ENOC 2011, 24-29 July 2011, Rome, Italy

Bobsleigh Optimization - A Customized Dynamic Vibration Absorber with Limit Stops

Pascal Arnold , Christoph Glocker Institute for Mechanical Systems, Department of Mechanical and Process Engineering, ETH Zurich, Zurich, Switzerland

Summary. This paper presents a dynamic vibration absorber for a bobsleigh, that is based on vibration measuring runs on a real track. Observed chassis resonances are reduced by means of an absorber with the assumption that smaller vibration amplitudes should enhance the performance and controllability of the bobsleigh. The device features unconventional elements such as dry friction damping or limit stops for the tuned mass.

Bobsleighs and Vibrations

The performance optimization of bobsleighs is a challenge that engages a variety of research elds. Whereas aerodynamic optimization and athletic training are advancing in almost every competing team, the merit of vibration analysis of bobsleighs is a topic of much debate. To gain more information about the nature of the vibrations occurring in a bobsleigh, track experiments with a bobsleigh depicted in gure 1 have been conducted using a variety of sensors. In this paper the sensors around the steering headset are of interest.
3

2 1 Figure 1: Raw bobsleigh chassis with the canopy removed; accelerometer at sliders (1), strain gauge to measure the slider load (2) and accelerometer at the steering headset (3).

Figure 2 shows the logged slider acceleration in the upwards direction of the bobsleigh. After smoothing the signal that features rather large variation (b), one can see the centrifugal accelerations (a) of the bobsleigh in corners. When the bob is driving through corners, the upwards acceleration direction is inclined to the gravity direction. At times t = [62 . . . 64.5]s, the bob is on a straight part of the track with a little bump around 63.3s. In the time window between 65s and 67.5s the bob passes a corner.
Slider Acceleration[m/s2]
Measured Smoothed a b

20 10 0 -10 58 59 60 61 62

63 Time t [s]

64

65

66

67

68

Figure 2: Measured slider acceleration on the Cesana bobsleigh track. Centrifugal acceleration (a) is of the same magnitude than the variation of the signal (b).

Standard FFT analyses of several similar signals lead to unsatisfactory results, as it is impossible to nd a balance between analyzed signal length, windowing and FFT length, such that the FFT variance is acceptable. As the acceleration data has been sampled at 2kHz and the data set size is approximately 180000 points or 60s, it is expected that a reasonable frequency resolution and FFT precision could be achieved between 5 and 100Hz. This means that the signal sampling is appropriate.

ENOC 2011, 24-29 July 2011, Rome, Italy

Another measure for the frequency domain properties is the Power Spectral Density (PSD) derived after Welchs Method [4]. In this analysis, the signal is chopped into several (50% overlapping) segments which are individually subjected to a FFT analysis. Each segment is used to compute its modied periodogram after [4], and the nal PSD measure is an averaged value of these periodograms with units Power per Hz. This power spectrum estimate features a reduced variance, as well as reduced signal amplitude loss due to the windowing process compared to a single FFT analysis. Therefore a signal with high variance is much more convenient to interpret. The drawback of this method is a lower frequency resolution because of shorter individual FFT segments. To compute the PSD for the bobsleigh measurements, the signal is chopped into 88 segments (4096 points each or approx. 60m track length) with 50% overlap, weighed by a Hamming window. As a consequence, one can clearly distinguish the resonance peaks from noise, despite the rather low frequency resolution of 0.48Hz. Figure 3 depicts the PSD of three signals logged by accelerometers and strain gauges in the bobsleigh during a complete track run. The power spectrum of the upwards slider acceleration makes clear, that the hard contact between slider and rough ice surface induces pretty much a wide-band excitation into the bobsleigh chassis with the expected natural decreasing signal energy for higher frequencies.
30 20 Slider Acceleration Steering Headset Acceleration Slider Load

PSD [dB/Hz]

10 0 -10 -20 Frequency Resolution 0.48Hz -30 10 20 30 40 50 60 Frequency f [Hz] 70 80 90

Figure 3: Power Spectral Density (PSD) of three measured signals during a complete bobsleigh run.

It is also found that a characteristic resonant phenomenon occurs in the complete front substructure of the chassis. Starting at the upwards slider load sensor (see gure 1, (2)), the vibration is transferred to the front axle and the upwards headset acceleration in gure 1, (3). From gure 2 it becomes clear that the vibrations are of the same order of magnitude than the centrifugal acceleration, thus ab. Based on the assumption that large vibrations in the bobsleigh chassis might have a negative impact on controllability or performance, a Dynamic Vibration Absorber (DVA) system is developed to reduce the vibration energy of the upwards steering headset acceleration.

Identication of the Dynamic Vibration Absorber Target Frequency

In order to design a DVA featuring optimal efciency, it is crucial to assign a relevant resonance peak that is present on any bobsleigh track. Figure 3 shows two predominant peaks at 16 and 25Hz that could be observed in several tests. Laboratory measurements using hammer strokes at different spots for excitation reveal that the well-reproducable 25Hzpeak is related to the rst bending mode of the chassis. It covers less than 3% of the signal energy contained between 5 and 100Hz (application of Plancherels theorem [5]). On the other hand, the 16Hz-peak is clearly a consequence of the interaction between the bobsleigh and the track, as it is absent in all laboratory tests, but present on the eld tests. This peak contains substantial 13% of the signal energy. Several test runs on different bobsleigh tracks have been performed and the named peak frequency has always been found in the range between 14 and 16.5Hz. Possible reasons for variations are different ice conditions during the test sessions or different track foundation properties (natural track vs. articially cooled track). The DVA target frequency is set to 16Hz mainly due to the higher energy absorbing potential. Even if the resonance peak is not perfectly matched for all tracks, the DVA absorbs more energy than if it was tuned for the 25Hz peak.

ENOC 2011, 24-29 July 2011, Rome, Italy

DVA with Limit Stops and Dry Friction Damping

To determine appropriate parameters of the DVA a simplied model of the bobsleigh front substructure is required (see gure 4). As vibrations have xmax only been analyzed in the upwards direction, the model should also be oneFN m directional. The mass of the surrounding bobsleigh structure is condensed 2 in m1 = 31kg and parameters [k1 = 3.25 105 N/m, d1 = 1.6 103 Ns/m] xmax are chosen such that the 16Hz peak can be reached and the damping is corx2 k2 responding to the measured PSD in gure 3. In contrast to many engineering applications, viscous dampers are forbidden m1 by the bobsleigh federation rules [3]. As a consequence, DVA damping is realized by dry friction ( = 0.3, FN ) between the moving mass m2 and the front structure m1 . Also the spring k2 cannot be implemented as a simple x1 d1 k1 coil spring, but has to be designed as a leaf spring in order to be consistent with the rules. The measured acceleration data (e.g. gure 2) of the sliders is r(t) used to generate the base excitation r(t). An initial analysis and optimization of this simplied model leads to DVA parameters showing relatively high amplitudes (x2 x1 ) between the tuned Figure 4: Dynamic vibration absorber with limit stops (xmax ) & dry friction damping (). mass and the underlying body, which would result in interference of m2 with the surrounding structures of the bobsleigh. The DVA system is therefore extended by two limit stops (xmax = 0.01m in gure 4) to bound the amplitude of m2 . The equations of motion m1 x 1 = k1 (r x1 ) + d1 (r x 1 ) + k2 (x2 x1 ) + N1 N2 T m2 x 2 = k2 (x1 x2 ) N1 + N2 + T (1)

of this specic multibody system with impacts at the stops (Ni ) and dry friction (T ) are formulated within the nonsmooth dynamics aproach [1]. The set-valued force laws of normal cone type N1 Upr(x1 x2 + xmax ) N2 Upr(x2 x1 + xmax ) T FN Sgn(x 2 x 1) (2)

are formulated as inclusions using Upr and Sgn functions (gure 5). The unilateral contacts induced by the limit stops (eq. (2)) are considered to be hard constraints with a Newton-type of impact law with a corresponding impact coefcient N = 0.2. For the tangential contact, T = 0 is chosen. Moreaus timestepping algorithm, which is a time-discretization on velocity impulse level, is used for numerical simulation [2].
Upr(gN) Sgn(gN) 1 gN -1 gT

Figure 5: Upr and Sgn-functions used in the inclusions of equation (2).

For the further analysis the one-dimensional model with the DVA attached as seen in gure 4 is named the 2-DOF system and the reduced model only considering m1 , k1 , d1 and r(t) without DVA attached is dened as the 1-DOF system.

ENOC 2011, 24-29 July 2011, Rome, Italy

Numerical Optimization of the DVA

The main purpose of the DVA is to reduce vertical vibrations of the steering headset. Because there exist no optimal design rules for a system with dry friction and limit stops, a cost function J is chosen to be minimized, taking the acceleration root mean square value J= ( x1 )2 /N

of the underlying mass m1 . N is the number of simulation time steps. A close connection between the cost function J and the power spectrum is given, because the integral over all frequencies f of the power spectral density is equal to J 2 . An alternative cost function such as the amount of absorbed energy of the DVA results in similar values for the optimization parameters. Figure 6 shows that the 1-DOF model approximates the measured 16Hz peak adequate, and that the 2-DOF system shows absorbing properties in the frequency band between 15 and 22.5Hz (see detail in gure 6).
40

Detail of the 16Hz Peak

1 Measured 2 Simulated 1-DOF Model 3 Simulated 2-DOF Model

20 2 3 1

20
PSD [dB/Hz]

10
1 -20 3
-40

Absorbing Region 0
Frequency Resolution 0.48 Hz

10

20

30

40 50 60 Frequency f [Hz]

70

80

90

10

15 20 Frequency f [Hz]

25

Figure 6: PSD signals of measured upwards steering headset acceleration and simulation results for the 1-DOF and 2-DOF model.

To quantify the absorption effect of the DVA, one can estimate the reduction of vibration energy by a comparison of the signal energies of the 2-DOF system with the 1-DOF system. The corresponding total signal energies are equal to the integral of the PSD functions, calculated in gure 6, over all frequencies. However after studying gure 6 it becomes clear that the 1-DOF model is tting the measured data only in a small frequency band ftrust = [5 . . . 22]Hz accurately enough. In this region of trust, only about 23% of the total signal energy of the measurements are contained. Anyhow, the frequency interval of trust is always similar, no matter what measurements the 1DOF-model is based on. Because there only exist PSD values for discrete frequency intervals (0.48Hz), the signal energy is a sum and therefore the vibration energy reduction
22

P SD2DOF (f ) DV A = 1
f =5 22

[0 . . . 1]. P SD1DOF (f )

f =5

In order to verify the optimal parameter choice, the same DVA setup is tested for other data on 3 different tracks. Considering the frequency band ftrust = [5 . . . 22]Hz, the signal energy can be reduced by DV A = 15.6% on the Cesana track, DV A = 16.8% on the Igls track and DV A = 7.2% on the St. Moritz track when a DVA is tted to the system. It is found, that the performance of the DVA is relatively insensitive to parameter uncertainties, and that the version with limit stops seems to damp away also frequencies that are lying outside of the working frequency band of the DVA without the stops - mainly due to the high energy absorption of the impacts. Because on the natural ice track in St. Moritz a less dominant 16Hz-peak is observed, the corresponding signal energy reduction DV A is obviously smaller.

ENOC 2011, 24-29 July 2011, Rome, Italy

With optimal parameters, the movement of m2 can feature both sticking at the dry friction contact and bumping into the limit stops (xmax ), as can be seen in gure 7. It shows a simulation of the system with the optimized DVA attached as described in gure 4, excited by the measured slider acceleration r(t). Optimal performance was found with k2 = 5 105 N/m and FN = 71.5N.
0.01

x 2 - x 1 [m]

0.005 0 -0.005 Impact -0.01 224.6 224.8 225 225.2 225.4 225.6 225.8 226 226.2 226.4 Stick

Time t [s]

Figure 7: Simulated time signal of the relative motion (x2 x1 ) of the DVA on a bobsleigh track. Sticking at x2 x1 = const., impacts at x2 x1 = 0.01m.

Realization and Testing

The promising and robust simulation results of the optimized DVA led to its production, that, thanks to the limit stops for the motion of m2 , can be realized in a relatively small package. Up to now, only four test runs with a 4-man bobsleigh on the tracks of Winterberg (Germany) and St. Moritz (Switzerland) have been conducted. The DVA has a noticeable inuence at high speeds and stronger vibrations, according to the testpilots observations, but the number of test runs is by far too small to give well-founded statements about the performance of the device. Even if the simulation showed a large parameter tolerance of the model, it remains an open task to verify the vibration reduction effect by acceleration measurements on m1 and m2 or even direct displacement measurements (x2 x1 ) during a bobsleigh run. Also a complete test campaign to evaluate the DVA performance on the track with 2-man bobsleighs is planned.

Conclusions
The analysis of measured vibration signals during a bobsleigh run reveals resonance phenomena in the steering headset. Due to the high variance of the data the Power Spectral Density is applied for interpretation of the signal frequency contents. A dynamic vibration absorber with dry friction damping and limit stops has been designed and optimized for energy absorption in a bobsleigh chassis. Limit stops to bound the tuned mass movement have been implemented mainly due to mechanical constraints, but they also enhance the working bandwith and parameter tolerance of the device. Simulations using a one-dimensional model predict that the vertical steering headset vibration energy is reduced by 7.2-16.8% when the absorber is attached, depending on the track that is simulated. Not enough experiments have yet been conducted to verify the effect of the device on race performance.

References
[1] Glocker Ch. (2001) Set-Valued Force Laws: Dynamics of Non-Smooth Systems. Lecture Notes in Applied Mechanics 1, Springer. Berlin Heidelberg. 222 pages. [2] Leine R.I., Nijmeijer H. (2004) Dynamics and Bifurcations of Non-Smooth Mechanical Systems, Lecture Notes in Applied and Computational Mechanics Vol. 18. Springer. Berlin Heidelberg New York. [3] International Rules Bobsleigh, F ed eration Internationale de Bobsleigh et de Tobogganing (FIBT) http://www.bt.com/leadmin/Rules/Rules%202010-2011/Reg.BOB-2010-E.pdf, p. 29-49. [4] Welch P.D. (1967) The Use of Fast Fourier Transform for the Estimation of Power Spectra: A Method Based on Time Averaging Over Short, Modied Periodograms. IEEE Trans. Audio Electroacoustics AU-15: 70-73. [5] Wiener N. (1988) The Fourier Integral and Certain of its Applications. Cambridge University Press.