In: Proceedings of the 5th National Turbine Engine High Cycle Fatigue Conference, Chandler, Arizona, March 2000

Component Mode Modeling of Mistuned Bladed Disk Vibration

Mr. Ronnie Bladh Graduate Student Research Assistant Dr. Matthew P. Castanier * Assistant Research Scientist Dr. Christophe Pierre Professor Department of Mechanical Engineering and Applied Mechanics The University of Michigan Ann Arbor, MI 48109-2125 ABSTRACT Mistuning in bladed disks can result in blade forced response amplitudes and stresses that are much larger than those predicted for a perfectly tuned assembly. Thus, mistuning has a critical impact on high cycle fatigue in turbine engines, and it is of great importance to be able to predict the mistuned forced response in an accurate and efficient manner. This work considers the current state of the art in reduced order modeling of mistuned bladed disk vibration, as defined by relatively new research codes developed in the GUIde Consortium. In particular, a critical analysis is offered for the REDUCE code and its underlying modeling method. It is shown that while REDUCE has produced highly accurate results for many test cases, it has some limitations that must be addressed. Based on the presented results, it is argued that a component mode approach with secondary modal reduction offers an ideal framework for developing accurate, low order models of mistuned bladed disks. The advantages of this approach include straightforward, efficient modeling of blade mistuning, incorporation of experimental measurements, and minimized model sizes. 1. INTRODUCTION Since the 1960s, several researchers have documented the effects of mistuning on blade vibrations by analyses of representative lumped parameter models. Unfortunately, in order to accurately represent an actual bladed disk design with a lumped parameter model, one must perform a difficult parameter identification. This identification procedure becomes infeasible as the number of model degrees of freedom (DOF) increases. Hence, to gain practical usefulness, there is a pressing need to employ accurate finite element models (FEMs) of rotor designs in mistuning studies. This important issue is addressed by the computationally efficient component-mode-based method of Castanier et al. (1997), which generates reduced order models systematically from FEMs of arbitrary complexity. This method has been particularly well received, and the corresponding computer code, REDUCE, is actively used by industrial members of the GUIde Consortium. Hence, it may be said that REDUCE represents the current state of the art and the industry standard for the vibration analysis of mistuned bladed disks. More recently, Yang and Griffin (1999) presented a promising technique that is based on a mistuning projection approach. This approach had been developed specifically for shrouded assemblies in Bladh et al. (1999).

Unfortunately, both the REDUCE and the mistuning projection techniques have important limitations that must be addressed: the method of Castanier et al. (1997) exhibits relatively slow modal convergence, and the mistuning projection method as formulated by Yang and Griffin (1999) is limited in scope in terms of realistic modeling of blade mistuning. Prompted by these limitations, as well as the continuous search for optimal (accuracy versus size) reduced order models, a recent study (Bladh et al., 2000) investigated alternative techniques for the vibration analysis of mistuned bladed disks. This paper provides a brief review of the most important findings from this study. 2. ALTERNATIVE TECHNIQUES The classical Craig-Bampton (C-B) component mode synthesis (CMS) method (Craig and Bampton, 1968) may be re-formulated specifically for the analysis of mistuned bladed disks. Each blade may be taken as a component structure. The entire disk may be treated as a single component by using a cyclic symmetry description, which significantly reduces the final size of the C-B CMS model by elimination of the sector-to-sector interface DOF. Note that the modal properties of the individual blades are directly accessible in this formulation, which makes the input of mistuning data trivial. After generating the mass and stiffness matrices using the C-B method, a secondary modal analysis may be performed in three different ways: (a) on the partitions of the matrices that pertain to the constraint modes (); (b) on the partitions of the matrices that pertain to the disk normal modes (D) plus the constraint modes; or (c) on the entire C-B CMS model. These three options are depicted schematically in Fig. 1, with the secondary modes labeled above the arrows. Only the secondary modes of interest are retained, which further reduces the size of the model. The blade mistuning is then introduced in the intermediate CMS matrices (blade component DOF) and projected from the primary modal coordinates to the secondary modal coordinates. This type of approach is referred to as Secondary Modal Analysis Reduction Technique (SMART). Clearly, SMART in any of the forms (a)-(c) yields further model order reduction, and it may be applied to any complex structure. However, in this study, only the minimized SMART model (c) is presented. The (a) and (b) approaches will not be discussed any further. In addition to SMART, a straightforward non-CMS method is considered that consists of a classical modal analysis of the nominal (tuned, cyclic) system, with a subsequent projection of the blade mistuning data onto the retained nominal system modes of vibration. This technique is here denoted mistuning projection method. This mistuning projection approach is a generalization of the mistuning formulation for shrouded blade assemblies developed in Bladh et al. (1999). Also, Yang and Griffin (1999) presented an analogous technique for the case in which each blade is mistuned by a small deviation in its Youngs modulus. Both the SMART and the mistuning projection methods employ the assumption of a common deformation space for tuned and mistuned mode families. This results in minimized reduced order models, where the set of retained system modes typically consists of a complete set of modes (all harmonics) for the blade-dominated mode family under investigation. Hence, the resulting model size is usually on the order of the number of blades. 3. SPEED VERSUS ACCURACY COMPARISON Figure 2 compares the relative costs of the different methods through estimates of the required number of floating point operations (flops) during a typical analysis of mistuned forced response statistics. First, it is noted that the effort involved in the secondary modal analysis on the C-B model is negligible the setup costs for the standard (cyclic) C-B and the SMART C-B models are virtually indistinguishable. As the mistuning patterns are run, it is clear that even though the standard C-B model features direct mistuning input, it suffers tremendously from carrying all the disk-blade interface DOF. It is also noted that while the mistuning projection method is minimal in final size, this method suffers from carrying out the projection of mistuning data in the physical (FEM) domain, even in the simplistic case of mistuning via Youngs modulus offsets. In contrast, studying a sequence of mistuning patterns is relatively effortless with the SMART model. 2

Finally, REDUCE demonstrates good analysis speed. However, it will be shown that REDUCE lacks the high accuracy of the other methods. Figure 3 depicts (a) the finite element mesh for the used test model and (b) a representation of its tuned vibration characteristics. The modal convergence for each method (except for the mistuning projection method, which is irrelevant for the tuned case) is presented in Fig. 4. This figure shows the average percent error among the six distinct eigenfrequencies in the Modal Convergence Region indicated in Fig. 3b versus the number of retained modes in the reduced order model (ROM). Figure 4 clearly illustrates the remarkable difference in both accuracy and modal convergence rate between the C-B model and REDUCE. However, it must be emphasized that the superiority of the C-B model is achieved at a considerable expense in terms of minimum model size relative to REDUCE (approximately 200 DOF versus 100 DOF). The vertical line in Fig. 4 represents the SMART approach with a C-B model as intermediate CMS basis. The implication of the vertical line is that the tuned SMART model is always as accurate as the intermediate CMS model from which it derives, while its size stays constant at the number of modes selected by the analyst. Hence, it may be viewed as collapsing the C-B model onto an arbitrarily smaller, ndimensional subspace. Note that using SMART for tuned free vibration analyses makes no sense; it is included here only to aid in demonstrating the approach. 4. MISTUNED FREE AND FORCED RESPONSE REPRESENTATIONS Next, the model in Fig. 3a is mistuned by offsets in blade Young's modulus. The mistuning values are taken as samples from a uniformly distributed random variable of mean zero and standard deviation 3.0%. Figure 5 shows an example mistuned mode, which belongs to the second family of blade-dominated system modes located around 3300 Hz (2F, see Fig. 3b). Except for REDUCE, note the excellent agreement among eigenfrequencies listed in the key for Fig. 5 the relative error is less than 0.02%. These results translate to the mode shapes as well. The approximated mode shapes in Fig. 5 are virtually indistinguishable from the mode shape obtained by full finite element analysis. Although REDUCEs mode representation is not as accurate, the method does an excellent job in capturing the relative peaks and their locations. Finally, for a forced response case, an engine order one (1E) excitation is considered in Veering #1 (see Fig. 3b), using a unit nodal load applied to one of the nodes on the blade tip. The resulting mistuned forced response amplitudes from a frequency sweep through Veering #1 are shown in Fig. 6. Clearly, the excellent accuracy displayed by C-B, SMART, and the mistuning projection method in free vibrations translates nicely to the forced response. As demonstrated in Fig. 6, the mistuned response approximations are practically indistinguishable from the exact (FEM) solution for these methods. Also, the capture of the maximum amplitude and the general resonance behavior over the frequency range is outstanding for these three methods. Again, REDUCE lacks accuracy in comparison, although it does provide a good prediction of the mistuned peak amplitude, which is the key quantity in studies of forced response statistics. Based on the presented results, it is concluded that the C-B, SMART, and the mistuning projection methods are all consistent, high-performance reduced order modeling techniques suitable for mistuned bladed disks. Taking into consideration the differences in efficiency and the final ROM sizes, it is clear that SMART must be considered to be the premier method examined in this study. 5. CONCLUSIONS This paper discussed the application of two novel approaches for the reduced order modeling of mistuned bladed disks. These two methods the mistuning projection method and the secondary modal analysis reduction technique (SMART) utilize the assumption that tuned and mistuned mode families span the same deformation space. This results in very small reduced order models (ROMs), with matrix dimensions on the order of the number of blades. In the mistuning projection method, a classical modal analysis is performed on the finite element model of a bladed disk. A coordinate transformation is performed to project the blade mistuning 3

from the finite element domain onto the tuned system modes. In the SMART approach, a primary ROM is generated via component mode synthesis, and then a secondary modal analysis is performed to generate a smaller, secondary ROM. The mistuning data is implemented directly in the blade-component modal coordinates of the primary ROM, and the mistuning is then projected onto the secondary ROM. Using a low-order, modal-domain projection makes the SMART approach highly efficient. The methods were compared in terms of modal convergence, mistuned mode shape representation, and mistuned forced response amplitude predictions. It was demonstrated that the mistuning projection and SMART methods outperform the current state of the art technique, REDUCE, as well as a cyclic Craig-Bampton approach. The new methods exhibit comparable or improved accuracy, while being far superior in computational efficiency. The SMART approach was shown to be exceptionally fast for running simulations of mistuned rotor forced response. Thus, SMART appears to be the most appealing method to date for comprehensive studies of forced response statistics for mistuned bladed disks. REFERENCES Bladh, R., Castanier, M. P., and Pierre, C., 1999, Reduced Order Modeling and Vibration Analysis of Mistuned Bladed Disk Assemblies with Shrouds, ASME Journal of Engineering for Gas Turbines and Power, Vol. 121, No. 3, pp. 515-522. Bladh, R., Castanier, M. P., and Pierre, C., 2000, Component-Mode-Based Reduced Order Modeling Techniques for Mistuned Bladed Disks, Part I: Theoretical Models, Part II: Application, Proceedings of the 45th ASME Gas Turbine and Aeroengine Technical Congress, Exposition and Users Symposium, Munich, Germany (to appear). Castanier, M. P., ttarsson, G. S., and Pierre, C., 1997, A Reduced-Order Modeling Technique for Mistuned Bladed Disks, ASME Journal of Vibrations and Acoustics, Vol. 119, No. 3, pp. 439-447. Craig, R. R., and Bampton, M. C. C., 1968, Coupling of Substructures for Dynamic Analyses, AIAA Journal, Vol. 6, No. 7, pp. 1313-1319. Yang, M.-T., and Griffin, J. H., 1999, A Reduced Order Model of Mistuning Using a Subset of Nominal System Modes, Proceedings of the 44th ASME Gas Turbine and Aeroengine Technical Congress, Exposition and Users Symposium, Indianapolis, Indiana.

DD D DB D B BD B BB
Craig-Bampton Basis

DD D DB D B BD B BB

Characteristic Constraint Modes

DD DC DB CD CC CB BD BC BB

(a)

DD D DB D B BD B BB

Disk plus Interface Modes

DB DD BD BB

DD D DB D B BD B BB

Global Modes

[S]

(b) (c) Figure 1: Schematic representation of secondary modal analysis reduction technique (SMART), using a Craig-Bampton model as intermediate basis. The secondary modal analysis may be performed on (a) the partitions of the matrices that pertain to the constraint modes (); (b) the partitions of the matrices that pertain to the disk normal modes (D) plus the constraint modes; (c) the entire intermediate model.

10

16

Standard C

raigBampt

on

Number of Floating Point Operations

10

15

10

14

Mistuning Proj

ection (Individ

ual Mode Mist s Modulus O

uning)

Mistuning Proj

ection (Young

ffsets)

10

13

REDUCE

10

12

SMART CraigBampton o = Model Setup Cost

10

11

10

20

30

40

50

60

70

80

90

100

Number of Mistuning Patterns

Figure 2: Comparison of estimated cumulative numbers of floating point operations during statistical analyses including the model setup cost.

8 7

Eigenfrequency [kHz]

3F

6 5
Veering #1

Veering #2

4
2F

3 2
Modal Convergence Region

1
1F

0 0

Nodal Diameter

(a) (b) Figure 3: (a) Finite element mesh for the used test model. (b) Natural frequencies versus nodal diameters, where the character of each family of blade-dominated modes is indicated on the right (e.g., 1F=1st Flex mode).
1

10

REDUCE CraigBampton SMART 10

2

Average Veering Error

10

10

10

10

10

100

200

300

400

500

ROM DOF

Figure 4: Modal convergence trends in the region surrounding Veering #1 (Fig. 3b).

1 0.8 FEM (2016 DOF): CraigBampton (192 DOF): REDUCE (96 DOF): Mistuning Projection (17 DOF): SMART (17 DOF): 3417.2 Hz 3417.1 Hz 3425.4 Hz 3417.5 Hz 3417.4 Hz

Relative Max Blade DOF Deflection

0.6 0.4 0.2 0 0.2 0.4 0.6 0.8 1

10

12

Blade Number

Figure 5: Reduced order model representations of mistuned mode shape number 37.

Max Blade Deflection Norm [mm]

0.2

FEM 2016 DOF SMART 17 DOF Mistuning Projection 17 DOF CraigBampton 300 DOF REDUCE 96 DOF Tuned FEM

0.15

une
0.1

d Mist
3250 0.05 0 3200 3300
Tun e d FE M

3350

3400

3450

3500

Excitation Frequency [Hz]

Figure 6: Forced response frequency sweep through Veering #1 (Fig. 3b) for engine order one (1E) excitation of the mistuned test model (tuned FEM solution included for reference).