INCAST 2008 - IT21

International Conference on Aerospace Science and Technology 26-28 June 2008, Bangalore, India

ROTOR DYNAMICS OF AIRCRAFT GAS TURBINE ENGINES

J. S. Rao
Chief Science Officer, Altair Engineering, Mercury 2B Block, 5th Floor, Prestige Tech Park, Bangalore, India js.rao@altair.com

Rankine defined critical speed of a rotor in 1869. Laval built the first impulse turbine in 1883 and ran his rotor upto 40,000 rpm by using an extremely flexible shaft. Jeffcott in 1919 formulated the rotor problem as one of forced vibration..Jeffcott models on 8 coefficient bearings are extensively used to study unbalance response, stability of practical rotors. Holzer in 1921 presented a tabular method to determine the torsional natural frequencies of systems, which can be discretized in the form of several rigid inertias, connected by massless torsional springs. Myklestad in 1944 extended Holzers approach for bending vibrations of aircraft wing type structures. Prohl adopted a similar tabular method in 1945 to determine the critical speeds of rotating shafts. This method has been used for a considerable period of time until it was replaced by transfer matrix forms for application by using computers. With the availability of computers by 1950s, both torsional and bending problems were initially expressed in Transfer Matrix form, and extensively used for torsional vibration under transient conditions of excitation such as electrical disturbances. The gyroscopic moment has a significant effect on the rotor dynamic behavior. Each natural frequency is split into two, one attributing to forward whirl and the second to backward whirl. The rotor whirls in a backward direction between these two critical speeds. For whirling rotors, bending in two planes is to be accounted. For solving unbalance response of rotors, 17x17 size matrices are employed. Fortunately, the overall transfer matrix still remains 17x17, and many industries still use codes based on transfer matrices, because of simplicity and less CPU time. Transfer matrices have been enhanced to account for twin spool rotor systems as adopted in modern aircraft engine power plants. The matrix size becomes 33x33. The inter-shaft bearing conditions have to be derived for each specific case and therefore the transfer matrix method has a limitation in making the procedure general in nature. Transfer matrix methods have serious limitations, e.g., in considering the gear mesh stiffness in a train. Another limitation is to consider the effect of casing/foundation stiffness on the rotor dynamics. Modern day design of rotors couples all these motions and it appears that the transfer matrix method will gradually be replaced by others in the very near future. The modern trend is to use finite element method to rotor dynamics, taking into account Timoshenko shear correction factor and gyroscopic effects. However, most use beam models even now which suffer from many disadvantages: 1.Real life rotors are not one-dimensional. 2. Considerable time and effort are involved in making the model, 3. The influence of disks on shafts, vice versa is not possible, 4. Centrifugal effects of distributed shafts and mounted parts cannot be accounted, 5. Gyroscopic effects are calculated as separate elements and the speed does not enter into the determination of critical speeds and 6. The foundation and casing effects are to be determined by sub-structuring analysis. An accurate rotor dynamic analysis needs solid models for the rotors. Analysis of solid rotor models progressed rapidly in the last few years. Initially an example of an aero-engine power plant twin spool rotor shown in Fig. 1 was considered to benchmark the results. The Campbell diagram is shown in Fig. 2. Forward whirl gradually increased from 100.2 Hz to 111.4 Hz, whereas the backward whirl dropped to 5.018 Hz. At 630 rad/s spin speed, there is only one frequency detected which is forward whirl 111.4 Hz. Thus the backward whirl is dominated by spin softening effect compared to the forward whirl frequencies where the stress stiffening plays a significant role. Beam models cannot predict the backward whirl modes as stress stiffening and spin softening influences cannot exist with the structure as a line element. Deriving beam models for real life engines is time consuming and arguable; the casing and support structure is very light in construction and therefore participates in coupled motion with the rotor. As a first step, for civil transport aircraft, this casing may be considered rigid. For modern state-of-art military aircraft engines, however, such a simplification creates problems. Recent advances give us the ability to use solid models directly resulting in Model-Mesh-Analyze approach to rotor dynamics. An aircraft engine mode shape is illustrated.

3000

2500 WHIRL SPEED (rad/s)

2000

1500

1000

500

0 0 500 1000 1500 2000 2500 3000 R O T O R 1 S P IN S P E E D ( ra d/s )

Fig. 1 A Twin Spool Solid Rotor Model

Fig. 2 Campbell Diagram

Fig. 3 Rotor Dynamics of an Aircraft Engine High speed cryogenic pumps in some cases run up to 100,000 rpm; the bearings become highly nonlinear and seals are responsible for instability regions. Internal pressures can also affect the stiffness of the flexible casing. Today it is possible to perform a nonlinear analysis and typical response of a high speed rotor as shown. High speed rotors are also required to accelerate from start up to full speed in 3-5 seconds. The response at which peak response is obtained can be obtained by transient analysis. The transient analysis for the unbalance response of a highly accelerating rotor is depicted.

Fig. 4 Rotor Dynamics of A High Speed Pump and Its Transient Response Rotors have mounted parts and Turbomachine Bladed disks are most stressed components, globally elastic but locally plastic and therefore subjected to fatigue failures, particularly when they have a material or manufacturing defect in the stress riser regions. Strain Based method is most appropriate to estimate life. Life estimation depends on how accurately the resonant stresses at a critical speed are determined. A major factor affecting this process until recently is an appropriate and realistic damping model. Todays technology allows us to analytically determine the damping due to material as well as friction in a given mode of vibration at a given speed of operation and as a function of reference strain amplitude, as shown in Fig. 5

0.18 0.16 0.14 0.12 0.1 0.08 0.06 0.04 0.02 0 0 0.00001 0.00002 Amplitude (m) 0.00003 0.00004 200 RPM 500 RPM 1000 RPM

Fig. 5 Analytically Predicted Combined Material and Friction Damping as a function of reference Strain Amplitude at a given Speed and in a given Mode of Vibration Life estimation process has become crucial at design stage and is illustrated in the following chart.
Steady Loads Centrifugal Loads Static Analysis Mean Stress Natural Frequencies Mode shapes Modal Analysis Dynamic Stress Critical speeds Damping Models Life Estimation Excitation Forces

Thermal Loads

Goodman Diagram S-N Law Failure Surface Cumulative Damage

Damping Ratio

True Stress Neubers Law Local Strain Strain based Crack Initiation Life

LEFM - SIF Crack Initiation

Crack Propagation Unstable Crack Fracture Life

The steady load analysis in the upper left hand side of chart 1 for mean stresses is well established; the main sources are centrifugal loads, thermal loads and gas loads. The dynamic stress module is usually complicated, particularly while estimating the alternating force definition and damping definition. Current practices of CFD allow an accurate determination of the force field in a turbine or compressor stage with moving row of blades. Modal analysis of rotating blades is a complex subject that attracted attention of researchers over the last five decades. The governing equation of this system can be derived from basic energy principles as given. Contributions of each term in the equation are identified in the same equation. Solutions for such highly nonlinear equation are complicated; they have been achieved for special cases, e.g., the response in fundamental mode under transient conditions. The nonlinear terms become important for long slender blades such as helicopter blades under acceleration. Fortunately many of these nonlinear and Coriolis terms have negligible influence in aircraft engine blades as far as forced vibration at critical speeds is concerned

in determination of stress field.

Translation and Rotational Inertias Beam Stiffness Stiffness due to rotation Hardening Softening

& &+ y

xx & & + y A

EI

xx y + + t 2 (R + Z ) y y R y + 0 2 A
Nonlinear Coriolis Forces

){

+2

l z & & dz y y & dz + t y y y y + 0 0 0

l 1 z 2 + y y y dz y ydz + (R + Z ) = 0 20 0 Force due to acceleration

independent of y Nonlinear terms due to acceleration

The natural frequencies and mode shapes of turbine and compressor blades are determined following well established Finite Element methods and commercial codes. Campbell diagram can then be drawn to identify the critical speeds. At these critical speeds, the damping is estimated as outlined. The resonant stress is determined using the nonlinear damping model by an iteration process and the stress variation around the critical speed is obtained by using dynamic magnifier relation. The steady state and dynamic stresses together are used in a cumulative damage calculation process to assess the life. This tedious and time consuming life estimation is recently developed as a user friendly code. Aircraft engines work in limits; they demand as high possible life that is permissible under globally elastic and locally plastic conditions and in addition minimum weight consistent with structural integrity. In earlier practices, dedicated codes are developed to achieve a specific optimization problem but are limited to one or two objective functions and few design constraints in a real life structural problem. Several attempts have been made in recent years to develop commercial user friendly optimization codes. One such general optimization code is Altair HyperStudy which uses global optimization methods. These methods use higher order polynomials to approximate the original structural optimization problem over a wide range of design variables. The polynomial approximation techniques are referred to as Response Surface methods. A sequential response surface method approach is used in which, the objective and constraint functions are approximated in terms of design variables using a second order polynomial. One can create a sequential response surface update by linear steps or by quadratic response surfaces. The process can also be used for non-linear physics and experimental analysis using wrap-around software, which can link with various solvers. A recent study demonstrated optimization of a blade root shape, the parameters varied consistent with manufacturing constraints and available space. The optimized root shape reduced plastic strain at the singularity by 26%. Using strain based method of lifing this strain reduction increased the life as much as 4 times. Another significant optimization reported recently is concerned with weight optimization. The lean stress regions are used to decrease the weight of the blade shank by 10%

Region of interest Optimized Shape Baseline Shape

(a) Shape

(b) Weight

Fig. 6 Blade Root Optimization