology, Goteborg, Sweden, 1982. 7. F.W. Galbraith and M.V. Barton, Ground loading from footsteps, Journal of Acoustical Soc. of America, Vol. 48 no.5, 1970. 8. F.C. Har per, The mechanics of walking, Research Appl. In Industry, Vol. 15, no.1, 1962.

Conclusions
Semiconductor facilities require low vibration environments, and higherfrequency floors are necessary in order to control footfall-induced vibration amongst other factors. Whilst current simple methods to predict footfallABOUT THE AUTHORS
Michael Willford has 30 years experience in engineering design of major building, civil and offshore structures. He is a specialist in structural dynamics and advanced numerical analysis. He is a world leader in the application of performancebased design in seismic and vibration engineering. He is an Arup Fellow (in recognition of his technical innovations) and currently leader of Arups Advanced Technology and Research Group. Caroline Field is an Associate Structural Engineer/ Vibration Consultant based in Arups San Francisco Office. She has 10 years structural engineering experience and specializes in earthquake and vibration consulting. She has a Masters Degree in Earthquake Engineering and Structural Dynamics from the University of London, Imperial College.

ENQUIRIES
Michael Willford Arup 13 Fitzroy Street, London W1T 4BQ UK Tel: +44 (0)20 7755 2224 E-mail: michael.willford@arup.com Caroline Field Arup 901 Market Street, San Francisco CA USA Tel: +1 (415) 957-9445 E-mail: caroline.field@arup.com

68

SEMICON D U C T O R F A B T E C H 28 TH E D I T I O N

WWW.FABTECH.ORG

You might also like