1. “VARIATION RESPONSE METHOD”
- CAE SIMULATION SUITE -
Authors: Pasquale Franciosa, Sumit Sinha, Manoj Babu, Emile Glorieux, Darek Ceglarek
4th February 2020 | Professor Lord Bhattacharya Building (NAIC), WMG | University of Warwick, Coventry, CV4 7AL
Presented by: Pasquale Franciosa
Get Connected: When CAE Simulation Meets Artificial Intelligence
WMG, University of Warwick, Coventry, UK
1
2. Agenda
PART A: Introduction
Challenges
Framework
Examples of application
PART B: Hands-on session
Software architecture
Applications:
• Virtual tooling optimization in automotive body-in-white
• Part fit up in aerospace body assembly
• Integration to “Deep Learning for Manufacturing” module
PART C: Engineering game
Let’s challenge the Simulation toolkit – stop by the PC booth at the back of the room
2
4. FRAMEWORK
MAIN APPLICATION AREA: MULTI-STAGE PRODUCTION SYSTEMS
OBJECTIVE: Diagnosing and root causes of
quality defects generated and propagated in
multi-stage systems
CHALLENGES:
(1) Generation and propagation of defects
(2) Large parameters space
(3) Multi-physics nature of quality defects
TYPICAL APPLICATIONS:
CONCEPT:
TYPICAL TOPICS:
(1) Zero-defect manufacturing
(2) Variation reduction
(3) Quality control
(4) Tooling and process optimisation
Automotive Aerospace
System
𝑌 = 𝐹(𝑋)
𝑋
Input control
parameters
𝑌
Output system
response
𝑇 Process
constraints
PART A - Introduction 4
5. Forward Propagation
𝑋
Takes as input a set of
control parameters
𝑌
Gives as output a
system response
𝑇Constraints such as
specification limits
System
𝑌 = 𝐹(𝑋)
CAE SIMULATION
- VARIATION SIMULATION METHOD -
ARTIFICIAL INTELLIGENCE
- DEEP LEARNING FOR MANUFACTURING -
FRAMEWORK
MAIN APPLICATION AREA: MULTI-STAGE PRODUCTION SYSTEMS
𝑋
Control parameters
have to be estimated
𝑌
Given the output is
known
Backward Propagation
System
𝑋 = 𝐹−1
(𝑌)
𝑇Constraints such as
specification limits
(1) To support system optimisation
(2) To improve process capability
(3) Variation reduction
(1) To support system optimisation
(2) Intelligent root cause analysis of defects
(3) Preventive control actions
PART A - Introduction 5
6. FRAMEWORK
MAIN APPLICATION AREA: MULTI-STAGE PRODUCTION SYSTEMS
PART A - Introduction
4. DL Deployment
X
𝑌 ?
N
Y
Adaptive sampling
3. DL Training1. Training Data 2. CAE Simulation
Set X and compute YReduce uncertainty Optimise DL prediction
Model training iteration: 1-st
6
7. FRAMEWORK
MAIN APPLICATION AREA: MULTI-STAGE PRODUCTION SYSTEMS
PART A - Introduction
4. DL Deployment
X
𝑌 ?
3. DL Training1. Training Data 2. CAE Simulation
N
Y
Adaptive sampling
Set X and compute YReduce uncertainty Optimise DL prediction
Model training iteration: i-th
7
8. FRAMEWORK
MAIN APPLICATION AREA: MULTI-STAGE PRODUCTION SYSTEMS
PART A - Introduction
4. DL Deployment
X
𝑌 ?
Set X and compute Y
N
Y
Adaptive sampling
3. DL Training1. Training Data 2. CAE Simulation
Reduce uncertainty Optimise DL prediction Ready to be used in field
Model training iteration: N-th Model ready to be deployed!
8
9. CHALLENGES
3. Deep Learning Model
1. Training Data
4. Model Deployment
1.(1) Model fidelity
2.(2) Model completeness
3.(3) Computational time
2. CAE Simulation
1.(1) Model fidelity
2.(2) IT deployment infrastructure
3.(3) Real-time data gathering
1.(1) System Dimensionality
2.(2) Uncertainty Quantification
3.(3) Large dataset and storage
1.(1) Architecture Selection
2.(2) System Collinearity
3.(3) Model Transferability
PART A - Introduction 9
10. CHALLENGES
3. Deep Learning Model
1. Training Data
4. Model Deployment
(1) Model fidelity
(2) Model completeness
(3) Computational time
2. CAE Simulation
1.(1) Model fidelity
2.(2) IT deployment infrastructure
3.(3) Real-time data gathering
1.(1) System Dimensionality
2.(2) Uncertainty Quantification
3.(3) Large dataset and storage
1.(1) Architecture Selection
2.(2) System Collinearity
3.(3) Model Transferability
PART A - Introduction 10
11. VARIATION RESPONSE METHOD (VRM)
Introduction
WHAT IT IS: Modular CAE simulation toolkit to
model, simulate and optimize multi-stage systems
with “real” rigid and compliant parts
WHAT IT CAN DO:
(1) To support system optimisation
(2) To improve process capability
(3) Dimensional variation reduction
KEY MODELLING CAPABILITIES:
(1) Stochastic uncertainty and quantification
(2) Variation propagation in multi-stage systems
(3) Model parametrisation and interface with AI/DL modules
PART A - Introduction 11
12. VARIATION RESPONSE METHOD (VRM)
..a bit of history…
PART A - Introduction
When: 2008
What: SVA-FEA (Statistical Variation Analysis – Finite Element Analysis)
Application: Variation simulation of single- and multi-station assembly system with compliant parts (ideal parts only)
Software platform: Matlab linked to MSC Nastran FEA processor
When: 2010
What: FEMP (Finite Element Method & Programming)
Application: Open-source FEM solver with capability to model thin and thick shell elements (applications to sheet-metal parts)
Software platform: first release in Scilab and then translated to Matlab
When: 2013
What: VRM 1.0 (Variation Response Method)
Application: Modular simulation toolkit with capability to model and optimize (stochastic optimisation) assembly system with non-ideal compliant sheet-
metal parts (single-stage assembly)
Software platform: C++ with OpenMP and Matlab
When: 2017
What: VRM 2.0 (Variation Response Method)
Application: Integration of new modules: (a) robotics (task sequencing; path planning; collision checking); (b) laser weld simulation in keyhole &
conduction mode
Software platform: C++ with OpenMP and Matlab
When: 2020
What: VRM 3.0 (Variation Response Method)
Application: Integration of new modules: (a) thermal simulation; (b) multi-stage assembly simulation; (c) interface for Deep learning
Software platform: C++ with OpenMP and Matlab
12
13. VARIATION RESPONSE METHOD (VRM)
Example of NPI process for fixture development
PART A - Introduction
Early design stage Late design stage Commissioning
Concept design FabricationDetailed design
… cascading of quality requirements: GD&T, cycle time, cost, etc.
13
14. VARIATION RESPONSE METHOD (VRM)
Example of NPI process for fixture development
PART A - Introduction
Objective: to determine optimum fixture layout by considering dimensional/geometrical quality, joint
quality, accessibility of robotic arm and cycle time requirements => needs for multi-physics
variation modelling
14
15. VARIATION RESPONSE METHOD (VRM)
Fixture development process
PART A - Introduction
Challenges:
(1)complex non-linear relationship between input process
parameters and output key performance indicators
(1)heterogeneous and coupled design models
I. Part variation (non-ideal parts)
II. Clamping
III. Welding
IV. Robot path
V. …
(2)manufacturing parts deviate from nominal CAD specs
Objective: to determine optimum fixture layout by
considering dimensional/geometrical quality, joint quality,
accessibility of robotic arm and cycle time requirements
(clamp closing)
(part placement – non-ideal parts)
(welding) robot path
Infeasible clamp
location
15
16. VARIATION RESPONSE METHOD (VRM)
Fixture development process
PART A - Introduction
“which clamp layout gives the smallest
part-to-part gap?”
…
joint clamp joint
Part form erorr(1)
Clamp layout(1)
Clamp layout(M)
How to model
stochastic
uncertainty
16
17. VARIATION RESPONSE METHOD (VRM)
Fixture development process
PART A - Introduction
“which clamp layout gives the smallest
part-to-part gap?”
…
Part form erorr(N)
Clamp layout(1)
Clamp layout(M)
joint clamp joint
?
How to model
stochastic
uncertainty
17
18. VARIATION RESPONSE METHOD (VRM)
Fixture development process
PART A - Introduction
o Product/process uncertainty (i.e., variation) impact final quality performance
o Can we optimise the quality performance for given product/process variation?
…
…
…
…
…
joint clamp joint
Part form erorr(1) Part form erorr(N)
Clamp layout(1) Clamp layout(1)
Clamp layout(M) Clamp layout(M)
How to model
stochastic
uncertainty
18
19. VARIATION RESPONSE METHOD (VRM)
Fixture development process
PART A - Introduction
How to model
stochastic
uncertainty
How do we model stochastic variations for sheet
metal parts if no scanning data available?
(1) Morphing mesh
(2) Parametric envelop
(3) Gaussian morphing mesh
… using polynomial chaos expansion
19
20. VARIATION RESPONSE METHOD (VRM)
Fixture development process
PART A - Introduction
How to model
stochastic
uncertainty
How do we model stochastic variations for sheet
metal parts if no scanning data available?
(1) Morphing mesh
(2) Parametric envelop
(3) Gaussian morphing mesh
… using polynomial chaos expansion
20
21. VARIATION RESPONSE METHOD (VRM)
Fixture development process
PART A - Introduction
Input part[2]
Input part[1]
How about if we
have a multi-stage
system?
21
22. VARIATION RESPONSE METHOD (VRM)
Fixture development process
PART A - Introduction
STAGE[1]
Place / Clamp
Fasten
Release
Input part[2]
Input part[1]
How about if we
have a multi-stage
system?
22
23. VARIATION RESPONSE METHOD (VRM)
Fixture development process
PART A - Introduction
STAGE[1]
Place / Clamp
Fasten
Release
Input sub-assembly[1]
Input part[3]
Input part[2]
Input part[1]
How about if we
have a multi-stage
system?
23
24. VARIATION RESPONSE METHOD (VRM)
Fixture development process
PART A - Introduction
STAGE[1]
Place / Clamp
Fasten
Release
Input sub-assembly[1]
Input part[3]
Input part[2]
Input part[1]
STAGE[2]
…
How about if we
have a multi-stage
system?
24
25. VARIATION RESPONSE METHOD (VRM)
Fixture development process
PART A - Introduction
STAGE[1]
Place / Clamp
Fasten
Release
Input sub-assembly[1] Output sub-assembly[2]
Input part[3]
Input part[2]
Input part[1]
STAGE[2]
…
How about if we
have a multi-stage
system?
25
26. x
y
z
Part A
Part B
Part D
Part C
VARIATION RESPONSE METHOD (VRM)
Fixture development process
PART A - Introduction 26
Deviations of Part A and D generated according to statistical input
deviations
Parts B and C assumed with initial zero deviations
27. VARIATION RESPONSE METHOD (VRM)
Fixture development process
Disp (mm)
Disp (mm)
Disp (mm)
Releasing first Stage
(A+B)
Releasing second Stage
(C+D)
Releasing third Stage
(A+B) + (C+D)
When parts are released after first and second stage, part B and D exhibit rigid rotation
When parts are positioned into the third stage, sub-assemblies (A+B) and (C+D) go in
contact with each other
Defect propagation accumulates in the third and final stage
PART A - Introduction 27
28. VARIATION RESPONSE METHOD (VRM)
Fixture development process
Disp (mm)
Single-stage configuration
Disp (mm)
Multi-stage configuration
… single-stage configuration gives un-realistic results…
Multi-stage configuration: at the final releasing phase, parts B and C are not aligned, due to deviations which occur in the
previous stations
Single-stage configuration: parts B and C are joined in their nominal position, where parts are correctly aligned => model
is conceptually wrong
vs.
PART A - Introduction 28
29. VARIATION RESPONSE METHOD (VRM)
Software development
PART A - Introduction
Modelling of part variation
modelling
Deviation
[mm]
Deviation
[mm]
Deviation
[mm]
Modelling of assembly process
29
30. VARIATION RESPONSE METHOD (VRM)
Applications
PART A - Introduction
AUTOMOTIVE CROSS MEMBER
Part A
Part B
Part C
Part D
Part E
Multi-stage assembly
Stage(1): Part A + Part B (Sub 1)
Stage(2): Sub 1 +Part C + Part D + Part E
Target: to minimise spring back
30
31. VARIATION RESPONSE METHOD (VRM)
Applications
PART A - Introduction
AUTOMOTIVE CROSS MEMBER
0
0,5
1
1,5
2
2,5
1 2 3 4 5 6 7 8
Inspection Point
MeanValue(mm)
Mean Measured Mean Simulated
Mean deviations comparison
Correlation Index: 0.7159
Point 2
Point 1
Point 7
Point 6
Point 5
Point 4
Point 8
Point 3
31
32. VARIATION RESPONSE METHOD (VRM)
Applications
PART A - Introduction
AUTOMOTIVE BONNET ASSEMBLY
Single-stage assembly
Hood + fender groups
Target: to control panel gaps and flush
32
33. VARIATION RESPONSE METHOD (VRM)
Applications
PART A - Introduction
AUTOMOTIVE BONNET ASSEMBLY
Std Dev Contour plot. Global Z direction Std Dev Contour plot. Euclidean Norm
0,1
0,14
0,18
0,22
0,26
0,3
0,34
0,38
0,42
DT1 DT3 DT5 DT7 DT9 DT11 DT13 DT15 DT17 DT19
Point Distance
StdDev(mm)
Distance Z Direction (Std Value) Distance Y Direction (Std Value)
Distance X Direction (Std Value) Profile [0.23 -0.09 0.969] (Std Value)
Gap [-0.3956 0.8573 -0.3294] (Std Value)
Inspection point
Std Dev Contour plot. Global Z direction
33
34. VARIATION RESPONSE METHOD (VRM)
Applications
PART A - Introduction
AUTOMOTIVE DOOR ASSEMBLY
Targets: to optimise clamp location to
minimise part to part gaps
34
35. VARIATION RESPONSE METHOD (VRM)
Applications
PART A - Introduction
AUTOMOTIVE DOOR ASSEMBLY
Targets: to optimise clamp location to
minimise part to part gaps
Results achieved:
Fixture design synthesis of assembly systems
Fixture design tasks’ integration at early-stage
design
Robust fixture layout optimisation with non-ideal
sheet-metal parts
Novel concept of fixture capability as a
representation of the most likely feasible design
solutions
35
36. VARIATION RESPONSE METHOD (VRM)
Applications
PART A - Introduction
AUTOMOTIVE SIDE MEMBER
Target: to optimise assembly sequence
36
37. VARIATION RESPONSE METHOD (VRM)
Applications
PART A - Introduction
AUTOMOTIVE SIDE MEMBER
Target: to optimise assembly sequence
37
38. VARIATION RESPONSE METHOD (VRM)
Applications
PART A - Introduction
STAMPING AND MATERIAL HANDLING
Target: Co-adaptive optimisation of the end-effectors’ structure with the robot motion planning to obtain the
highest productivity and to avoid excessive part deformations
38
39. VARIATION RESPONSE METHOD (VRM)
Applications
PART A - Introduction
AEROSPACE TURBINE BOX COVER
Target: Larger gaps are to be avoided to
reduce the spring-back force after the removal
of the clamps
39
40. VARIATION RESPONSE METHOD (VRM)
Applications
PART A - Introduction
(Fixture Capability map)
No. of iterations
Computational time
*(hours)
Monte Carlo 1000 ~26.5
Polynomial Chaos 30 ~1.0
*DELL Precision T7400 workstation - win 7 64bit
24 GB RAM, 2 Xeon E5420 quad-core processors
40
AEROSPACE TURBINE BOX COVER
41. VARIATION RESPONSE METHOD (VRM)
Applications
PART A - Introduction
AEROSPACE VERTICAL STABILISER
Target: to control part to part gaps
Note: Income variation of skin panel is modelled
using scanning data (cloud of points)
SKIN
SPAR + RIB sub-assembly
(internal skeleton)
41
42. VARIATION RESPONSE METHOD (VRM)
Applications
PART A - Introduction
AEROSPACE VERTICAL STABILISER
STAGE 1 Set non-ideal geometry
STAGE 2
Set rib placement
42
STAGE 3
Clamped assembly
44. VARIATION RESPONSE METHOD (VRM)
To summarise…
PART A - Introduction 44
WHAT IT IS: Modular CAE simulation toolkit to
model, simulate and optimize multi-stage systems
with “real” rigid and compliant parts
WHAT IT CAN DO:
(1) To support system optimisation
(2) To improve process capability
(3) Dimensional variation reduction
KEY MODELLING CAPABILITIES:
(1) Stochastic uncertainty and quantification
(2) Variation propagation in multi-stage systems
(3) Model parametrisation and interface with AI/DL modules
46. VARIATION RESPONSE METHOD (VRM)
Software architecture
PART B – Hands-on session
• Matlab-based toolkit compatible with
Matlab 2015a or later
• MEX-C++ routines, which takes
advantages of multi-core CPUs
capability and x64 platforms
• Parallel computing made possible
using the build-in Matlab parallel toolbox
• The software toolkit will be made
available to all attendees after the
event
http://www2.warwick.ac.uk/fac/sci/wmg/research/manufacturing/downloads
46
47. VARIATION RESPONSE METHOD (VRM)
Software architecture
Load product data
CAD
Load process
data CAM
(1) MODEL INITIALISATION
Load production
data
Set parameters Solver settings
(2) SIMULATE ASSEMBLY
(4) ANALYSIS (5) SYNTHESIS
(5.1) Calculate Regression
(5.2) Solve optimisation problem
(4.1) What-if analysis
(4.2) Plot results
(6) EXPORT COMPUTED SOLUTION
MANUALCONFIGURATION
AUTOMATICCONFIGURATION
PART B – Hands-on session 47
49. VARIATION RESPONSE METHOD (VRM)
Case-in-point
PART B – Hands-on session
AIMS:
• Study deformation patterns due to locator movement during the positioning, clamping, fastening and release
steps for assembly of door inner and hinge reinforcement
• Perform Deep Learning using 3D CNN on the deformation pattern data to estimate process variations
causing these deformation patterns
Stage(1): place Stage(2): clamp Stage(3): fasten Stage(4): release
49
50. VARIATION RESPONSE METHOD (VRM)
Case-in-point
PART B – Hands-on session
Process
Parameter
Description Unit Training
Range
Deployment
Range
X(1)
Rotation around
the pinhole
degree [-1,1] [-2, 2]
X(2)
Pin hole
displacement in x
mm [-1,1] [-4, 4]
X(3)
Pin hole
displacement in z
mm [-1,1] [-4, 4]
X(4)
Clamp 1
displacement in y
mm [-2,2] [-4, 4]
X(5)
Clamp 2
displacement in y
mm [-2,2] [-4, 4]
X(6)
Clamp 3
displacement in y
mm [-2,2] [-4, 4]
50
51. VARIATION RESPONSE METHOD (VRM)
Case-in-point
PART B – Hands-on session
FULL CAE SIMULATION OF RLW CELL
• Robot path planning
• Collision checking
• Fixture clamp layout optimisation
• Welding parameters selection
• Thermal simulation
Interface with Unity3D (VR engine)
51
52. VARIATION RESPONSE METHOD (VRM)
Take home messages
52
WHAT IT CAN DO:
(1) To support system optimisation
(2) To improve process capability
(3) Dimensional variation reduction
KEY MODELLING CAPABILITIES:
(1) Stochastic uncertainty and quantification
(2) Variation propagation in multi-stage systems
(3) “Virtual data generator” to train deep learning models
53. MAIN SOFTWARE CONTRIBUTORS AND DEVELOPERS:
Pasquale Franciosa (Warwick University), Abhishek Das (Warwick University), Manoj Babu (Warwick University), Emile Glorieux (Warwick University), Sumit Sinha
(Warwick University), Darek Ceglarek (Warwick University), Avishek Pal (Microsoft), Gerardo Beruvides (Hitachi), Vahid Shahi (University of Tehran), Ferdinando
Vitolo (University of Naples Federico II), Salvatore Gerbino (University of Campania), Achille Pesce (FCA), Chris Esposito (University of Naples Federico II)
53
Thanks for your attention!
Any questions and/or comments?
Dr. Pasquale Franciosa
Prof. Darek Ceglarek
WMG, University of Warwick, UK
WMG, University of Warwick, UK
Email: p.franciosa@warwick.ac.uk / d.j.ceglarek@warwick.ac.uk
M: +44(0) 7440022523 / 44(0) 7824540721
T: +44(0)24765 73422 / 44(0) 24765 72681
Acknowledgements