Académique Documents
Professionnel Documents
Culture Documents
546
The energy terms of F.q. 22 used in the context of This is done in a t&miqnc known as the Rayleigh-Ritz method,
Hamiltons equation, will gcncratc the equations of motion and which can k applied to any system that can bc characterized by
boundary canditim given in F+ 3 and 4. a self-adjoint (pz.&ivc dcfnite) differential writor L and
bamdary condition operators B, as given in Eqs. 5 and 6.[41 L
is of mder 2~. and Ll is of orda JI. For the fourth order F&r-
Bernoulli equation, p txluals 2.
547
A derivation based on energy terms, however, reveals km functions and the fmite clement method. In all caxs it
that the solution technique of Eq. II will work even if the slightly outperformed the cmpCsn function method and
functions used in the summation are only p-times differentiable signiticantly outperformed the tinite element methai.
and match the tamday conditions in sihxticns in which
displacement 1 slopes are tixed.[61 These functions are known The barn m&led in this study was m&led with
as admiisible functions. This lads to a difficulty of a different qua.+comparison functions. In an interview, Meirotitch
nature. There are now so many allowable bmctions that deter- suggested using combinations of tix mcdes generated by free
mining which fimctiom will work test is difficult. free and fmed-fixed tamdary condition% A mtimnm of four
tams are needed to match any pzsible boundary condition.
implementation SDecifics Wkile these mde shape can lx solved analytically, they contain
hypzrbiic trigonometric terms that result in difficult integmtions
over the tam. Therefore, linear combinations of the m&s
Quashm~arkwn Functions
resulting from pinncz-pinned and slide-slide boundary conditions
were used. With thex mcde shapes, boundary conditions could
While the reqnirements for a function to be admissible
still be mcdelled with four tams, but all of the mcde shapesare
are weak, there are strong rquirements that a set of admissible
camped of simple sine and cosine terms. This greatly simpli-
functions must n~eetif they are to accurately approximate the
ties the integrations required while retaining the pJwer of the
response of the structure. The szt of admissible functions must
qua+comparizan function method. Using thess m&s is
tc linearly independent and complete. Two examples of sets of
equivalent to a Fourier series with half-periodic tenm included,
functions that satisfy thex requirements are the Fourier series
as SW in Eq. 12.
and a polynomial seriesof increasing order. Either one of these
series can exactly match a given function if enough tams are
bxluded in the s&es.
548
inaccurate data storage form combined with mamix illxondition- presence of large, negative eigenvalues in the eigensystem
ing caused imaginary frequencies for mass and stiffness matrices solution, which correspond to imaginary frequencies. Th&
hug- than 1 I by 1 I. IXe second version, which worked occumed when a bigh numlxr of terms were present in the
directly with MAT tiles, was useful with systems as large as 19 &es. TIE exact numlxr of tams required for the singuIariw to
by 19. TIE third version of the program W&S formulated with take place was determined by the accuracy with which mmtters
direct design parameters mtber than the Eaditional combined were stored in the compter. The early text file program went
team pmmetm, such as El and nz. Finally, the fourth version singular at I3 terms, the Matlabbaxd solver went singular at 21
solved the eigenvalue problem direcdy using a Jacobi based tams, and the Pascal-based solver went singular at 29 terms.
symmewic eigensystem solvex. The utilization of symmetiy and
inaased precision available witb the Pascal compiler enabled Due to the Rayleigh-Ritz formulation of the beam, the
the progmn to work with systems as large zz 27 by 27. eigensystem should be either Ix&ix defnite or positive s&mi-
defmite. As a m.sult, negative eigenvalucs should not can.
The primary Alice of difficulty in solving the beam Therefore, when negative eigenvalws did cccw, the results of
poblem was iIl.conditioning that cccurred in the mass and the entire solution were suspect. The illanditioning of the
stiffness matxices. According to the Rayleigh-Ritz tbeoty, the mauices discwsed in the Ixevious section is probably the cause
discretized mzss and stiffness mat&x are Ixsitive derinite, of this problem. In order to alleviate this effect, an eigensystem
though the stiffness ma& is positive semi-definite if both es& solver specifk to symmetric, Izusitive semi-definite systems was
of the beam are free. However, if the series of functions used developed and incorpxated into the final version of tJx prc-
for approximation is not Iinealy indepzndenL positive detinite gmm.[lOl[lll Table 2 shows bow the symmetric solver
ness can lx lost. A lwk at the eigenvalues of the rna~ matrix performed in comparison to the Matlab eigensolver. Tbe
(which remains constant for a given number of tams) shows imaginary frequencies resulting from the negative eigenvalues
how the cigenvalues of the matix approach zero as the munlxr generated by Matlab are higbIighted to emphasize how it faiIs to
of tams in the series is incwased. This is indicative of a matrix calculate correct results for tbis Insitive deftite system
appmaching singukwity. Table I shows this wend.
Table 2: Comparison of eigensystem solvers
Table 1: Ill-conditioning of mass matrix
Matlab PC?.%%?1 Matlab PSCZII
Z7 n=9 = 11
0 2.6e-13 1206 1407
-- -. 0.0000000036 .7e-11 I 2.2e.12 1407 1623
-- 0.0000001101 0.0000003242 17.5 17.5 1623 1855
0.0000032794 0.0000076425 0.0000163275 46.2 46.2 1655 2106
0.0001576016 0.0002632472 0.0003195041 94.5 94.5 2107 2360
igew 0.0031573992 0.0032649727 0.0037480720 156 156 2432 2791
eS 0.0156133142 0.0163376762 0.0156133142 233 233 2625 3223
0.0290391704 0.0279616558 0.0261046567 326 326 3719 4449
0.0310666269 0.0310595565 0.0309071244 434 434 4699 5376
0.0456667222 0.0312169659 0.0312170274 536i 556 6791 10902
.- 0.0460192954 0.0312263043 556 697 13129 13880
-. -. 0.0462079125 697 651 51569 65016
651 1021 K7554i 69862
The reason for this tendency towards singularity is not
1021 1206 -- --
immediately apparent. It probably cccws because the whole
wavelength tams ccmpris~ the equivalent of a truncated Fourier
series. The baIf.wavelengtb terms, added through the WE of Small negative eigenvalues sometimes still resuhed from
quasi-comparison hmctions, comprise additional terms that can the use of the Pascal-based eigemolver, but these were slight
also te approximated with the Fomia series. As the numbs of pmxbatim from zero caused by rounding and tnmcation errors
terms is increased, this approximation becomes closa and closer during the solution pro-xss. These small negative values were
until a match with the added terms is obtained. At this In% safely rounded to zero for purIwses of this study, as they were
hear indepadence is lost and the ma& lxcomes singukir. clearly errors cccurring at the very limits of ma&ix precision.
A saond difficulty that occmred duibtg the quasi- A fmal difficulty encountered in accurately modeling the
comparison function m&lii of the beam mcde sbapzs was the beam was determining the achnl beam parameters to use in the
549
modeling mdy. Since the m&l was ultimately intcodcd to Lx each additional term crates anotber row and column to the
wed in comparisom with an actual steel tx%m, parameters for a ~y~tcmmatrice. This pushes all of the frequcncics down except
prticu!ar beam wed in the lab were &m&d. Table 3 gives for tbe new one ad&d. The frequencies nevcx drop below the
the best &mates available for the pxameters. actual dutiw.s frequencies if tbc solution is cowergent.
550
UK& shapzs for stiffoess values of I.54 lbflin. The d&shed lbxs
near zero represent the mm when compared to the theoretical
shapes.
M) @J 1 2 3 4
0.00 0.00 17.50 48.23 94.55 156.30
3.63 6.67 19.32 48.91 94.90 156.51
Figure 3: Fixed-free Mode Shapes
6.73 16.87 31.60 55.34 98.16 158.44
7.61 29.07 60.12 94.50 131.84 161.21 Validation bv Other ComputatIonal Model
7.71 30.69 68.53 120.47 185.28 261.03
7.72 30.87 69.46 123.47 192.89 277.72 Wlile zero and infinite stiffness tanday conditiow ale
easily tested analykally, imermediate values are not. To
7.72 30.67 69.47 123.50 192.97 277.87 validate the m&l oo intermediate values, another nomaical
package known as BEAM VI was nsed. BEAM I is a pagn,,,
based on transfer matrices that wlve~ team problenx[lZJ III
this approach, a finite number of onifom lxam sections are
joined together with joints of known stiffne~.
Finally, the &?&free &mJary conditions were studied Natural frequency of modes [Hz]
in a mamm similar to the pinne&pimted bamdary conditioos.
The tnnslational stiffoess was held war iofnity while the 0) w 1 2 3
rotatory stfffllcss as increased from n30 to near intiity. me 17.50 48.23 94.55
0.00 0.00
convergence for tbjs case is tab&ted in Table 7. Figue 3
21.07 21.07 27.39 52.64 96.87
shows the male shqes for a rotatory stiffoess of 1~6 lbf-in, with
the da&d lines mar zero representing the enor when compared 66.64 66.64 68.91 82.26 115.68
to thw~etical vahs. 210.74 210.74 211.47 216.19 230.98
Table R Fixed-free Convergence Study Since the results from the qoasi-comparison fooction
based Rayleigh-Ritz model very closely mat&d the results from
Natt~4freqwncy of modes [Hz] koovm analytical solutions and results from BEAM VI, it can be
stiffness concluded that the mxlel is effective for mcdeling the bxn and
[lbffin] any external stiffnews added to the ends of the beam
el) a 1 2 3 4
0 ~ 0.00 0.00 17.50 48.23 94.55 156.30
Gravitational Tension Convemence
10 3.63 6.87 19.32 48.91 94.90 156.51
1e2 6.73 18.87 31.60 55.34 98.16 158.44 Ibe convwgeoce and validation studies discosxd above
7.61 29.07 60.12 94.50 131.84 181.21 fccosed on variations of the fre&free beam problem for which
le3
iodependent solution were available. However, io the physical
le4 7.71 30.69 68.53 120.47 185.26 261.03 equiment in which the actual stainless steel beam was to be
le6 7.72 30.87 69.46 123.47 192.89 277.72 tested, there was a teosioo effect caused by gravity loading.
theory 7.72 30.87 69.47 123.50 192.97 277.87 This tension is described io Eq. 14, where W is the weight of the
- beam and x is the distance measured from the top of the barn.
551
Table l& Recommendations for Future Work
552