application of a functional adaptive simulation … · 2016-02-20 · taguchi principles...
TRANSCRIPT
05
./ 0
6.1
1.2
01
5 –
WO
ST
20
15
Institut für Maschinenelemente, Universität Stuttgart
Technologie Transfer Initiative, Universität Stuttgart
Rob
ust
Se
nsitiv
e
Robust Design
in commercial vehicle braking systems
12. Weimar Optimization and Stochastic Days 2015
WOST 2015
Dipl.-Ing. Stefan Kemmler
M.Sc. Alexander Fuchs *
Dr.-Ing. Tobias Leopold *
Prof. Dr.-Ing. Bernd Bertsche
* Knorr-Bremse Group
Systeme für Nutzfahrzeuge GmbH
Institute of Machine Components
Reliability Engineering
Technologie Transfer Initiative
APPLICATION OF A FUNCTIONAL ADAPTIVE
SIMULATION MODEL FOR THE ROBUST
PRODUCT-OPTIMIZATION
05
./ 0
6.1
1.2
01
5 –
WO
ST
20
15
Institut für Maschinenelemente, Universität Stuttgart
Technologie Transfer Initiative, Universität Stuttgart
S. K
em
mle
r; A
. F
uch
s; T
. L
eo
po
ld; B
. B
ert
sch
e
Outline of the presentation
Introduction to the optimization strategies RDO and Taguchi
Application example: adjusting unit
Functional adaptive simulation strategy SIM-SMAR2T
RDO and Taguchi in optiSLang
Summary and conclusion
2Agenda
05
./ 0
6.1
1.2
01
5 –
WO
ST
20
15
Institut für Maschinenelemente, Universität Stuttgart
Technologie Transfer Initiative, Universität Stuttgart
INTRODUCTION TO THE
OPTIMIZATION STRATEGIES RDO
AND TAGUCHI
3
05
./ 0
6.1
1.2
01
5 –
WO
ST
20
15
Institut für Maschinenelemente, Universität Stuttgart
Technologie Transfer Initiative, Universität Stuttgart
S. K
em
mle
r; A
. F
uch
s; T
. L
eo
po
ld; B
. B
ert
sch
e
Taguchi principles
4Introduction to the optimization strategies RDO and Taguchi
Source: Fowlkes, 1995
Step 1:
Reduce variability with the aim
of S/N Ratio.
Step 2:
Go back to your target.
S/N Ratio (i.e.):
Nominal-the-best (type II)
𝜂 = −10 𝑙𝑜𝑔 (𝜎2)
Target
Probability
density
f
Target
Probability
density
f
05
./ 0
6.1
1.2
01
5 –
WO
ST
20
15
Institut für Maschinenelemente, Universität Stuttgart
Technologie Transfer Initiative, Universität Stuttgart
S. K
em
mle
r; A
. F
uch
s; T
. L
eo
po
ld; B
. B
ert
sch
e
Robust Design Optimization (RDO)
Robust optimization
Using different optimization methods
Identification of global and local models of robustness
Usage of expected value as robust value
Target:
Parameter-Optimization
based on merged mean value
and corresponding variance.
5Introduction to the optimization strategies RDO and Taguchi
parameter x2
initial design
global optimum
robust optimum
variance of the target
valid proof of
parameter space
05
./ 0
6.1
1.2
01
5 –
WO
ST
20
15
Institut für Maschinenelemente, Universität Stuttgart
Technologie Transfer Initiative, Universität Stuttgart
S. K
em
mle
r; A
. F
uch
s; T
. L
eo
po
ld; B
. B
ert
sch
e
Comparison of both Robust Design methods
6Introduction to the optimization strategies RDO and Taguchi
Robust Design Optimization (RDO)
Exclusive consideration of sensitive designs
Target and constraint are predefined
Simultaneous optimization of µ and σ
Optimum mostly not feasible regarding manufacturing
Taguchi method(TM)
Clear separation of the parameter through arrays
No direct specification of constraints
Separate optimization of µ and σ
Parameters dimension based on manufacturing aspects
05
./ 0
6.1
1.2
01
5 –
WO
ST
20
15
Institut für Maschinenelemente, Universität Stuttgart
Technologie Transfer Initiative, Universität Stuttgart
APPLICATION EXAMPLE:
ADJUSTING UNIT
7
05
./ 0
6.1
1.2
01
5 –
WO
ST
20
15
Institut für Maschinenelemente, Universität Stuttgart
Technologie Transfer Initiative, Universität Stuttgart
S. K
em
mle
r; A
. F
uch
s; T
. L
eo
po
ld; B
. B
ert
sch
e
Adjusting Unit
Between the brake pad and the brake disc a defined clearance is
necessary
Protection against dragging of the brake
Too large clearance reduces braking efficiency
8Adjusting Unit
adjusting
unit
lever
bridge
brake
caliper
brake
carrier
05
./ 0
6.1
1.2
01
5 –
WO
ST
20
15
Institut für Maschinenelemente, Universität Stuttgart
Technologie Transfer Initiative, Universität Stuttgart
S. K
em
mle
r; A
. F
uch
s; T
. L
eo
po
ld; B
. B
ert
sch
e
Adjusting Unit
Purely mechanical effect relationships
Main function is to compensate the wear of the pads and the disc
Clearance is obtained by geometric elements in the adjusting unit
Function includes five modes of operation:
9Adjusting Unit
shift fork
freewheel
overload clutch
return spring
output (spindle)
Overcome constructive clearance
AdjustmentOverload –decoupling
Retracting movement
Service function
05
./ 0
6.1
1.2
01
5 –
WO
ST
20
15
Institut für Maschinenelemente, Universität Stuttgart
Technologie Transfer Initiative, Universität Stuttgart
FUNCTIONAL ADAPTIVE
SIMULATION STRATEGY SIM-
SMAR2T
10
FE model
Temperature
Fitting Curve
META model (MOP)
Response Surface
SIM-SMART (BASIS model)(1D model with GUI)
Varia
ble
A
Varia
ble
B
Varia
ble
C
Varia
ble
D
ANSYS
Wear
STEP 2:
META model (MOP) modeling with optiSLang
Target: continuous correlation
Variable A
Variable D
FIT
Parameter-
indentification of the
1D model
STEP 1:
SUB model X modeling
META model simulation
Target: Behaviour of input- and output-parameters
META model (MOP)
Response Surface
Variable B
Variable C
Coefficien
t of frictio
n
Dia
mete
r
Wear
Tem
pera
ture
SUB X
optiSLang
STEP 3:
META model (MOP) integration
Interaction between the models
Parameter variation / tolerance analyse
Direct parametersVariable / parameter in
1D model
05
./ 0
6.1
1.2
01
5 –
WO
ST
20
15
Institut für Maschinenelemente, Universität Stuttgart
Technologie Transfer Initiative, Universität Stuttgart
S. K
em
mle
r; A
. F
uch
s; T
. L
eo
po
ld; B
. B
ert
sch
e
Parameter Design – simulation strategy
Define the expedient simulation strategy!
11Functional adaptive simulation strategy SIM-SMAR2T
Source: Kemmler, 2014
Overlo
ad (
no
rm.)
[N
m]
1
0,8
0,6
0,4
0,2
0
12
34
56
7
515
2535
45
FE-
Model
SUB-
MODEL 2
SUB-
MODEL 1
SUB-
MODEL 3
- Forwards - - Adjust - - Overload -
SUB-
MODEL 4
- Backwards -
ϑ
µ
M
ϑ
µ
M
Φ
ϑ
µ
M
Φ
META-
ModelΦ
SIM-
SMAR2T
M M M M
+ + = ..Φ Φ Φ Φ
05
./ 0
6.1
1.2
01
5 –
WO
ST
20
15
Institut für Maschinenelemente, Universität Stuttgart
Technologie Transfer Initiative, Universität Stuttgart
S. K
em
mle
r; A
. F
uch
s; T
. L
eo
po
ld; B
. B
ert
sch
e
Simulation strategy - overview
12Functional adaptive simulation strategy SIM-SMAR2T
FE model
Temperature
Fitting Curve
META model (MOP)
Response Surface
SIM-SMART (BASIS model)(1D model with GUI)
Varia
ble
A
Varia
ble
B
Varia
ble
C
Varia
ble
D
ANSYS
Wear
STEP 2:
META model (MOP) modeling with optiSLang
Target: continuous correlation
Variable A
Variable D
FIT
Parameter-
indentification of the
1D model
STEP 1:
SUB model X modeling
META model simulation
Target: Behaviour of input- and output-parameters
META model (MOP)
Response Surface
Variable B
Variable C
Coefficien
t of frictio
n
Dia
mete
r
Wear
Tem
pera
ture
SUB X
optiSLang
STEP 3:
META model (MOP) integration
Interaction between the models
Parameter variation / tolerance analyse
Direct parametersVariable / parameter in
1D model
Source: Kemmler, 2014
05
./ 0
6.1
1.2
01
5 –
WO
ST
20
15
Institut für Maschinenelemente, Universität Stuttgart
Technologie Transfer Initiative, Universität Stuttgart
S. K
em
mle
r; A
. F
uch
s; T
. L
eo
po
ld; B
. B
ert
sch
e
FE model
Temperature
Fitting Curve
META model (MOP)
Response Surface
SIM-SMART (BASIS model)(1D model with GUI)
Varia
ble
A
Varia
ble
B
Varia
ble
C
Varia
ble
D
ANSYS
Wear
STEP 2:
META model (MOP) modeling with optiSLang
Target: continuous correlation
Variable A
Variable D
FIT
Parameter-
indentification of the
1D model
STEP 1:
SUB model X modeling
META model simulation
Target: Behaviour of input- and output-parameters
META model (MOP)
Response Surface
Variable B
Variable C
Coefficien
t of frictio
n
Dia
mete
r
Wear
Tem
pera
ture
SUB X
optiSLang
STEP 3:
META model (MOP) integration
Interaction between the models
Parameter variation / tolerance analyse
Direct parametersVariable / parameter in
1D model
Simulation strategy – STEP 1
13Functional adaptive simulation strategy SIM-SMAR2T – STEP 1
FE model
Temperature
Fitting Curve
ANSYS
Wear
Variable A
Variable D
FIT
Parameter-
indentification of the
1D model
STEP 1:
SUB model X modeling
META model simulation
Target: Behaviour of input- and output-parameters
Variable B
Variable C
05
./ 0
6.1
1.2
01
5 –
WO
ST
20
15
Institut für Maschinenelemente, Universität Stuttgart
Technologie Transfer Initiative, Universität Stuttgart
S. K
em
mle
r; A
. F
uch
s; T
. L
eo
po
ld; B
. B
ert
sch
e
FE model
Temperature
Fitting Curve
META model (MOP)
Response Surface
SIM-SMART (BASIS model)(1D model with GUI)
Varia
ble
A
Varia
ble
B
Varia
ble
C
Varia
ble
D
ANSYS
Wear
STEP 2:
META model (MOP) modeling with optiSLang
Target: continuous correlation
Variable A
Variable D
FIT
Parameter-
indentification of the
1D model
STEP 1:
SUB model X modeling
META model simulation
Target: Behaviour of input- and output-parameters
META model (MOP)
Response Surface
Variable B
Variable C
Coefficien
t of frictio
n
Dia
mete
r
Wear
Tem
pera
ture
SUB X
optiSLang
STEP 3:
META model (MOP) integration
Interaction between the models
Parameter variation / tolerance analyse
Direct parametersVariable / parameter in
1D model
Simulation strategy – STEP 2
14Functional adaptive simulation strategy SIM-SMAR2T – STEP 2
STEP 2:
META-Model (MOP) Modeling with optiSLang
Target: Continuous Correlation
META-Model (MOP)
Response Surface
05
./ 0
6.1
1.2
01
5 –
WO
ST
20
15
Institut für Maschinenelemente, Universität Stuttgart
Technologie Transfer Initiative, Universität Stuttgart
S. K
em
mle
r; A
. F
uch
s; T
. L
eo
po
ld; B
. B
ert
sch
e
FE model
Temperature
Fitting Curve
META model (MOP)
Response Surface
SIM-SMART (BASIS model)(1D model with GUI)
Varia
ble
A
Varia
ble
B
Varia
ble
C
Varia
ble
D
ANSYS
Wear
STEP 2:
META model (MOP) modeling with optiSLang
Target: continuous correlation
Variable A
Variable D
FIT
Parameter-
indentification of the
1D model
STEP 1:
SUB model X modeling
META model simulation
Target: Behaviour of input- and output-parameters
META model (MOP)
Response Surface
Variable B
Variable C
Coefficien
t of frictio
n
Dia
mete
r
Wear
Tem
pera
ture
SUB X
optiSLang
STEP 3:
META model (MOP) integration
Interaction between the models
Parameter variation / tolerance analyse
Direct parametersVariable / parameter in
1D model
Simulation strategy – STEP 3
15Functional adaptive simulation strategy SIM-SMAR2T – STEP 3
META model (MOP)
Response Surface
SIM-SMART (BASIS model)(1D model with GUI)
Varia
ble
A
Varia
ble
B
Varia
ble
C
Varia
ble
D
Coefficien
t of frictio
n
Dia
mete
r
Wear
Tem
pera
ture
SUB X
STEP 3:
META model (MOP) integration
Interaction between the models
Parameter variation / tolerance analyse
05
./ 0
6.1
1.2
01
5 –
WO
ST
20
15
Institut für Maschinenelemente, Universität Stuttgart
Technologie Transfer Initiative, Universität Stuttgart
S. K
em
mle
r; A
. F
uch
s; T
. L
eo
po
ld; B
. B
ert
sch
e
Transfer in optiSLang
16Functional adaptive simulation strategy SIM-SMAR2T
Step 1:
Store all system
parameters to sensitivity
Step 2:
Transfer the
respective para-
meters to the
respective MOP
Step 3:
Send MOP results to automatized
calculation script
Step 4:
Send results to
sensitivity module
05
./ 0
6.1
1.2
01
5 –
WO
ST
20
15
Institut für Maschinenelemente, Universität Stuttgart
Technologie Transfer Initiative, Universität Stuttgart
S. K
em
mle
r; A
. F
uch
s; T
. L
eo
po
ld; B
. B
ert
sch
e
Results of SIM-SMAR2T
Significant factors are level systems and
plays of the adjusting unit.
17Functional adaptive simulation strategy SIM-SMAR2T
ℎ𝑁 = 𝑓𝑁(ℎ𝑖𝑑𝑙𝑒 − 𝑠𝑐𝑐)
Components
K ball
CW cone washer
LB bearing bushing
LS bearing washer
PS adjusting washer
Abbreviation
AS output clearance
D_I Inner diameter
W angle
EZ short lever arm (X)
KS constructive clearance
ZFS flank clearance
y_EZ short lever arm (Y)
z_H distance lever-brake
Inp
ut P
ara
me
ter
E-Modulus_LS_LB (<1 %)
E-Modulus_CW (<1 %)
E-Modulus_K (<1 %)
D_I_PS (<1 %)
D_I_LB (<1 %)
Z_H (1 %)
W_ZFS (9 %)
y_EZ (12 %)
W_KS (36 %)
W_AS (41 %)
Coefficient of Importance (CoI) [%]
05
./ 0
6.1
1.2
01
5 –
WO
ST
20
15
Institut für Maschinenelemente, Universität Stuttgart
Technologie Transfer Initiative, Universität Stuttgart
RDO AND TAGUCHI IN OPTISLANG
18
05
./ 0
6.1
1.2
01
5 –
WO
ST
20
15
Institut für Maschinenelemente, Universität Stuttgart
Technologie Transfer Initiative, Universität Stuttgart
S. K
em
mle
r; A
. F
uch
s; T
. L
eo
po
ld; B
. B
ert
sch
e
Taguchi in optiSLang – Design of experiments
Taguchi - Workflow in optiSLang
19RDO and Taguchi in optiSLang
Outer array: L12(211)
Inner array: L81(332)
05
./ 0
6.1
1.2
01
5 –
WO
ST
20
15
Institut für Maschinenelemente, Universität Stuttgart
Technologie Transfer Initiative, Universität Stuttgart
S. K
em
mle
r; A
. F
uch
s; T
. L
eo
po
ld; B
. B
ert
sch
e
Taguchi and SIM-SMAR2T in optiSLang
20RDO and Taguchi in optiSLang
design system parameters
1
2
…
n
P1 P2 P3 … Pn
P1 P2 P3 … Pn
P1 P2 P3 … Pn
P1 P2 P3 … Pn
alpha
α1
α2
…
αn
Overlo
ad (
no
rm.)
[N
m]
1
0,8
0,6
0,4
0,2
0
12
34
56
7
515
2535
45
Angle [º]To
rqu
e [N
m]
05
./ 0
6.1
1.2
01
5 –
WO
ST
20
15
Institut für Maschinenelemente, Universität Stuttgart
Technologie Transfer Initiative, Universität Stuttgart
S. K
em
mle
r; A
. F
uch
s; T
. L
eo
po
ld; B
. B
ert
sch
e
Approach for RDO in optiSLang
ARSM-results:
21RDO and Taguchi in optiSLang
Reason: noise
Solution:
Meta-Meta-Model
05
./ 0
6.1
1.2
01
5 –
WO
ST
20
15
Institut für Maschinenelemente, Universität Stuttgart
Technologie Transfer Initiative, Universität Stuttgart
S. K
em
mle
r; A
. F
uch
s; T
. L
eo
po
ld; B
. B
ert
sch
e
30,129,829,5
18,16
18,08
18,00
1,00,80,5 1,00,80,6
766044
18,16
18,08
18,00
0,960,800,64 29,629,4529,3
2,01,851,7
18,16
18,08
18,00
1,41,10,8 1,21,00,8
D_H_V Z_I
Me
an
of
SN
ra
tio
s
R_H_V Z R_H_R_I
W_H_A S_H D_SpH_V Z_A
R_SpH_V Z R_SpH_R_A S_SpH
Main Effects Plot for SN ratios
Signal-to-noise: Nominal is best (-10*Log10(s**2))
30,129,829,5
1,00
0,75
0,5
1,10,80,5 1,00,80,6
766044
1,00
0,75
0,5
0,950,800,65 29,629,4529,3
2,01,851,7
1,00
0,75
0,5
1,41,10,8 1,21,000,8
D_H_VZ_I
Me
an
of
Me
an
s
R_H_VZ R_H_R_I
W_H_A S_H D_SpH_VZ_A
R_SpH_VZ R_SpH_R_A S_SpH
Main Effects Plot for Means
Results of both tools – S/N-Ratio
Taguchi-Method Robust Design Optimization
22RDO and Taguchi in optiSLang
30,129,829,5
18,16
18,08
18,00
1,00,80,5 1,00,80,6
766044
18,16
18,08
18,00
0,960,800,64 29,629,4529,3
2,01,851,7
18,16
18,08
18,00
1,41,10,8 1,21,00,8
D_H_VZ_I
Mean
of
SN
rati
os
R_H_VZ R_H_R_I
W_H_A S_H D_SpH_VZ_A
R_SpH_VZ R_SpH_R_A S_SpH
Main Effects Plot for SN ratios
Signal-to-noise: Nominal is best (-10*Log10(s^2))
05
./ 0
6.1
1.2
01
5 –
WO
ST
20
15
Institut für Maschinenelemente, Universität Stuttgart
Technologie Transfer Initiative, Universität Stuttgart
S. K
em
mle
r; A
. F
uch
s; T
. L
eo
po
ld; B
. B
ert
sch
e
Results of both tools – mean values
23RDO and Taguchi in optiSLang
0
5
10
15
20
25
30
35
40R_H_VZ
D_SpH_VZ_A
D_H_VZ_I
R_SpH_VZ
R_SpH_R_AW_H_A
R_H_R_I
S_SpH
S_H
Difference [%]
Parameter UnitBest Design
RDO
Best Design
TM
Difference
[%]
W_AS [°] 0,820 0,790 3,66
σ [°] 0,091 0,092 1,09
Design - Parameter Sensitivity
Results due to the target:Robustness-analysis with ALHS!
05
./ 0
6.1
1.2
01
5 –
WO
ST
20
15
Institut für Maschinenelemente, Universität Stuttgart
Technologie Transfer Initiative, Universität Stuttgart
Summary and conclusion
24
05
./ 0
6.1
1.2
01
5 –
WO
ST
20
15
Institut für Maschinenelemente, Universität Stuttgart
Technologie Transfer Initiative, Universität Stuttgart
S. K
em
mle
r; A
. F
uch
s; T
. L
eo
po
ld; B
. B
ert
sch
e
Summary and conclusion
Functional adaptive simulation strategy SIM-SMAR2T
Results of SIM-SMAR2T
General investigation of the approaches RDO and Taguchi Method
Application of both methods using the output clearance of the
adjuster unit as an example
Evaluation of the results
Construction of an universal flow chart for the application related
optimization according to RDO and/or Taguchi Method
Conclusion: Both approaches lead to the same objective
25Summary and conclusion
05
./ 0
6.1
1.2
01
5 –
WO
ST
20
15
Institut für Maschinenelemente, Universität Stuttgart
Technologie Transfer Initiative, Universität Stuttgart
S. K
em
mle
r; A
. F
uch
s; T
. L
eo
po
ld; B
. B
ert
sch
e
Summary and conclusion
Flowchart (printed in paper):
Detailed and universal approach
Provides the developer a clear
structure and decision aid
Approach for the pre-process to
design the right meta-model in
respect to:
Coupled terms
Regression-method
Interactions
Handling with sensitivities
Realization of RDO or TM
26Summary and conclusion
05
./ 0
6.1
1.2
01
5 –
WO
ST
20
15
Institut für Maschinenelemente, Universität Stuttgart
Technologie Transfer Initiative, Universität Stuttgart
S. K
em
mle
r; A
. F
uch
s; T
. L
eo
po
ld; B
. B
ert
sch
e
References
Fowlkes, 1995:
Fowlkes W.Y., Creveling C. M., “Engineering Methods for Robust
Product Design: Using Taguchi Methods in Technology and Product
Development”, Addison-Wesley, 1995
Kemmler, 2014:
Kemmler, S., et al., “Method for the development of a functional
adaptive simulation model for designing robust products”, 11th Weimar
Optimization and Stochastic Days, 2014
27
05
./ 0
6.1
1.2
01
5 –
WO
ST
20
15
Institut für Maschinenelemente, Universität Stuttgart
Technologie Transfer Initiative, Universität Stuttgart
THANK YOU FOR YOUR
ATTENTION.
Institute of Machine Components
University of Stuttgart
http://www.ima.uni-stuttgart.de
[email protected] STUTTGART
Knorr-Bremse
Systeme für Nutzfahrzeuge GmbH
http://www.knorr-bremse.com