Tolerance Analysis of

Compliant Metal Plate

Assemblies Considering
Welding Distortion
Hyun Chung, Ph.D.
Postdoctoral Research Associate
Dept. of Naval Architecture and Marine Engineering
University of Michigan

August 28 2006
Overview
 Motivation and Perspective
 Research Objective
 Background
 Current Research
 Inherent strain method
 Mechanistic variation simulation
 Three-bar model
 Modified method of influence coefficients
 Design Implication
 Conclusion and Future Work Recommendations
Motivation and Perspective (1)
 Ship block construction processes
 Thick metal plates ( 10mm~30mm )
 Compliant behavior due to large sizes
 Required accuracy is about 1/100,000
 Joined primarily by heat-flux welding that causes
significant distortions
 Intermediate products that fail to meet the tolerance
requirements are not scrapped, but reworked
 Dimensional quality of intermediate products not only
affects final product quality but also productivity of
shipyard
Motivation and Perspective (2)
 Excessive variation due to warpage, shrinkage, etc may
lead to tolerance stack-up that results in increased
assembly stresses
 Traditional tolerance analysis usually relies upon a rigid
body assumption, which cannot explain compliant
characteristics of metal plate assembly
 Various research work on tolerance analysis and tolerance
synthesis for automotive body and machining production
[S.C.Liu and S.J.Hu 1997]
 Assembly sequence can positively affect the welding
distortions during assembly process without any mitigation
processes [ J.K.Roh and J.G.Shin 1999]
Motivation and Perspective (3)
 Statistically predict dimensional quality of ship
block assemblies
 Non-nominal plate geometry
 Compliant metal plates
 Welding distortion
 Fast and efficient calculation
Motivation and Perspective (4)
 Ship block construction processes (cont.)
 Research has provided a means to determine statistical
variation in ship block assembly including distortion of
welded structure in minutes instead of hours with equivalent
accuracy compared to current methods.
 Usually requires many runs and table below shows research
benefit per run
Modified Method of
Method Direct Monte Carlo
Influence Coefficient

Average distortion 0.003398 0.003746

Standard deviation of distortion 0.0007 0.0004

No. of FEM runs 1000 3

Real CPU time (sec) 12082 25

Motivation and Perspective (5)
 Different assembly sequences make different dimensional
qualities
 Welding distortions are heavily affected by structural and thermal
boundary conditions
 Different sequences make different boundary conditions
Motivation and Perspective (6)
Research Objectives (1)
 Combination of tolerance analysis and finite
element analysis to develop compliant assembly
model of ship block construction

 Fast and efficient calculation

 The effect of heat-flux welding deformation and
residual stresses considered in variation simulation for
compliant assemblies
 Identification of factors that control the deformation of
non-nominal plate assemblies after being welded
Research Objectives (2)
 Validation of compliant assembly model through ship
block assembly example
Background (1)
Background (2)
 Variation simulation
 Traditional variation analyses
 Mechanistic variation analysis
 Method of Influential Coefficients
 Welding distortion prediction
 Simplified load
 Inherent strain method
 3D thermo-elasto-plastic FEA

 Statistical representation of plate part variations

 B-Spline Beam/Plate element
Background (3)
 Traditional variation analysis
 Worst case [Greenwood and Chase 1988]
 Assumes that all parts are built to its extreme values
 Too restrictive
 Statistical analysis [Chase and Parkinson 1991]
 Variations of parts are specified as statistical distribution
 Root Sum Square (RSS)
 Monte-Carlo simulation [Craig 1988, Doepker and
Nies 1989, Early and Thomson 1989]
 Straightforward
 Computationally expensive
Background (4)
 Traditional variation analysis (cont.)
 Individual parts are considered as rigid bodies
 Aggregate behaviors are determined by geometric
and/or kinematic relations
 Assembly sequence does not affect the dimensional
variations of final product

 Mechanistic variation simulation [S.C.Liu and S.J.Hu

1995]
 Compliant sheet metal assemblies
 Sequence of assembly can affect the tolerance of final
product [ Liu, Hu and Woo 1995]
Background (5)
 Method of Influence Coefficients as the more
efficient method of calculating variations due to
compliant assemblies [Liu and Hu 1995]
 Statistically combining the sensitivity matrix with root
sum square method to determine mean and standard
deviation, without Monte-Carlo simulation
 Use of sensitivity matrix to interpret geometrical
imperfection to equivalent external forces
 Applied only to simple geometries
 Valid only in elastic deformation range
 Welding distortions not considered
Background (6)
 Welding deformation prediction
 Simplified load method
 Simple rule-of-thumb based on previous production data

 Only applicable to existing cases

 Cannot predict transient shape change

 Numerical simulation method (Inherent strain/load

method)
 Assumption of inherent strain distribution
 Cannot predict transient shape change
 3-Dimensional thermo-elasto-plastic FEA
 Computationally expensive

 Can predict transient shape, temperature, stress/strain

distribution change
Background (7)
 Statistical representation of plate part variations
 Random Bézier Curve [Merkley K., 1998]
 Good for plate gap modeling
 Not applicable to plate shape
 Finite Element Mesh
 Straightforward
 Cannot accurately represent slope and curvature
 B-Spline Beam/Plate Element
 B-Spline shape functions
 Not applicable to statistical shapes
Current Research (1)
Current Research (2)
Current Research (3)
Current Research
Inherent Strain Method (1)
 Originally developed to predict residual stress within
structure [Ueda, 1975]
 Equivalent to ‘Eigenstrain’ [Mura, 1982]
 Hybrid method (partially analytic, partially experimental)
 Inelastic and non-compatible strain that accounts for
plastic deformation of the body
 Defined as dS 2  dS 0
 
*

dS 0
Inherent Strain Method (2)
 What are the stress/strain states when the temperature is
increased by 1 degree given that the individual bars have
different thermal expansion coefficients?
 Sequence
 Stress free initial state
 Stress free state after unit temperature increase
 Final state
 Total strain = elastic strain + inherent strain

  e *

 Residual stress
  E (   * )  E e
Inherent Strain Method (3)
 Three-bar model for welding distortion
prediction
 Welding distortion is caused by highly non-
uniform temperature distribution in the welded
region
 Only middle bar undergoes temperature change
 Elastic-perfect plastic material assumed
 Stress history in the middle bar
 OA: elastic compression
 AB: plastic compression
 BC: elastic tension
 CD: plastic tension
Inherent Strain Method (4)
Inherent Strain Method (5)
Inherent Strain Method (6)
Inherent Strain Method (7)
 Based on three-bar model, welding
distortion is only dependant on
 Material
 Maximum temperature
 Degree of restraints,
KS

KW  K S

 DOR is determined by
 Plate thickness
 Welding type
 Butt or fillet welding
 Welding conditions
Inherent Strain Method (8)
 Equivalent loading based on Inherent
Strain [Lee, 2002]
 Equivalent loads are forces and moments
relevant to the deformation modes of welded
plates, which are the integration of inherent
strain distribution

1 1 0 bw
fy       E *y dydz
*
Ey dA
bw bw h 0
At

1 * h 1 0 bw h
my 
bw  y2
At
E (  z ) dA 
bw  
h 0
E *y (  z )dydz
2
0 bw
Fx   E x* dA    E x* dydz
h 0
At

h 0 bw h
M x   E x* (  z )dA    E x* (  z )dydz
2 h 0 2
At
Inherent Strain Method (9)
 Compared to 3D FEA result
 ~1/500 computation time
 Relatively high accuracy
 Longer weld line results in higher accuracy
Current Research
Mechanistic Variation Simulation (1)

 Method of Influence Coefficients [Liu and

Hu, 1995]
 Elastic deformation only
 No welding distortion (spot welding)
 Linear mapping [S ] between part variation {V }
and assembly deviation {U }
Fw  [ K w ]{U w}  [ Ku ]{Vu }  Fu

{U w }  [ K w ] 1 [ K u ]{Vu }  [ S wu ]{Vu }

 The sensitivity matrix [S ] is obtained by

unit force response
Mechanistic Variation Simulation (2)
 Method of Influence Coefficients [Liu and Hu, 1995]
 Small variations are assumed
 Unit force is applied at j-th source of variation
 Deformation under unit force is calculated by FEM
 Obtain the influence of j-th variation to N-th node
 Spring-back is calculated by FEM

 V1  N  c1 j   c11  c1N   F1 
     
{V }         F j           [C ]{F }
V  j 1 c  cN 1  c NN   FN 
 N  Nj 

[ F ]  [C ]1{V }  [ K ]{V }

 U1  N  s1 j   s11  s1N   V1 
     
{U }         V j           [ S ]{V }
U  j 1 s   sM 1  sMN  VN 
 M  Mj 
Current Research

How part variation

affects welding
distortion?

How welding
distortion affects
variation
simulation?
Three Bar Models (1)
 Clamped-clamped 3-bar model with initial
part variation
 The structure is clamped to its nominal length
and undergoes temperature changes, and then
clamping is released
 Displacement is confined
 Model for the region adjacent to clamping
devices
 Spring-clamped 3-bar model with initial
variation
 External force is applied to the structure and the
structure undergoes temperature changes, and
then the external force is removed
 Clamping force is constant
 Model for the region far from clamping devices
Three Bar Models (2)
 Procedure
 Step1: initial variation
 Step2: clamping to nominal position
 Elastic deformation
 Step3: welding
 Elastic state during heating
 Plastic state during heating

 Elastic state during cooling

 Plastic state during cooling

 Step4: releasing clampings

 Elastic deformation
Three Bar Models (3)
Three Bar Models (4)
Three Bar Models (5)
Three Bar Models (6)
Three Bar Models (7)
Three Bar Models (8)
Modified Variation S imulation (1)

 Modified method of influence coefficients

[ K w ]{U w }  [ K u ]{Vu }  Fd 
{U w }  [ K w ]1 [ K u ]{Vu }  [ K w ] 1 Fd   [ S wu ]{Vu }  [ K w ] 1 Fd 
{U w}  [Swu ]{Vu }  D

 Initial part variation affects welding distortion as it slightly changes

degree of restraints
 Due to the Linearity between the residual strain and initial variation, the
welding distortion term of non-nominal parts can be replace as the welding
distortion term of nominal parts
Dvar   D   AVu 

 The differences could be absorbed in the sensitivity matrix based on the

value found on the inherent strain chart
Current Research
Plate variation representation (1)
 Cubic B-Spline finite element
 New basis function that require two more control points is adopted
 Equally spaced splines
 1
 6h3 ( x  xi  2 ) xi  2  x  xi 1
3


 1 {h3  3h 2 ( x  x )  3h( x  x ) 2  3( x  x )3} xi 1  x  xi
 6h3 i 1 i 1 i 1


i ( x )   1 3
 6h3 {h  3h ( xi 1  x)  3h( xi 1  x)  3( xi 1  x) } xi  x  xi 1
2 2 3


 1 ( x  x )3 xi 1  x  xi  2
 6h3 i  2
0
 otherwise
Plate variation representation (2)
 Cubic B-Spline plate element
 Displacement m1 n 1
w    cij i ( x) j ( y)   N  s 
j 1 i 1

T
 s   c1 c0 c1 cm1 cm cm1 
T T T T T T
where

c j  c1, j
T
c0, j cn 1, j 

   1 0 n1 

    1 0  m1 

 N      
Plate variation representation (3)
 Cubic B-Spline plate element
 Strain-displacement relationship
 2w 
 2 
 x        
  2 w  
     2           s    B  s 
 y   2     
 2w   
2 
 xy 
 Strain-stress relationship

 
1  0 
Et 3
 
    1 0    D     D  B  s 
12(1  2 ) 
(1  ) 
0 0 
 2 
Plate variation representation (4)
 Cubic B-Spline plate element
 Stiffness matrix with respect to Spline parameters
 K s   y x  B   D  B  dxdy
ym xn T

0 0

 Displacement relationship
1
wi , j  {16ci , j  4(ci , j 1  ci 1, j  ci , j 1  ci 1, j )  (ci 1, j 1  ci 1, j 1  ci 1, j 1  ci 1, j 1 )}
36

 w  1
ix, j     {4(ci 1, j  ci 1, j )  (ci 1, j 1  ci 1, j 1  ci 1, j 1  ci 1, j 1 )}
 x i , j 12hx

 w  1
iy, j     {4(ci , j 1  ci , j 1 )  (ci 1, j 1  ci 1, j 1  ci 1, j 1  ci 1, j 1 )}

 i , j
y 12hy

 2w  1
 xy
   (ci 1, j 1  ci 1, j 1  ci 1, j 1  ci 1, j 1 )
 
i, j
 x y i , j 4 h h
x y
Plate variation representation (5)
 Cubic B-Spline plate element
 Stiffness matrix with respect to global coordinate system
 K   T   K s T 
T 1

where K  is (n  3)(m  3)  (n  3)(m  3) square matrix

 1 4 1 4 16 4 1 4 1 
 36 36 36 36 36 36 36 36 36 
 
 1 0
1

4
0
4

1
0
1 
 12hx 12hx 12hx 12hx 12hx 12hx 
Tc i , j 
 1 4 1 1 4 1 

  0 0 0
 12h 12hy 12hy 12hy 12hy 12hy 
 y

 1 1 1 1 
 4h h 0  0 0 0  0
 x y 4hx hy 4hx hy 4hx hy 
Plate variation representation (6)
 Cubic B-Spline plate element example
 Plate deflection under concentrated load
(C-C-C-C) and (S-S-S-S)
 Plate variation representation (7)
 Cubic B-Spline plate element example
Varmean   wA  0.02 wB  0.01 wC  0.01 wD  0.01
T

Varstdev   0.01 0.01 0.01 0.01

T
Design Implication
Design Implication (1)
Method Direct Monte Carlo Modified Method of Influence Coefficient

Mean value of center distortion 0.003398 0.003746

Standard deviation of center distortion 0.0007 0.0004

No. of FEM runs 1000 3

Real CPU time (sec) 12082 25

Design Implication (2)
Design Implication (3)
Design Implication (4)
Design Implication (5)
Design Implication (6)
Design Implication (7)
ANSYS 0.0022 0.0013665 0.0014263 0.00135 0.0019

Modified Method of
0.0028 0.001312 0.0014762 0.001312 0.0028
Influence Coefficients
Conclusions
 Inherent strain concept based on three-bar model is well
applied to the prediction of welding distortion since it
captures the core mechanism of welding distortion
generation
 Welding distortion is affected by initial part variation as
the force required to clamp the part into their nominal
position is absorbed in the degree of restraints
 The modified method of influence coefficients can
predict ship block assembly variation including welding
distortion with equivalent accuracy with FEA in much
less computation time
 Statistical B-Spline plate element is developed and it can
be effectively applied to variation simulation when slope
and curvature of final assembly are important
Future Research
 Application to free-from plate assemblies
 Current equivalent loading method based on inherent
strain only works for flat plate assemblies
 Higher order B-Spline element is required
 Optimal assembly sequence determination
 Including welding distortion mitigation techniques
 Further development of inherent strain method
 3D Poisson effects
 Covariance representation
 Material covariance
 Geometrical covariance
Thank you for your attention
Hyun Chung
 Research Interest
 Tolerance Analysis/Optimization for compliant metal plate assembly
considering welding distortion
 3D FEM welding distortion simulation and Equivalent loading method
based on eigenstrain concept
 Statistical tolerance representation of 3D surface using B-Spline FEM
 Sensitivity matrix model for compliant tolerance analysis
 Simulation based tools to support decision making in ship production
systems
 Object-oriented modeling of production systems
 Artificial intelligence tools for production planning
 Simulation based manufacturing facility design
 Education
 2000.8 MSE in NAME, University of Michigan
 2006.3 Ph.D in NAME, University of Michigan
 Publications
 Simulation-based Performance Improvement for Shipbuilding
Processes, JSP Vol. 22, No. 2, 2006
 A Generic Shipyard Computer Model Development – A Tool for Design
for Production, JSP, Vol. 16, No. 3, 2000.