Charles W. Engelhardt Structural Dynamics Research Corporation 11995 El Camino Real, Suite 200 San Diego, California 92130 ABSTRACT An alternative approach to traditional random vibration analysis is discussed. Recent improvements in computer technology have made it practical to simulate random vibration using sufficiently long transient analyses. The transient approach uses the probabilistic definition of the random excitations to generate sufficiently long transient excitations which are statistically equivalent to the random excitations. The transient simulation approach has significant advantages in aiding the engineer to understand the behavior of the structure in the random environment and qualifying the structure for the random environment. Dynamic analysis of a vehicle traversing a rough road is used to illustrate the advantages of the transient method compared to the traditional random analysis method. NOMENCLATURE FFT 1FT PSD Nt M T lit SRSS RMS forward Fourier transform inverse Fourier transform power spectral density number of frequency points in a spectrum frequency increment = delta frequency total time (duration) of a transient function time increment = delta time square root of sum of squares root mean square INTRODUCTION Recent improvements in computer technology enable us to simulate transient responses of complex structures due to dynamic excitations with many thousands of timesteps, simulations which were impractical just a few years ago. This is a powerful alternative to the two traditional random simulation methods: random analysis in the frequency domain, or random vibration testing such as a shake table test or vehicle road test. The recently practical transient analysis approach has tremendous advantages over the traditional approaches. Cross-country road surface PSD at 15 mph 1.00E+00 1.00E-01 N' 1.00E-02 :r: ;:;- 1.00E-03 :. 0 (/) Q. 1.00E-04 'E 1.00E-05 Q) E Q) u 1.00E-06 <1l a. (J) 0 1.00E-07 1.00E-08 0.0 2.0 4.0 6.0 8.0 10.0 Frequency (Hz) Figure 1. Road Surface PSD is used as excitation for the random vibration analysis. Shake table testing simulates a random environment by converting the random excitations to transient excitations which are used to drive the shake table. The new computer technology makes it practical for us to do this analytically for large computer models. We can use the transient excitations which previously drove the shaker table to analytically drive the computer model. However, this transient analysis approach is not limited by shake table stroke limits and is not limited to base excitations. This "virtual testing" has many of the advantages of an actual random vibration test compared to a random analysis, without the hardware limitations. An example of a traditional random analysis is the simulation of a vehicle driving over a road surface, where the road surface is defined with a power spectral density (PSD) function (Figure 1 ). The random analysis generates PSD responses of requested items, such as cargo acceleration or a particular stress component at some structural location (Figure 2), but a combined stress, such as von Misses stress can be calculated but the statistical significance is not well understood. 1852 0.20 Cargo Vertical Acceleration PSD N' 0.10 RMS = 0.22 g I ;;;- 3cr =0.66 g
0 en a. c: 0
CJ) Q) 0 0 <( 0.01 1---- f- - 1.0E-3
2.0E-4 0.0
.. 5.0 10.0 15.0 20.0 <il CD .t::. 0 , c CJ) E CJ) [il a. .!!l 0 Frequency (Hz) Figure 2. Acceleration PSD is a typical result of a random Analysis. : c: 0
CJ) Q) 0 0 <( -2.0 \i I -4.0 0.0 10.0 20.0 25.0 Time (seconds) Figure 3. Road Surface profile is used as excitation for the transient analysis. 1.00 0.50 0.00 -0.50 Cargo Vertical Acceleration RMS =0.22 g Pea =O.ng -1.00 ____ _j_ __ _j 0.0 10.0 20.0 25.0 Time (seconds) Figure 4. RMS of the transient acceleration response is equivalent to RMS of the random response. Alternatively, the transient approach uses a transient excitation (Figure 3) with the frequency content equivalent to the specified road surface PSD. Transient structural responses are computed, from which standard deviations and PSD functions can be computed equivalent to the random analysis results. (Figure 4) Additionally, transient results may also be recovered which are not available from the random analysis, such as a deformed shape at a time that the peak stress occurred (Figure 5). The trans1ent approach has the advantage over the traditional random approach by providing additional insight into the structure's physical response to the random excitations. Deformed shape at 4.45 sec Figure 5. Deformed shape from the transient analysis is useful for understanding causes of peak stresses in the random environment. TRANSIENT VS. RANDOM ANALYSES The traditional random analysis calculates the responses in the frequency domain. Then peak values are calculated for an assumed probability level, such as 3-sigma. Responses due to simultaneous random excitations are typically combined using some method such as SRSS. Also, response spectra are typically approximated from the random responses using Mile's equation. The transient approach to random analysis uses excitations which have frequency content defined by the random PSD and correlation functions. The PSD functions define the amplitude at each excitation. The PSD functions are normally interpolated so that the number of spectral lines (Nt) in the PSD and the spacing between each spectral line result in the desired timestep and duration for the resulting transient excitations, according to the standard 1FT procedure. A reference excitation is chosen for which random phasing is defined at each spectral line. The use of random phasing results in a Gaussian transient excitation assuming the number of spectral lines is large and none dominant, according to the central limit theorem. Phasing of the additional excitations is created relative to the reference excitation using the given correlation functions. The transient excitations are then calculated from these amplitude and phasing functions using the inverse Fourier 1853 transform (1FT). Then the transient simulation is performed to generate the desired transient structural responses. The root-mean-squared (RMS) value may be calculated directly from the transient results, which will be equivalent to the RMS value computed from the traditional random analysis, assuming reasonable transient duration and timestep values were used. The new computer technology makes it practical to conservatively choose a long duration and short timestep for the transient analysis. If conservative values for the transient timestep and duration are used, then differences between RMS values from the transient responses and RMS values from the random analysis will be insignificant compared with uncertainties due to modeling, damping, and forcing function assumptions. Transient displacement responses may be used to animate the structural response to the random environment. Transient results may also be computed which are not available from the random analysis, such as combined stresses, which are useful for the engineer to evaluate his structure. ADVANTAGES OF THE TRANSIENT APPROACH Similar to a random vibration test, the transient approach has many advantages over traditional random analysis because the analysis is done in the time domain rather than the frequency domain. These advantages include: (1) The analyst has greater visibility into the physical behavior of the structure in the random environment. The analyst can literally animate the structural response due to the random excitations. This visibility also includes insight into combined responses, such as von Mises stress, which are directly available from the transient analysis. Combined responses such as von Mises stress are difficult to calculate from traditional random analyses and the statistical significance is not well understood. (2) Responses due to simultaneous random excitations are typically combined using some method such as SASS. These excitations may be applied simultaneously in the transient approach using an estimate of the correlation (or lack of correlation) of these excitations. (3) Response spectra can be calculated directly from the transient results without the approximations involved in using Mile's equation to convert PSD functions to a response spectra. (4) Nonlinear effects can be included in direct transient solutions, resulting in a more accurate calculation of responses to the random excitations. GUIDELINES FOR CONVERTING RANDOM EXCITATIONS TO TRANSIENT EXCITATIONS The transient excitations are simply calculated from the random PSD and correlation functions using the inverse Fourier transform (1FT). The number of spectral lines (Nt) in the PSD, and the frequency increment between the lines (M) should be set appropriately to achieve the desired transient timestep and duration. This results in excitation functions with frequency content equivalent to the random PSD functions. An outline of the steps follows: (1) The desired transient excitation duration (T) and timestep ~ t ) must be chosen. These are subjective, but generally the duration should be long enough to include 20 - 100 cycles of the lowest frequency of interest. The lowest frequency of interest may be defined by the lowest frequency of the excitation, or may be the lowest structural mode of interest. The timestep should be short enough so that at least 5 - 1 0 timesteps occur within one cycle of the highest frequency of interest. (2) The random excitation PSD should be interpolated to produce the required frequency increment (M) and total number of spectral lines (Nt). The required frequency increment is the inverse of the desired transient duration and the number of spectral lines is half of the number of timesteps in the transient excitation. (3) For each excitation, the appropriate PSD and correlation function(s) are used to create a complex (amplitude and phase) function. Calculate the amplitude of each excitation spectrum from the PSD by multiplying the PSD by the frequency increment (M) and taking the square root. (4) Pick one of the excitations to be the "reference excitation", and use a random number generator to generate a random phase angle at each spectral line of the reference excitation. (5) Define the phase part of every other excitation relative to the reference excitation using the appropriate correlation functions. For example, for the vehicle road test, road surface excitation under a rear tire is phase shifted behind the front tires based on the vehicle speed, axle spacing, and frequency of the spectral line. (6) Convert each of the complex excitation functions to a transient excitation using the I FT. (7) Window the excitation functions as desired to eliminate the initial non-zero values and slopes. If not eliminated, the initial non-zero values may generate initial high- frequency "spikes" in the transient response. A cosine taper window applied to the initial and final 1% of the excitation is recommended. See reference [1] for a description of a cosine window. 1854 (8) Perform the transient analysis using these excitations. The timestep and duration of the excitations are appropriate for the transient solution. To verify that the excitations were created without error, the RMS values for each excitation should be calculated from the transient analysis results and compared with the RMS values of the original random PSD functions. CONCLUSIONS Current computer technology enables transient simulations of complex models which were computationally prohibitive in the recent past. The transient simulation approach may be used to simulate structural responses to random excitations, which can be considered a "virtual" random vibration test. The transient approach to random analysis has significant advantages over the traditional random analysis method or actual random vibration testing. These advantages include insight into the physical response of the structure in the time domain, calculation of time consistent deformed geometry and stress contour plots, and calculation of combined stresses such as von Mises stresses. REFERENCES [1] Bendat, J.S.; Piersol, A.G., Random DATA: Analysis and Measurement Procedures, John Wiley and Sons, Inc., New York, 1971, 407 pp. 1855