Dynamic Analysis Of Rotors Supported

On Journal Bearings By Solving Reynolds

Equation Using Pseudospectral Method

Sudhakar Gantasala

Dr. Praveen Krishna I R

Indian Institute of Space Science and Technology

Prof. A S Sekhar
Indian Institute of Technology Madras

IFToMM ICORD
23-Sep-2014
 Objectives
 To solve Reynolds equation using pseudospectral method (PSM).

 To compare accuracy, computational efficiency of

pseudospectral method (PSM) with the finite element method
(FEM).
 1. Literature survey
 Vance (1988) gave analytical expressions for short bearing fluid
film forces with π film cavitation.
 Kirk & Gunter (1976) studied rigid rotor supported on journal
bearings using short bearing theory for its stability and
predicted transient behavior.

 Booker & Huebener (1972), Allaire et al. (1977) solved

Reynolds equation using FEM.

 Boyd (2000), Trefethen (2000), Shen & Tang (2006) explained

about application of pseudospectral method for solving partial
differential equations.
 Raad and Karageorghis (1993) solved Reynolds equation using
Chebyshev spectral collocation method to study loads
supported by ultra-thin gas bearings used in hard disk drives.
 2. Fluid film forces
Reynolds equation for journal bearing fluid film pressure
1   3 p    3 p   1 Uh  h 
 h    h   6    2  (1)
R 2     z  z   R  t 
Θ

h  c  e X cos   e Y sin  
'
θ

ω
Fluid film forces are calculated using
Y
 Short bearing theory
ϕ
 Finite element method (FEM) Fr
(3 node triangular element,
9 node quadrilateral element) Ft
X
 Pseudospectral method (PSM)
 2. 1 Short bearing theory
Short bearing fluid film pressure
3  1 Uh h   2 L2 
p 3   2   z   (2)
h  R  t   4
Short bearing fluid film forces
2 L/2
FX    p(, z) R cos d dz
0 L / 2 (3)
2 L/2
FY    p(, z) R sin  d dz
0 L / 2

Short bearing fluid film forces

L  2  1  2 2  
2

Fr    R L     2   5/ 2 
 c   1    2 1    
2 2 2
FX  Fr 
(4)    T  
    2     FY   Ft 
2

Ft   R L     2 
L
 2
 c   4 1    1    
2 3/ 2 2
 2. 2 Finite element method (FEM)
Finite element solution of Reynolds equation yields
(Childs, 1993)

K p  K U  K h

p Ux x h
(5)

where 3 node triangular

 Li L j Li L j 
h 3 element
Kpij       dA,
A 12  x x z z 
L
KUxij   h i L j dA,
A x
KH ij    Li L j dA,
A

9 node quadrilateral
element
 2. 3 Pseudospectral method (PSM)
Consider a differential equation
Lu ( x)  s( x) subjected to boundary condition, Bu ( y )  0 (6)

Assume a solution in terms of sum of orthogonal polynomials

N
u ( x)  uˆ ( x)   uk k ( x) (7)
k 0

k ( x)  e jkx, Fourier polynomials

k ( x)  Tk ( x), Chebyshev polynomials
Residual obtained after substituting uˆ ( x) in differential equation
RN ( x)  Luˆ ( x)  s( x) (8)
In weighted residual method, integral sum of weighted error function
is made zero.
 Luˆ ( x)  s( x)  w( x) dx  0 (9)
 Luˆ ( x) w( x) dx   s( x) w( x) dx
In collocation method, delta function at a specific location is
considered as weight function which results in algebraic equations.
N

  L(u  ( x))  ( x  x ) dx   s( x)  ( x  x ) dx
k 0
k k n n

 L( ( x )) u  s( x )
k 0
k n k n (10)

Du  s
Here the operator L on variable u is transformed to a matrix [D].
The differentiation is achieved by a matrix multiplication on the
vector u.
Consider a signal p0, p1, p2, …. pN-1 then its discrete and inverse
Fourier transform pair are given by
1 N 1 2 i k t / T
Pk   pl e l

N l 0 (11)
N 1
pl   Pk e 2 i k t / T
l

k 0

Differentiation on pl gives
dpl 2 N 1

dtl

T
i k P e 
k 0
k
2 i k tl / T

2 N 1
 1 N 1 2 i k t / T  2 i k t / T
  i k   pl e j
e
l

T k 0  N j 0 
(12)
N 1
 2 1 N 1 2 i k ( l  j ) / N 
   i k e  pj
j 0  T N k 0 
d
p  Dp
dt
Reynolds equation for journal bearing fluid film pressure
1   3 p    3 p   1 Uh  h 
 h    h   6    2 
R 2     z  z   R  t 

Assume pressure in terms of Fourier and Chebyshev polynomials as

N 1 M
p( , z )   pnm e jn Tm ( z ) (13)
n 0 m0
L

z
0
0 2π
θ
Journal bearing
Unwrapped bearing geometry

Reynolds equation results below algebraic equations after inserting

differentiation matrices through tensor products
[ DF ]  [ I M 1 ][ A]  [ I M 1 ][ DF ]  [ I M 1 ]
 [ I ]  [ D ][ B]  [ I ][ I ]  [ D ]  p   f  (14)
 N C M 1 N C 
3. Comparison of fluid film forces obtained
from different methods
Journal bearing details
Diameter – 50.8 mm
Y Length – 12.7 mm
Radial clearance – 50.8 μm
Oil viscosity – 0.00689 Pa-s
Journal speed – 10000 rpm
X
Journal bearing

Journal bearing forces are calculated at an eccentricity (ε) and phase

angle (ϕ).
eX   c cos( ), eY   c sin( ), eX  0.1 eX  , eY  0.1 eY 
 3 node triangular element mesh 9 node quadrilateral element mesh
24 x 10 elements 10 x 5 elements

PSM grid points

24 x 10 polynomials
Gauss-Lobatto
Quadrature points
 
z j  cos j
M 

2
Fourier quadrature points  i  i
N
Table 1: Fluid film forces calculated for different triangular element mesh sizes in FEM
Journal position Mesh 24 x 10 Mesh 36 x 20 Mesh 48 x 30
Eccentricity Attitude
FX (N) FY (N) FX (N) FY (N) FX (N) FY (N)
(ɛ) angle (φ)
0.3 45˚ -48 10 -48 11 -47 11
0.5 45˚ -138 -4 -137 -2 -137 -1
0.7 45˚ -468 -128 -470 -124 -470 -123
0.8 45˚ -1101 -465 -1126 -466 -1135 -465
0.9 45˚ -4107 -2546 -4486 -2724 -4646 -2812

Table 2: Fluid film forces calculated for different quadrilateral element mesh sizes in FEM
Journal position Mesh 12 x 5 Mesh 18 x 10 Mesh 24 x 15
Eccentricity Attitude
FX (N) FY (N) FX (N) FY (N) FX (N) FY (N)
(ɛ) angle (φ)
0.3 45˚ -47 11 -47 11 -47 11
0.5 45˚ -136 0 -136 -1 -136 -1
0.7 45˚ -469 -118 -469 -119 -469 -119
0.8 45˚ -1139 -454 -1140 -458 -1141 -459
0.9 45˚ -4709 -2749 -4809 -2875 -4821 -2885

Table 3: Fluid film forces calculated using different number of polynomials in PSM
Journal position Polynomials Polynomials Polynomials
24 x 10 36 x 10 48 x 10
Eccentricity Attitude
FX (N) FY (N) FX (N) FY (N) FX (N) FY (N)
(ɛ) angle (φ)
0.3 45˚ -47 11 -47 11 -47 11
0.5 45˚ -136 0 -136 0 -136 0
0.7 45˚ -469 -119 -469 -119 -469 -119
0.8 45˚ -1139 -459 -1140 -459 -1140 -459
0.9 45˚ -4729 -2853 -4830 -2899 -4830 -2896
 Table 4: Computational time required for one calculation of fluid film forces
using FEM and PSM
FEM: Triangular elements FEM: Quadrilateral elements PSM
No. of No. of No. of
Time (s) Time (s) Time (s)
elements Elements Polynomials
24 x 10 0.147 12 x 5 0.194 24 x 10 0.015
36 x 20 0.458 18 x 10 0.595 36 x 10 0.027
48 x 30 0.934 24 x 15 1.302 48 x 10 0.043
Note: All these are computed on Intel Core i5-3337U CPU @ 1.80GHz, (4 GB
RAM, 64 bit, WINDOWS 8) using Scilab 5.5.0. Tic, toc commands of Scilab are
used to get time taken for one calculation of force values.

-4000
1 2 3 Triangular element

-4200 Quadrilateral element

PSM
Force (Fx in N)

-4400

-4600

-4800

-5000
Case number

Convergence plot of force values for 0.9 eccentricity with different number of elements used
in FEM and number of polynomials used in PSM
 Table 5: Computational time required for one calculation of fluid film forces
using FEM and PSM
FEM: PSM
Short bearing Short bearing
Journal Quadrilateral (Number of
(Analytical (From Pressure
position elements (Mesh polynomials
forces) integration)
size 12 x 5) 24 x 10)
ɛ φ FX (N) FY (N) FX (N) FY (N) FX (N) FY (N) FX (N) FY (N)
0.3 45˚ -49 12 -49 11 -47 11 -47 11
0.5 45˚ -142 1 -146 -4 -136 0 -136 0
0.7 45˚ -509 -120 -533 -149 -469 -118 -469 -119
0.8 45˚ -1289 -487 -1382 -595 -1139 -454 -1139 -459
0.9 45˚ -6073 -3392 -6886 -4325 -4709 -2749 -4729 -2853

 At low eccentricities, forces predicted by short bearing theory and

numerical methods matches well.

 But at high eccentricities, short bearing theory over predicts forces. So

at high eccentricities numerical methods are accurate for values.

 We can achieve same accuracy as FEM using PSM and also with less
computational time.
 FEM PSM
Input Input
(Bearing (Bearing
geometry, geometry,
speed) speed)

Element matrices Film thickness matrix

Tensor products to
Global matrix assembling calculate differentiation
matrix
Boundary conditions,
Boundary conditions,
Matrix inversion Matrix inversion

Pressure integration to Pressure integration to

calculate forces calculate forces

 Matrix assembling in FEM is the time consuming part but these matrices are
sparse matrices which needs less memory and results in fast matrix inversions.
 Tensor products in PSM are the time consuming part and it results dense matrix
which takes more time for matrix inversion.
 PSM replaces costly matrix assembly process in FEM with tensor products
which needs less computational effort.
 4. Conclusions
 Reynolds equation is solved for fluid film pressure using Pseudospectral
method (PSM).

 PSM is accurate as FEM (for the plain bearing case studied here) and it
calculates fluid film pressure with less computational effort than FEM
(both triangular and quadrilateral elements).
 REFERENCES
1. Vance, J. M., 1988, “Rotordynamics of turbomachinery”, John Wiley & Sons Ltd.
2. Kirk, R. G. and Gunter E. J., 1976, “Short bearing analysis applied to rotor dy-
namics, Part 1: Theory”, ASME Journal of Lubrication Technology, Vol. 98(1),
pp. 47-56.
3. Kirk, R. G. and Gunter E. J., 1976, “Short bearing analysis applied to rotor dy-
namics, Part 2: Results of journal bearing response”, ASME Journal of Lubrica-
tion Technology, Vol. 98(2), pp. 319-329.
4. Booker, J. F. and Huebner, K. H., 1972, “Application of finite element methods
to lubrication: An engineering approach”, ASME Journal of Lubrication Technol-
ogy, Vol. 94(4), pp. 313-323.
5. Allaire, P. E., Nicholas, J. C. and Gunter, E. J., 1977, “Systems of finite elements
for finite bearings”, ASME Journal of Lubrication Technology, Vol. 99(2), pp.
187-194.
6. Childs, D., 1993, “Turbomachinery Rotordynamics: Phenomena, Modeling, and
Analysis”, John Wiley & Sons Ltd.
7. Boyd, J. P., 2000, “Chebyshev and Fourier spectral methods”, Dover publica-
tions.
8. Trefethen, L. N., 2000, “Spectral methods in MATLAB”, Society for Industrial
and Applied Mathematics (SIAM).
9. Shen, J. and Tang, T., 2006, “Spectral and high-order methods with applications”,
Science press, Beijing.
10. Raad, P. E. and Karageorghis, A., 1993, “A Chebyshev spectral collocation
method for the solution of the Reynolds equation of lubrication”, Journal of
Computational Physics, Vol. 106, pp. 42-51.