Sunteți pe pagina 1din 5

7th GRACM International Congress on Computational Mechanics Athens, 30 June 2 July 2011

Nonlinear System Identification of Vibroimpact Dynamics


M. Kurt1, Y.S. Lee2, D.M.McFarland3, L.A. Bergman3, A.F.Vakakis1
2

Department of Mechanical Science and Engineering, University of Illinois, Urbana, Illinois,USA Department of Mechanical and Aerospace Engineering, New Mexico State University, Las Cruces,USA 3 Department of Aerospace Engineering, University of Illinois, Urbana, Illinois, USA

Keywords: Nonlinear System Identification, Vibroimpact Dynamics, Reduced Order Modeling Abstract. This paper investigates the system identification of cantilever beam excited by a pulse force and undergoes two-sided vibro-impacts due to symmetrically placed rigid constraints. The empirical mode decomposition method (EMD) is used for the system identification of the dynamics, identifying sets of intrinsic mode functions (IMFs) corresponding to different mode frequencies. In the presence of nonsmooth effects, a nonsmooth IMF can be extracted out of the original time signal to obtain a smooth remander. Based on this analysis reduced order models of the dynamics can be constructed.

1 INTRODUCTION When analyzing a data signal, obtained either from experimental measurements or numerical simulations, problems such as short total data span, nonstationarity and nonlinearity of the data are almost inevitable. Although being the most commonly used technique in data signal analysis for, Fourier spectral analysis may exhibit spurious harmonic components that give a misleading energy-frequency distribution even if the signal consists of purely harmonic components. This is because Fourier spectral analysis assumes a linear system and a stationary data set. The EMD method combined with the Hilbert spectral analysis, which was introduced in Huang et al. [1] identifies in the characteristic time scales in the data. This decomposition method is claimed to be adaptive and highly efficient, which makes it more suitable for nonlinear and nonstationary data sets. Recently [2,3] a physics-based foundation for EMD has been proposed, making it suitable for reduced order modeling of nonlinear systems. In this paper we analyze the response of a 1.3 m long cantilever beam, fixed at one end, constrained by two rigid barriers from clearances at 1.185 m away from the clamped boundary condition, and excited by a pulse force at 0.4 m from its fixed end. EMD is utilized for nonlinear system identification. The general overview of the system can be seen below in Figure 1. The red points indicate the accelerometers where the response of the system are recorded, the forcing location is indicated by an arrow at approximately 0.4 m away from the support and the obstacles are marked at 1.815m. The blue point indicates the Laser Displacement Sensor, which is for experimental measurements. In this work only simulated data are considered.

Figure 1. The general overview of the system

M. Kurt, Y.S. Lee, D.M.McFarland, L.A. Bergman, A.F.Vakakis

Below in Table 1, the properties of the simulated beam are tabulated. Table 1 Properties of the beam [4] Beam Properties A E I L 2. Analysis The system in Figure 1 is simulated using an assumed modes simulation code, and the results are recorded at specific locations. The code has the flexibility of adjusting the clearances of the rigid constraints. Keeping this in mind, analysis is conducted for different clearance values, varying from infinitely large to very small clearances where the nonlinear effects are highly present. Then, for each case EMD analysis is performed in the measured data series for system identification purposes and the corresponding IMFs are extracted out of the time series, corresponding to the fast frequency components [2,3] present in the dynamics. 2.1 Case of Large Clearances Linear Dynamics Value 7850 Kg/m3 3.57 x 10-4 m2 200 GPa 1.9 x 10-9 m4 1.311 m

When there is a high clearance between the rigid constraints and the beam is large, no vibro-impacts occur and the dynamical response of the system is linear. In Figure 2, the acceleration time series together with the corresponding numerical wavelet transforms and frequency response function (FRF) are plotted for the sensing location at 1.215m along the beam, which is the closest measurement point to the location of the rigid constraint. As seen from the FRF and the wavelet transform, the system has 10 fast frequencies corresponding to the ten modes in the frequency range considered.

Figure 2. Acceleration response at 1.215 m location.

M. Kurt, Y.S. Lee, D.M.McFarland, L.A. Bergman, A.F.Vakakis

The EMD method is capable of extracting IMFs out of the original time series, for each of the 10 fast frequencies. IMFs with the two highest frequency content , approximately at 926 and 750 Hz are plotted in Figures 3a,b along with their wavelet transforms.

Figure 3. EMD analysis of the time series of Fig. 2: (a) IMF at ~920 Hz, (b) IMF at ~750 Hz.

Although there is the presence of the closely spaced modes, the linear case is relatively easier to handle as there are no nonsmooth effects in the time series data. As seen in the above figures of this section, the EMD analysis for the 1.215 m position(9th accelerometer) came out to be pretty straightforward and clean. The wavelet transforms revealed an almost perfect monochromatic frequency component for each IMF, with the almost same modulation pattern in each IMF graph. As seen below in Figure 4 a,b IMFs lower modes exhibit perfect monofrequency behaviour without any contamination by the neighboring modes, as evident by the wavelet transforms.

Figure 4. EMD analysis of the time series of Fig. 2: (a) IMF at ~64 Hz, (b) IMF at ~23 Hz.

2.2

Case of 4mm Clearance Nonlinear Case

By decreasing the clearance to 4mm the beam undergoes vibro-impact oscillations and strongly nonlinear effects appear in the dynamics. This is clearly evident in the time series and the corresponding Fourier transform of Figure 5 which depicts the acceleration time series at the same 1.215m location. It is evident that no system identification can be performed based on the Fourier transform. On the contrary, applying EMD we are able to decompose the data into a non-smooth component (containing the highest frequency effects due to localized vibroimpacts) and a nearly smooth residual containing the response of the modes due to vibro-impacts. Reduced models can then be extracted from the smooth residual providing us with modal responses due to vibro-impact excitations.

M. Kurt, Y.S. Lee, D.M.McFarland, L.A. Bergman, A.F.Vakakis


4000 2000 0 -2000 -4000

1000 500 0 10
0

0.5

1.5

200

400

600

800

1000

1200

Figure 5 Vibro-impact acceleration at 1.215 m location: (a,b) Time series and FRF

Before proceeding with the EMD analysis to extract the 10 monofrequency components out of the signal, the nonsmoothness of the data should be removed by again applying the EMD method to extract, what is called a nonsmooth IMF. Although nonsmooth IMF does not have a certain definition nor a methodology to follow during its extraction, it was used by Lee et. al in the literature [3] which was caused by the application of the impulsive disturbance, acted like a catalyst to excite all the modes of the mentioned system in the paper. EMD analysis for the VI case, which corresponds to a 4 mm clearance value between the stiff obstacles and the cantilever beam, is harder to analyze, due to the nonsmooth effects caused by the impacts. This is especially evident for the closest measurement (simulated) point on the beam, 1.215 m positioned accelerometer. Below in Figure 6 a, you can first see the nonsmooth IMF extracted from the time signals, As you can see, it basically contains the frequency jumps caused by the impacts, and also some harmonic component of 926 Hz, which is due the limitations of the EMD analysis. Note that, in Figure 6 b, the remainder, that is the part of the time signal after the nonsmooth IMF is subtracted from the original time signal, is pretty smooth, even around the global impact, which is almost at 0.4 s.

3000

1000
2000

0 -1000 -2000 1500 1000 500 0 0


accel(m/s 2) 1000

-1000

-2000

-3000

0.2

0.4

0.6

0.8

1.2

1.4

-4000

0.5

1.5

Figure 6 (a) Nonsmooth IMF for 1.215 m position (b) Smooth remainder

M. Kurt, Y.S. Lee, D.M.McFarland, L.A. Bergman, A.F.Vakakis

4000 2000 0 2000 4000 1500 1000 500 0 0

1000 500 0 -500 -1000 1500 1000 500 0 0

0.2

0.4

0.6

0.8

1.2

1.4

0.2

0.4

0.6

0.8

1.2

1.4

Figure 7 EMD analysis of the time series of Fig. 5: (a) IMF at ~920 Hz, (b) IMF at ~750 Hz.

The EMD method is capable of extracting IMFs out of the original time series, for each of the 10 fast frequencies. IMFs with the three highest frequency content , approximately at 926 and 750 Hz are plotted in Figures 7 a,b along with their wavelet transforms. Note that, even in the case of strongly nonlinear dynamics, EMD technique is strong enough to capture the IMFs with the corresponding fast frequencies of the system. 3.Conclusions In this paper, empirical mode decomposition (EMD) as a nonlinear system identification technique has been utilized for a vibroimpact beam simulation. Linear and nonlinear dynamics cases have been examined for the simulated beam. It was observed that, especially in the case of a finite clearance value between the obstacle and the beam, EMD technique proved to be a practical and strong tool for system identification.

Acknowledgement
This work was supported in part by Grant EPSCoR Grant CMMI 09-27995 ARRA provided by the US National Science Foundation. Dr. Suhada Jayasuriya is the Grant monitor.

References
[1] N. Huang, Z. Shen, S. Long, M. Wu, H. Shih, Q. Zheng, N.-C. Yen, C. Tung, H. Liu, The Empirical Mode Decompostion and the Hilbert Spectrum for Nonlinear and Non-stationary Time Series Analysis, Proc. Royal Soc. London A, 454, 903995, 1998. [2] Y.S. Lee, S. Tsakirtzis, A.F. Vakakis, D.M. McFarland, L.A. Bergman, Physics-Based Foundation for Empirical Mode Decomposition: Correspondence Between Intrinsic Mode Functions and Slow Flows. AIAA J. 47(12), 2938-2963, 2009. [3] Y.S. Lee, A.F. Vakakis, D.M. McFarland, L.A. Bergman, Non-linear System Identification of the Dynamics of Aeroelastic Instability Suppression Based on Targeted Energy Transfers, J. Royal Aero. Soc. 114(1152), 6182, 2010. [4} M.Mane, Experiments in Vibro-impact Beam Dynamics and a system exhibiting a Landau-Zener quantum effect ,unpublished thesis (M.S.), University of Illinois at Urbana Champaign

S-ar putea să vă placă și