The Relationship Between Finite Element Analysis and Modal Analysis

Neville F. Rieger, Stress Technology Incorporated, Rochester, New York The properties and techniques of Modal Analysis and of Finite Element Analysis are identified, together with the present advantages and shortcomings of both methods. The interrelationship between these techniques is described, and the contributions of modal analysis to efficient finite element analysis are reviewed. It is noted that the term modal analysis is used to describe, a) a test procedure for obtaining structural data, b) an analytical procedure for efficient solution of structural dynamics problems, and c) the same solution procedure for rotordynamics analysis. Four case histories are included which describe the joint use of modal analysis and finite element analysis to diagnose and confirm the results obtained in selected practical problems of structural analysis. During the past ten years, the rapid development of specialized test equipment and efficient numerical methods for modal calculation of structures has revolutionized vibration analysis. The purpose of this article is to discuss the interrelationship between the various modal methods and finite element analysis. Certain limitations, which affect the results attained by each procedure, are identified, and the restrictions, which these limitations impose on modal analysis and its application are discussed. The extent to which modal analysis and finite element analysis can be coordinated into an effective diagnostic procedure for vibration analysis is demonstrated by several case histories. Some definitions to clarify the terminology of this subject are discussed in the following section. Types of Modal Analysis As used in the general literature of vibration analysis, modal analysis may refer to either: a) A formalized test procedure for identifying the dynamical properties of structures. b) A mathematical procedure for increasing the efficiency of structural dynamics calculations. c) A technique for rotor balancing. Modal testing is a formalized method for identification of natural frequencies and mode shapes of structures. It utilizes dedicated modal test equipment, and requires a formalized procedure for disturbing, e.g., rapping, the structure into motion, and then recording the distribution of the resulting motions throughout the structure. The end results of a modal test are the various natural frequencies, mode shapes, and impedance data of the structure. These data are identified from the digitized input signals using efficient curve-fitting routines. The results are subsequently displayed as impedance plots and mode shapes (possibly animated). Mathematical modal analysis is an analytical procedure used to uncouple the structural equations of motion by use of a known transformation, as outlined in the following section. Details are given in standard textbooks such as Hurty and Rubinstein.1 The resulting analysis is then readily achieved by solution of the uncoupled equations. The modal response of the structure is then found through a reverse transformation, followed by a summing of the respective modal responses, in accordance with their degree of participation in the structural motion. Modal balancing is a rotor balancing procedure in which the respective modes of a rotor system are first isolated and then corrected for residual unbalance in sequence. The balance corrections used for one mode are carefully arranged in accordance with modal principles so as not to re-introduce the other modes of the rotor system. The above modal procedures have the following items in common: a) Identification of structural modes and frequencies for further analysis. b) The theory of each procedure is based on an analytical technique known as modal analysis, which uncouples the equations of motion to make possible their efficient solution. c) The orthogonal properties of structural dynamics matrices are utilized either directly in the analysis, or implicitly in (he practical test procedure. Notation A1 = [C] = [C] = c = [D] = E = {F} = [I] = [K] = [K*] = k = [M] = [M*] = m = N = P = R1 = {S} = T1 = Y = b = e = h = i r s t u & u & & u a [] = = = = = = = = = = = = = = = =

First Axial Mode Damping Matrix Modal Damping Matrix Modal Damping Coefficient Dynamical Matrix, [D] = [K] -2 [M] + i [C] Youngs Modulus Force Vector Unity Matrix Stiffness Matrix Modal Stiffness Matrix Modal Stiffness Mass Matrix Modal Mass Matrix Modal Mass Normal Force Applied Force First Torsional Mode State Vector of Displacements First Tangential Mode Displacement Distance 2.71828 Distance
1

Applied Force Location in Modal Test Displacement Location in Modal Test Time Displacement Velocity Acceleration Real Part of (Damping Term) Phase Angle Modal Force Modal Displacement Angular Displacement Complex Eigenvalue (a + i ) Coefficient of Friction Circular Frequency - Rad/Sec Imaginary Part of Modal Displacement Matrix

Practical Modal Analysis Procedures Practical modal analysis, or modal testing, involves the following operations: a) The structural response amplitude is acquired in digital format throughout a prescribed frequency domain, at a given displacement point r for excitation applied at a point s. b) The modal mini-computer automatically develops and stores this digitized frequency response data in a designated memory for subsequent processing. c) Curve-fit routines are applied to the frequency response data to identify the natural frequencies within the given frequency range. The corresponding mode shapes are extracted from the digitized amplitude data at the natural frequencies. d) The mode shapes may be animated in terms of the simplified structural model, corresponding to those locations at which the response has been determined. e) The modal damping is estimated from the magnitude of the response at each natural frequency. This is often the most approximate structural parameter obtained by modal testing. f) Modal matrix data are identified for the structure. Output is developed for mass, stiffness, and damping matrices suitable for further computations, based on the structural modal properties. These data are printed out for subsequent use. g) Some software packages permit modifications to be made to the matrix data, to evaluate the influence of possible changes on the natural frequencies and mode shapes. These packages can be run on certain commercially available modal analyzers. Finite Element Analysis Finite element analysis is a computerized procedure for the analysis of structures and other continua. Rapid engineering analyses can be performed because the structure is represented (modeled) using the known properties of standard geometric shapes, i.e., finite elements. Efficient, large, general-purpose computer codes now exist with appropriate matrix assembler routines and equation solvers for calculation of the following structural properties: a) Static displacement and static stress. b) Natural frequencies and mode shapes. c) Forced harmonic response amplitude and dynamic stress d) Transient dynamic response and transient stress. e) Random forced response, random dynamic stress. Finite element analysis used in this manner provides the dynamic properties of structures, including mode shapes and corresponding natural frequencies. General purpose finite element codes such as NASTRAN, ANSYS, SAP, ADINA, etc., are programmed to develop and solve the matrix equation of motion for the structure, viz:

Typically the matrix equations of motion for the structure contain off-diagonal terms. The matrix equation may be de-coupled by introducing the transformation:

{u} = [ ] {?}
and writing the following expressions:

(4)

[]T [M] [] = [M*] ] [K] [] = [K ]

T *

(5) (6) (7)

[] [C] [] = [C ]
T *

which contain only diagonal terms. The damping matrix may be uncoupled on the condition that the damping terms are proportional to either the corresponding stiffness matrix terms and/or the corresponding mass matrix terms. The uncoupled matrix expressions are:

&} + [C* ]{? & } + [K* ]{? } = [ ]T {F cos (? t + ?)} [M* ]{& ?
Each equation in this expression then has the form:

(8)

&r + c r ? & r + k r ? r = dr cos (? t + ? r ) m r& ?

To which the solution is:

(9)

nr =

dr cos(? t + ? r ) (k r m r ? 2 ) 2 + (cr ? )2

(10)

The dynamic displacements, u at frequency may then be obtained from the transformation {u} = [] {}. Modal testing complements finite element analysis, both by obtaining the natural frequencies and mode shapes directly by measurement, and by providing the matrix data [M] [C] and [K] for response analysis. Modal test results can be used to confirm the mode shapes and natural frequencies predicted by finite element analysis and the test data for the natural modes may be used to obtain the modal mass matrix and the modal stiffness matrix for the structure. To improve the efficiency of finite element calculations, a socalled eigenvalue economizer routine (Guyan reduction) is often used. These routines reduce the size of the dynamical matrix:

[D] = [K] ? 2 [M] + i ? [C]

(11)

&} + [C]{u & } + [K]{u} ={F cos (? t + ?)} [M]{& u

(1)

where the terms used are defined in the Notation section above. The model details are entered by the analyst in a standardized format. The computer then assembles the matrix equation of the structure. The first part of the solution to a given problem is to solve the matrix equation:

by condensing it around selected master nodes having master degrees of freedom (DOF), for which the modal displacements are needed. These master DOFs, are critical locations whose displacements participate strongly in the modes of interest for a given case. For example, the tip of a cantilever beam would be a typical master DOF location for most modes of a beam system. Guidance in the selection of such nodes for analysis can be obtained from the results of modal testing, where the response of the structure to dynamic forcing at such locations has been determined by test. Advantages of Modal Analysis The mode shapes and natural frequencies of a structure are its basic dynamic properties. Modal testing is used to rapidly identify these modes and their natural frequencies, and to provide the structural matrices, which govern the modes and natural frequencies. Thus the basic structural dynamic data, when obtained accurately from a valid test also provides a true identification of the structural properties for the modes of interest. These derived matrices are based on the measured participation of the mass, stiffness and damping properties in the modes of interest, for the actual boundary conditions, which the structure is experiencing. These data can then be used directly in a finite element model for the structure or component, for subsequent problem solving, or re-designing the equipment for more optimum dynamic response. Modern modal analysis test equipment has been developed to provide the maximum convenience in testing and data reduction, and to provide the above-mentioned dynamic properties of the structure. All modal analyzers contain dedicated mini-computers for efficient

&} + [K]{u} = 0 [M]{& u

(2)

for the free vibrations of the structure. The solution to Equation (2) gives the natural frequencies (eigenvalues) and the undamped mode shapes (eigenvectors). These parameters are the basic dynamical properties of the structure, and they are needed for use in subsequent analysis for dynamic displacements and stresses. For harmonic motions, {} = -2{u}. Substituting gives the matrix eigenvalue expression:

[K] [M] {u} = [I] {u}

(3)

where [I] is the unity matrix. Standard procedures (Jacobi, QR, SubSpace iteration) exist for extracting the eigenvalues: see Bathe and Wilson.2 The corresponding matrix of eigenvectors, [] are obtained by back substitution.

high-speed data processing, performed in a prescribed manner in accordance with a specialized test routine. In the hands of an experienced modal analyst, this leads to economical extraction of the data mentioned above. The advantages of modal analysis are, first, that a modal test provides the most rapid and effective procedure available for the acquisition of data on the dynamic properties of a structure. Such testing can often be performed by a skilled technician for later interpretation by a dynamics engineer. Second, modal analysis is an effective analytical procedure for the solution of large sets of structural dynamics equations because it reduces coupled matrix equations (which must otherwise be solved by some iterative procedure) to a set of independent linear equations, each with the well-known closed-form solution given above. Modal solutions can therefore be obtained directly, without further numerical operations. These solutions are then re-combined to form the complete solution to the structural response problem in question. It should here be noted that solutions to harmonic, transient, and random forced vibration problems can all be obtained using this modal analytical procedure, by means of simple extensions to the theoretical procedure outlined above: see Reference [1] for details. Shortcomings of Modal Analysis The output from modal testing consists of natural frequencies, mode shapes, modal stiffness, modal damping, and modal mass matrices. The main assumption involved in the acquisition of this information is that the structural system is linear, i.e., structural displacements are directly proportional to applied loads. In practical structures this condition is not always met. Structural systems may be non-linear to some degree, due to those causes listed below. Nonlinearities complicate the extraction of modal data and, where their effect is strong, they may invalidate the results obtained by linear analysis. Non-linear effects may be present in a structural system due to several causes: a) The material properties may be non-linear, e.g., composite structures, viscoelastic materials, elastic-plastic materials, where displacement is non-linearly related to force. b) Where large amplitudes are involved, the geometry may result in displacements, which are non-linearly related to load, e.g., large deflections of plate and shell-type structures. c) The structural boundary conditions may introduce nonlinearities, e.g., structures where the number of support points changes, or where the structure is a rotor mounted in fluid-film bearings experiencing relatively large whirl amplitudes. Such non-linear effects complicate the analysis and tend to introduce errors into the data reduction and curve-fitting estimates of natural frequencies. Such results cannot always be adequately represented by a linear analysis, because the properties change according to the magnitude of the applied load. Errors can range from small errors where minor non-linearities are present to large errors where the non-linear effects are substantial, such as in multiple support structural contact problems (load-dependent indeterminacy). A further limitation to modal testing is that it does not directly address the forced response problem, nor problems of transient response nor of random response. For problems in which the response to such loadings is of interest, modal amplitude data can be obtained by testing to formulate an efficient structural model for finite element analysis. Once the structural model is available in matrix form, the forcing data can be loaded into a finite element program, and the response to dynamic loading (harmonic, transient, or random) can then be obtained by calculation. The accuracy of such analyses depends of course on the validity of the model, which is generated from the modal test data. It is good practice to make a preliminary natural frequency/mode shape calculation with such data, to verify that the test modal data is consistent with the structural modes and frequencies upon which it is based. Another limitation of modal testing is that it cannot, by itself, predict threshold conditions for structural stability problems, such as structural buckling, and rotor whirl stability in fluid-film bearings.

Again, the modal test structural matrix data from such problems can be developed for subsequent (linear) finite element analysis, such as the prediction of stability threshold conditions. However, the nonlinear limitation again applies to the post- threshold behavior of such structures. Following the development of an unstable condition, e.g., buckling or rotor whirl, the structure characteristically undergoes large displacements until a new equilibrium condition is found. Such behavior may be highly non-linear, and so beyond the capabilities of modal analysis, and of the structural matrices developed by modal testing. Advantages of Finite Element Analysis Finite element analysis in conjunction with the high-speed digital computer permits the efficient solution of large, complex structural dynamics problems. As the majority of structural dynamics problems are linear they can be solved in the frequency domain using a modal transformation as noted above, subject to certain simplifying assumptions concerning the nature of damping.* Many efficient and comprehensive finite element computer codes are now available to perform structural dynamics response calculations involving harmonic response, transient response, and random response of complex structures. Provision is made in many large codes for storing specific solutions on tape and using these solutions as input to a second related problem, involving the same structure. For example dynamics problems where high temperatures cause changes in the elastic properties of the structure may be addressed by solving for the temperature distribution prior to the natural frequency calculations. The temperature distribution is first obtained for known input conditions, and this solution is used to solve the structural dynamics problem with temperature-dependent elasticity. Similar comments apply to fluid/structural interactions, where the equivalent mass properties of the fluid must be incorporated within the structural mass matrix. The finite element method therefore offers a very efficient procedure for the calculation of complex linear structures under a variety of dynamic excitation conditions, and under environmental conditions, which may include temperature effects and entrained fluid effects. Where the structure is nonlinear, modal testing may still be used (with caution) to estimate initial values for mass, stiffness, and damping parameters, which can then be modified to suit more advanced structural models. Shortcomings of Finite Element Analysis Although most linear structural dynamics problems may now be solved accurately and economically, it is still costly to solve most non-linear problems. For such cases a solution strategy must usually be developed on a case-by-case basis. In such instances the structural geometry and elasticity may be needed in considerable detail in the input data, and the formulation time for such cases may be significant unless suitable pre-processors are available within the code. The finite element analysis of recurrent structures, i.e., where a specific segment of the structure geometry is repeated a number of times, are still costly to solve. No general-purpose codes (or preprocessors) are yet in use, which specifically address this problem. Problems of recurrent geometry are relatively common, e.g., bladed turbomachine structures, axisymmetric structures, building structures, and many types of rotating machinery. The geometry of such structures often closes on itself ('ring' structures). The total structural matrix is still symmetrical and tri-diagonal, but the dynamical matrix contains off-diagonal elements, which may substantially increase the local matrix bandwidth. This causes a corresponding increase in computation time. Efficient computation of such recurrent components has been undertaken by special finite difference procedures,3 but sub-routines to undertake such computations are not yet in widespread use.

So-called Rayleigh (or proportional) damping conditions must be used for the simple modal transformation to apply. Other Techniques are available when the damping matrix is non-symmetrical.

Case History 1 - Vibrations of a Three Blade Group Purpose. Vibrations of a group of three turbine blades were studied to determine whether resonance was likely between a natural frequency of the group and some per-rev harmonic of running speed in the machine. Procedure. A modal test was made on the three-blade group shown in Figure 1. The group was welded to a massive steel block attached to a concrete floor slab at the level of the first hook contact, to simulate the rim attachment flexibility. The blades were also welded together at their cover sections. The cover was attached to the blades by a single large rivet. The structure had no tiewire. The measured natural frequencies of the group were obtained without centrifugal stiffening, but with some root flexibility. A finite element calculation was then made of the blade group natural frequencies, without centrifugal stiffening, and then with centrifugal effect. The modal test natural frequencies confirmed the finite element natural frequencies, as shown in Table 1. This indicated that the finite element model was appropriately scaled. Natural frequency values were then calculated with stress stiffening added, corresponding to 3600 rpm, as shown in Table 1. The influence of the centrifugal rise on the natural frequencies is evident. Mode Shapes. The first three test mode shapes are shown in Figure 2. These were obtained with Structural Measurement Systems, Inc. (SMS) Modal Analysis System. The first mode is a sideways, or tangential, motion of the group, al T1 = 170.2 Hz. Viewed from the top (not shown) the motion occurs at about 45 to the tangential. The second mode occurred at AI = 263.5 Hz, mostly in the axial direction, i.e., normal to the tangential direction. The third mode RI occurred at 369.7 Hz, and is mostly a group torsional motion, in which the cover lends to rotate as a rigid body around the tenon of the middle blade.

Table 1. Natural frequencies of three-blade group. Closest Mode Modal Finite Per-Rev Frequency, Element Frequency, Hz Hz Frequency (no CF), Hz

Finite Element Frequency (with CF), Hz

180 (3x) + 6% 300 (5x) + 5% 420 (7x) + 2%

1 2 3

170.2 263.5 369.7

175.4 270.7 393.1

190.3 286.4 413.1

Figure 2. First three-mode shapes for three-blade groups.

Figure 1. Three blade group for frequency testing.

Comments. The results show that the frequency margins for T1 and A1 are 6% and 5% respectively from the closest excitation harmonics. This difference is adequate for effective detuning from either resonance condition. The apparent frequency margin for R1 is only 2%, which in practice would be less than desirable to ensure safe operation, although the present detuning from resonance is still very significant. Also the attachment boundary conditions at the welded root attachment and at the welded rivet, as used to calculate this mode are thought to be somewhat stiffer than with actual blades in practice. In

practice this would tend to lower the R1 mode further away form any third mode resonance. The correlation between modal results and finite element results for the first two modes is typical for such blade structures with complex airfoil shapes and imprecise boundary conditions. Potential improvements in correlation through improved knowledge of boundary conditions are often offset by the statistical variance of frequencies between actual blades, and by the specific boundary conditions of each attachment under rotating conditions. Having verified the blade group model, the next step in such studies would be to determine the group response to specified per-rev steam stimulus. Case History 2 Vibrations of Sewage Pump and Piping System Problem Description. Excessive rotor vibrations were reported for the sewage pump and piping system shown in Figure 3. These vibrations were the suspected cause of rapid gland seal wear and frequent seal replacement, which led to unacceptable restriction of the operating speed and pump discharge range. A program of vibration tests and modal tests, supported by rotor system calculations was made to determine the reasons for this problem.4

b)

c)

d)

Modal tests were made to determine the natural frequencies and mode shapes of the pump and piping system. These tests were made with and without fluid in the pump, but not with the pump rotating. Displacement sensor tests were made near the gland seal, between the rotor shaft and the pump casing, to determine the conditions under which the shaft vibrations were worst, for the test matrix. Strain gage telemetry tests were made to determine the location of the torsional natural frequencies of the motor drive and pump rotor. Strain gages were placed on the pump shaft between the bearings. Induced-power radio telemetry was used to bring out the shaft torsional signals.

Figure 4. Rotor and support model details.

Figure 3. Sewage pump and piping system.

Pumping and Piping System. The pump and piping system shown in Figure 3 raises sewage from a wet well sump through a vertical riser pipe into a treatment tank. The pump inlet and outlet pipes both have a 30 inch I.D., and the vertical pipe has a 42 inch I.D. A check valve is situated 65 ft. above the outlet pipe centerline. The horizontal pipe to the treatment tank is 165 ft. above the pump centerline. The pump is driven by an 850 Hp motor mounted on a 30 ft. vertical pipe column supported from the pump volute casing. The motor drives the pump through a vertical shaft. The pump rotor also has a casing with cantilever bearing supports and a gland packing seal above the rotor, which experienced the wear and sewage leakage. The pump impeller is overhung, with two outlet flow passages (twovane impeller). The volute casing surrounds the impeller. Test Program. A series of tests to determine the cause of the pump system vibrations was undertaken as follows: a) Accelerometer tests were made on the motor casing, on the gland seal casing and pump outlet flange, and at various locations on the pipes, including the vertical bend and the check valve flange. Data were recorded throughout a matrix of test conditions relating to pump head and operating speeds.

Figure 5. Amplitude vs. frequency response for drive system.

Calculations. The forced response properties of the pump and casing structure were investigated using a multi-level rotordynamics code. The rotor system model details for the pump and casing are shown in Figure 4. An arbitrary unbalance of 1 oz. in. was added to the impeller to provide unbalance excitation at once per-rev. The model also contained a small amount of viscous damping in each of the rolling element bearings in the drive system. This calculation was performed to determine the location of the natural frequencies of the pump structure, including those of the cantilevered motor, drive coupling and cantilevered pump rotor (multi-level system). The

frequency response of the system in its first mode is shown in Figure 5. The mode shape is shown in Figure 6. These natural frequencies and mode shapes are compared with the modal test natural frequencies and mode shapes in Table 2, and with the system resonances from he accelerometer tests. Torsional drive system calculations were also made to determine the torsional natural frequencies of the drive rotating components. Calculated torsional results are compared with the telemetry test results in Table 2. Modal Test Results. The first mode of the pump and piping system is shown in Figure 7. In this mode, the pump structure vibrates against the riser pipe, in the direction of the outlet pipe axis, at 24.5 Hz. The drive rotor system also vibrates against the motor support structure. The second mode at 28 Hz is similar. The pump structure and riser pipe again vibrate against each other, but this time in a direction transverse to the outlet pipe axis. The rotating components again vibrate against the motor support structure. Other pump and piping system modes were found at frequencies given in Table 2.

Table 2. Calculated and test natural frequencies. Mode Calculated Modal Frequency, Hz Frequency, Hz Reed 1 Reed 2 Shaft Bending 1 Shaft Bending 2 Torsional Structural 1 Structural 2 Structural 3 Structural 4 24 28 (32)* 35.5 44.5 24.5 28.1 32.8 35.2 44.8 53.1 82.4 88.1 96.6

Accelerometer Frequency, Hz 28.0 30.5 46.3 54.8 81.5 86.5 98.0

*Depends on assumed casing support stiffness Table 3. System resonance frequencies Mode Natural Frequency, Hz Drive Shaft (Lateral Mode 1) Drive Shaft (Lateral Mode 2) Reed Mode 2 32.0

Excited By

Speed, rpm

35.5

28.0

2x 3x 4x 2x 3x 4x 2x 3x 4x

960 640 Op. Range 480 1050 690 Op. Range 510 840 Op. Range 560 420

Figure 6. Mode shape of drive shaft and pump casing at 1 = 35.5 Hz (rotordynamic calculation).

Figure 8. Typical accelerometer response spectrum (bump test at motor).

Figure 7. Mode shape of piping and drive at 2 = 24.5 Hz (modal test).

Vibration Test Results. Resonant frequencies found during operation of the pump are listed in Table 3. These data were obtained and confirmed at a number of locations. The results indicate the frequency of the excitation (1, 2, etc.) related to pump speed of rotation. A typical accelerometer response spectrum is shown in Figure 8, which indicates the characteristically strong 2 excitation, which was consistently observed in the accelerometer and displacement sensor results. The maximum relative shaft vibratory displacements occurred around 700 rpm and especially above 800 rpm. Similar results were found for the impeller displacements. Comments on Test Results. Resonance frequencies are summarized in Table 3. At 736 rpm the 2 impeller frequency coincides with the motor support reed frequency of 24.54 Hz, and the 3 coincides with the shaft lateral natural frequency at 35 Hz when the impeller speed is 700 rpm. The occurrence of these two lightlydamped resonances within this speed range is thought to be the cause of the observed strong vibrations around 700-740 rpm. At 795 rpm the 4 harmonic coincides with the system structural mode 1 at 53 Hz. At 840 rpm the 2 harmonic coincides with the second-motor support reed mode at 28 Hz. This response was observed strongly in several readings. The shaft torsional mode at 45 Hz is also

approached at 885 rpm. Torsional resonance would occur at 900 rpm, which was beyond the test speed range. Each of the above resonances appear to have been excited by the impeller jet transient loadings on the casing and pipe system. These resonances appear to have been the cause of the observed structural vibrations of the pump and piping. It is noted that the system has little inherent damping, and the resulting transient vibrations will thus be sustained by the recurring impeller jet impacts as rotation occurs, especially under resonant conditions. The conclusions are as follows: The measured structural, lateral and torsional responses of the pump and piping system are all caused by forced vibration. The primary cause of this forced vibration is the rotating impeller jet discharge, which impacts on the pump structure at twice per impeller revolution. The time-variation of each jet discharge forcing resembles a saw tooth shape. Two saw tooth loadings correspond to a single rotation of the impeller. This saw tooth forcing is the cause of the observed structural vibration and the shaft lateral and torsional vibration. The saw tooth Fourier spectrum contains the 1, 2, 3, and 4 and higher components. The strongest source of excitation is the two-per-rev harmonic of pump rotational speed. This excitation causes large vibrations to develop when it becomes resonant with any of the natural frequencies of the pump piping structure. The 24.5 Hz reed frequencies can be resonated by the strong two-per-rev forcing at 720 rpm. The 28 Hz reed frequency can be resonated by the strong twoper-rev forcing at 840 rpm. The 32.8 Hz lateral mode was excited but not resonated by the 2 pump harmonic. It could be resonated by the 3 at 656 rpm. The 45 Hz torsional mode was excited but not resonated by the 2 impeller harmonic. It could be resonated by the 4 at 675 rpm, which is close to the operating speed range of the pump. These modal tests and calculations identified and explained the cause of the pump-system vibrations, and allowed replacement modifications to be specified which are expected to overcome this problem, through a generally stiffer system with a five-vane impeller. Case History 3 Modal Analysis of a Turbine Generator Foundation Description of Foundation. A 330 MW turbine generator is mounted on the massive reinforced concrete foundation shown in Figure 9. The total weight of the turbine foundation is 2180 tons. The foundation is supported on 50 ft. concrete piles which extend down to bedrock. The natural frequencies and mode shapes of the unit were obtained by modal test, and by finite element analysis to satisfy seismic requirements.5

Modal Test. A total of 68 nodal test sites were selected to provide modal rap test data (selectively) in the x, y, z directions shown. Analysis of the modal test data led to the identification of 20 modes between 3.88 Hz (fundamental, x-longitudinal) and 229 Hz (complex platform bending torsion). Finite Element Calculation. A finite element calculation was also made for the foundation based on the construction drawing, and assuming that the concrete columns were elastic for 3 ft. below the concrete floor, and were rigidly secured below this depth. The calculated natural frequencies are compared with the measured frequencies in Table 4.
Table 4. Frequencies of natural modes, modal analysis and finite element calculations. Measured Calculated Measured Calculated Frequency, Frequency, Frequency, Frequency, Mode Hz Hz Mode Hz Hz 1 2 3 4 5 6 3.88 4.12 5.05 9.52 15.86 19.11 3.41 4.27 5.36 9.89 17.44 20.61 7 8 9 10 11 12 27.30 35.61 40.19 55.22 73.71 90.23 29.62 40.33 45.87 56.21 79.68 99.64

Figure 9. Turbine generator concrete foundation.

Figure 10. A-Mode 1, 3.88 Hz, x-translational, in-phase, rigid platform. B-Mode 2, 4.12Hz, y-lateral, in-phase, rigid platform. C-Mode 3, 5.05 Hz, ytorsional, platform warping. D-Mode 4, 9.52 Hz, y-lateral, platform bending.

Foundation Modes. The first four calculated modes of the structure are shown in Figure 10. Mode 1 is a lateral translation of the structure in the x-direction at 3.88 Hz. Mode 2 is the y-translation of the structure at 4.12 Hz. Mode 3 is rigid torsion of the platform on its columns at 5.05 Hz, and mode 4 is platform free-free bending at 9.52 Hz. Higher modes show further platform warping and column vs. column bending. Significance of Modes. For seismic considerations the most significant vibration responses are expected to occur in modes 1 and 2, because these modes can most readily couple with the strongest earthquake spectral components in the 2 to 5 Hz domain; see the typical earthquake spectrum in Figure 11. Above this frequency range the earthquake excitation strength is less, and the modal forms are also less likely to couple with the ground motions, e.g., column vs. column modes are unlikely with ground vibration, as are platform warping modes, and turbine generator vertical modes. Discussion of Results. The correlation obtained between the frequencies shown in Table 4 confirms both analyses (modal and F.E.), plus the column fixity assumption. The finite element model was formed from solid elements with reinforcement rods situated along element interfaces. The turbine generator was modeled with discrete masses. The contribution of the turbine generator is expected to be minor below 30 Hz, at which frequency any excessive rotating unbalance may couple with mode 7. The results of this study identify and confirm the location of the natural frequencies of the foundation structure, and reveal the associated mode shapes. Further calculations concerning possible response to earthquake excitation can next be made, using the nowconfirmed finite element models.

Table 5. Modal parameters identified for disk. Frequency Kd1, Mode Hz Md, Kg 108N/m 1 2 3 4 5 6 7 8 9 10 272 754 1414 1689 2495 2628 3158 3376 3388 4194 52.78 11.29 5.48 7.25 12.43 5.25 733.80 38509.00 124.76 104.84 1.55 2.55 4.35 8.22 30.70 14.40 2906.60 174000.00 564.40 731.30

Id, Kgm2 0.2457 0.1215 0.0289 0.0173 0.0126 0.0298 0.4929 2.4352 0.0101 0.0383

Kdr, 106Nm/rad 0.7226 2.7443 2.2919 1.9563 3.1073 8.1733 19.5214 1101.8000 4.5733 26.6830

Figure 12. Brake and wheel arrangement.

Figure 11. El Centro, CA earthquake, May 18, 1940, NS ground acceleration, velocity, and displacement.5

Case History 4 Brake Squeal in Subway Trains Problem Description. An unpleasant squeal problem developed in the braking system of subway cars in the new Washington, DC transit system. Sharp squealing occurred as the cars were brought to rest. An investigation of the braking system arrangement was made to determine the cause of this squealing and to seek a remedy.6 Analysis Procedure. Detailed modal tests were made on a typical brake assembly to identify potential natural frequency components in the squeal spectrum. A series of operational tests were made to determine the brake assembly natural frequencies, and to provide spectrum data for confirming the squeal model, for which a special computer code was developed for parametric studies. Modal Tests. The brake and wheel arrangement shown in Figure 12 was tested to determine its modal properties. A total of 26 modal test locations were used. A typical response spectrum for the brake pad is shown in Figure 13. The mode at 268 Hz shown in this spectrum is plotted in Figure 15. These tests identified the principal modes of the brake structure, and the modal parameters associated with them. These parameters are listed in Table 5. They were used later in this study for the parametric model studies.

Figure 13. Modal response.

Finite Element Calculations. A finite element model of the brake disk was made to calculate the disk natural frequencies, and to confirm the disk modal parameters obtained from the modal tests. A 180 segment of this disk model is shown in Figure 14. The 360 disk was made from 462 plate elements, 126 solid elements and 142 beam elements. The one diameter finite element mode shape for this disk (180 segment) is shown in Figure 15. The correlation obtained between modal frequencies and finite element frequencies is given in Table 6.

&} = [D]{S} where {S}is a This leads to the system of equations {& S state vector of displacements, and [D] is a matrix of stiffness mass and damping coefficients. The solution for {S}is obtained by writing:

{S} ={S0 }e ?t

(18)

which leads to the standard eigenvalue problem [D]{S0 } = ? 2 {S0 } . When the brake system is stable the real part of the eigenvalues is negative. When the mode in question is unstable the real part is positive. In general:
? 1,2 = a i ?

(19)

Figure 14. Finite element model of disk.

where a represents system damping (positive, unstable; negative, stable) and is the circular frequency of the mode in question. A computer program EIGRDISK was written and used to find instability threshold conditions for specified conditions of the brake system.
Table 6. Correlation between modal and finite element natural frequencies.

Mode 1 2 3 4 5 6 7 8

Frequency Element, Hz 268.45 950.76 1679.78 1984 - 2260 2582 - 2599 -

Modal Rap Tests, Hz 225 750 1362 1875 - 1937 2462 - 2500 3050 -3225 3737 4400

Figure 15. Mode 1, 268 Hz, finite element calculation.

Squeal Model. The modal data in Table 6 as confirmed by the finite element calculations were used to prepare the squeal model shown in Figure 16. This model had six degrees of freedom, i.e., a lateral and a translatory DOF for the disk, caliper, and wheel. Writing M, K, and I for mass, stiffness and inertia; Y, for displacement and rotation; , N, h, b for friction coefficient, normal force, and distance; and l, r for linear and rotational direction leads to the following equations of motion for the system:

&& + K (Y Y ) + K Y N? = 0 MdY d pe d p de d d
Disk

Figure 16. Squeal model of brake system.

(12) Discussion of Theoretical Results. Youngs modulus of the pad friction material and the coefficient of friction of the pad were varied in the analysis as these values change from time to time depending on brake pressure, temperature and wear of the pad. The distance between pad center and cylinder centerline was also varied. This parameter was found to exert a strong effect on the squeal propensity. Three different groups of squeal frequencies were found in the analysis: the first around 3000 Hz for large values of Youngs modulus from 0.5 to 3.0 108 N/m2, the second was 400 Hz for a narrow range of Youngs modulus around 0.2 108 N/m2 and the third around 4000 Hz for lower values of Youngs modulus around 0.1 107 N/m2. 3000 Hz Squeal Frequency. This squeal frequency was observed with the 6th natural mode of the disk, 2628 Hz. Figure 17 shows the squeal propensity and squeal frequency of one brake system as a function of pad Youngs modulus, for four values of friction coefficient. The squeal frequency is independent of the

& + K ? + K (? ? ) + hK (Y Y ) = 0 I d& ? d dr d pr d p pe d p && + K (Y Y ) + P = 0 M pY p pe p d

Pad

(13) (14)

& + K (? ? ) + 2 dK (Y Y ) + Pa = 0 I p& ? p pr p d pe d p

(15) (16)

&& + K Y P = 0 M cY c cl c
Caliper

& + K ? Pa + bP = 0 Ic& ? c cr c b

(17)

coefficient of friction and increases with Youngs modulus in a linear manner in the instability zone. For a large value of friction coefficient the instability regime covers a wide range of Youngs modulus of the pad. This range decreases with reduction in the coefficient of friction. The intensity of squeal, characterized by the squeal propensity, is high with large values of coefficient of friction.

condition, in terms of generating loud unpleasant noise around 400 and 4000 Hz. Figure 17 also shows the effect of friction coefficient on the squeal instability.

Figure 19. Disk frequency spectrum during squeal initiation.

Figure 17. Squeal instability regimes for disk brake system.

Figure 20. Disk frequency spectrum after initiation of squeal.

Figure 18. Schematic of field test instrumentation.

400 Hz Squeal Frequency. Squeal frequency was observed with the 10th mode of the disk at 4194 Hz, for a pad Youngs modulus of 0.1 108 N/m2. An interesting feature observed for this instability was that the disk coupled frequencies are also very close together which could result in another squeal frequency, 4200 Hz. The natural frequencies obtained for the coupled system with E (pad) = 0.1 108 N/m2 were 4204.8, 4114.0, 436.2, 436.2 (squeal frequency) and 442.2 Hz. The squeal propensity for the 436.2 Hz frequency was only 21.24 as compared to three digit values for the 3000 Hz squeal frequency considered before. However, because this is near to a possible second squeal frequency (4210 Hz), and further that the caliper frequency of 442.2 Hz is also near the 436.2 Hz squeal frequency, this system instability could clearly be a dangerous

Field Test Instrumentation. Figure 18 shows the instrumentation used to collect the field data. PCB Series 303A quartz accelerometers were used to convert the shock and vibratory motion into high level, low impedance signals used in these tests. 303A02 PCB accelerometers were stud mounted on the disk and on stationary parts. 302A03 PCB accelerometers were also used with adhesive mounts. Coaxial shielded cables were used to eliminate cable noise. AB&K type 2203 precision sound level meter with one inch condenser microphone was used to record the noise signature. The system was mounted on a portable floor stand, with the microphone pointing downwards near the front open window of the rail car used for testing. The brake pressure and truck speed was obtained from the car brake system directly. Two tape recorders, Racal Store 4 and Store 7 were used to store the data on magnetic tapes. A special bracket was used to accommodate slip rings, thermocouple amplifiers, accelerometer amplifiers and connecting leads from the rotating section. This bracket was clamped onto the axle using three dust cover bolts. A Rockland Scientific 444A Mini-Ubiquitous FFT computing spectrum analyzer was used to analyze the vibration and noise signatures, along with a Tektronix 4622 interactive digital plotter to obtain a permanent graphic record.

The two tape recorders and noise measuring equipment were located inside the car along with the spectrum analyzer, oscilloscope, and a plotter. The brake pressure and speed signals were taken from the circuits in the driving car. Field Test Results. The disk modal analysis indicated that squeal would occur mostly in the range of 2800-3300 Hz. Occasionally squeal was observed around 4250 Hz. In every case, the squeal frequency was first observed in the vibration signals of disk, pad and caliper limb, and in the noise signal, as the brakes were applied and the train began to slow down. After the squeal frequency appeared in the real time spectra, the amplitudes of vibration around 500 Hz grew rapidly as shown in the vibration signals, and in the noise signals. Two typical sets of the frequency spectra are discussed below. As soon as squeal appeared, the disk vibration was almost a harmonic motion as shown in Figure 19. Immediately following the appearance of the squeal at frequency 2825 Hz, the disk vibration levels increased strongly. The corresponding frequency spectrum is shown in Figure 20. The main frequency components in this spectrum are in the 250-350 Hz band, which is at the fundamental frequency of the disk, as determined by calculation and modal test. Higher mode components of the disk vibration are also present. The pad response during the initiation of squeal is shown in Figure 21 and the response after squeal started in Figure 22 shows the frequency spectrum of noise when the squeal began to appear with a frequency component of 2950 Hz. The noise spectrum then spread rapidly to a lower frequency range, which contains lower disk modes and lower system modes. The maximum temperature at the outer radius of the disk front face is 460F, decreasing to 335F at the inner radius. During these tests, it was observed that squeal was present for one direction of train motion but not in the reverse direction. This was also validated by the mathematical model; no instability occurred when the kinematic constraint of pad-piston contact point was reversed. Another brake system was also calculated and found that no squeal occurred for a wide range of Youngs modulus values of the pad. Field tests with this system also produced no squeal. This further confirms the analysis and the squeal model.

The squeal instability in the system is dependent on the coefficient of friction of the pad, Youngs modulus of the pad, and the point of application of brake pressure. The coefficient of friction of the pad is a function of temperature, wear, relative speed, and humidity conditions. The Youngs modulus of the pad is a function of brake pressure and temperature. These conditions vary quite randomly during operation and may give rise to different squeal behavior under apparently similar conditions of operation. The coefficient of friction exerts a dominant influence on squeal and it should be kept as low as possible. The squeal frequency is not significantly affected by any changes in coefficient of friction. For a given set a parameters of the system, squeal instability occurs over certain ranges of Youngs modulus of pad. The squeal frequency increases with Youngs modulus of pad in a given zone. For a given system, squeal can be eliminated by reducing the distance between pad center and cylinder centerline. The observations of squeal from the field data are substantiated by the mathematical model. The guidance provided by this study established the cause and the mechanism of the squeal problem, and provided several alternative paths for its solution. Washington Metro is now squeal-free.

Figure 22. Pad frequency spectrum after initiation of squeal.

Figure 21. Pad frequency spectrum during signal initiation.

Comments of the Test Results. Based on the theoretical results and field data diagnosis, the mechanism of squeal in disk brakes has been confirmed, and defined in detail. The instability of the brake system was found to initiate in the higher disk modes (frequencies), typically in the 6th and 10th modes in the present analysis.

References 1. Hurty, W. C., and Rubinstein, M. F., Dynamics of Structures, Prentice-Hall, Inc., Englewood Cliffs, NJ, 1964. 2. Bathe, K. J., Wilson, E. L., Numerical Methods in Finite Element Analysis, Prentice-Hall, Inc., Englewood Cliffs, NJ, 1976. 3. Salama, A. M., Petyt, M., and Mota Soares, C. A., Dynamic Analysis of Bladed Disks by Wave Propagation and Matrix Difference Techniques, ASME Monograph, Structural Dynamic Aspects of Bladed Disk Assemblies, presented at the Winter Annual Meeting of the ASME, December 5-10, 1976. 4. Rieger, N. F., and Kimber, A. W., Dynamic Interaction Between a Vertical Pump and Its Piping System, Proceedings of International Conference on the Hydraulics of Pumping Stations, BHRA, The Fluid Engineering Center, Manchester, England, pp. 253-266, September 17-19, 1985. 5. Newmark, N. M., Fundamentals of Earthquake Engineering, Prentice-Hall Inc., Englewood Cliffs, NJ, 1971. 6. Rao, J. S., and Rieger, N. F., Brake Squeal Problems in Underground Trains, Inst. Mech. Engrs. (London) Third Intern. Conf. Vibrations of Rotating Machinery, York, England, September 11-13, 1984.