Why device, WIFI, and ISP insights are crucial to supporting remote Microsoft...
Murad Muradi - Quantum Annealing based Optimization of Robotic Movement in Manufacturing
1. Murad Muradi
04.04.2019
WORKING DRAFT.
BASELINE FOR CHARGE NOW.
BLOCKCHAIN –
ENABLING
TECHNOLOGY
FOR
AUTOMOTIVE? IoT
IAM
Blockcharge
DRM
ledgers
Smart
contract payment
transport
Proof of
ownership
BLOCKCHAIN.
BMW ACTIVITIES AND PROOF OF CONCEPTS.
LT-1 RUNDE.
DAQARI CORP.
HARD- UND SOFTWARE
QUANTUM INSPIRED OPTIMIZATION OF ROBOTIC
MOVEMENTS IN MANUFACTURING.
04.04.2019
2. Seite 2Quantum inspired optimizationof roboticprocesses in manufacturing | 04.04.2019
AGENDA.
Experimental Setup3
Results4
Use Case Introduction1
Formal Description2
3. Seite 3Quantum inspired optimizationof roboticprocesses in manufacturing | 04.04.2019
Use-Case
Introduction
4. Seite 4Quantum inspired optimizationof roboticprocesses in manufacturing | 04.04.2019
USE-CASE INTRODUCTION.
PVC SEALING PROCESS.
PVC Sealing servesto sealthe gaps of joined sheets,thereby preventing the ingress of corrosive media.
5. Seite 5Quantum inspired optimizationof roboticprocesses in manufacturing | 04.04.2019
USE-CASE INTRODUCTION.
DEFINITION OF TASK.
RoboticTask
Calculation of a production plan in which all seams are processed within a given cycle time.
b
c
1
a
d
f
g
2
e h
3
Scheduling:
Efficient allocation of tasks to available robots.
Sequencing and Motion Planning:
Minimizing the routes for all robots and creation of collision-
free robot movements.
OptimizationTasks
Goals
Increasing efficiency of existing production facilities.
Reduction of planning effort for integration of new
workloads.
Increasing flexibility in production.
Improved dynamic design of robot movements.
6. Seite 6Quantum inspired optimizationof roboticprocesses in manufacturing | 04.04.2019
Formal
Description
7. Seite 7Quantum inspired optimizationof roboticprocesses in manufacturing | 04.04.2019
FORMAL DESCRIPTION.
PROBLEM MODELLING*.
Modelling as weighted graph
𝐺 = 𝑉, 𝐸, 𝑊
𝑉 = 𝑉𝑟 ∪ 𝑖=1
𝑁
𝑉𝑖
1
, 𝑉𝑖
2
𝐸 = 𝑖=1
𝑁
𝑗=1
𝑗≠𝑖
𝑁
𝑉𝑖
1
, 𝑉𝑗
1
, 𝑉𝑖
1
, 𝑉𝑗
2
, 𝑉𝑖
2
, 𝑉𝑗
1
, 𝑉𝑖
2
, 𝑉𝑗
2
∪ 𝑖=1
𝑁
𝑉𝑟, 𝑉𝑖
1
, 𝑉𝑟, 𝑉𝑖
2
𝑊: 𝐸 → [0, ∞)
Logical variables
𝒙 = 0, 1 2𝑁2
Meaning of logical variable 𝑥𝑖,𝑡
𝑑
:
Robot should process ith task at stept in direction d
taks 1
taks 2
*) QC Ware
9. Seite 9Quantum inspired optimizationof roboticprocesses in manufacturing | 04.04.2019
Experimental
Setup
10. Seite 10Quantum inspired optimizationof roboticprocesses in manufacturing | 04.04.2019
EXPERIMENTAL SETUP.
QUANTUM ANNEALING*.
Hardware: DWave 2000Q
Annealing Cycle:
Lessthan 4tasks 10.000 cycles.
Morethan 4tasks 1.000.000 cycles.
For testing the use case the robotic tasks are defined in 2D.
Following isthe definition of thetasks along with start and end coordinate of eachtasks, along withthe velocity of robot.
Task Start End Velocity
a (287, 619) (17, 479) 94
b (595, 627) (592, 52) 58
c (488, 353) (43, 565) 68
d (450, 142) (688, 580 63
e (136, 403) (630, 170) 70
*) QC Ware
11. Seite 11Quantum inspired optimizationof roboticprocesses in manufacturing | 04.04.2019
EXPERIMENTAL SETUP.
DIGITAL ANNEALING*. 12 fixed seams
for Robot 1
12 fixed seams
for Robot 2
10variable seams
for Robot 1 and 2
24 fixed seams
10variable seams (1024 possible seam allocations)
Examined Scope:
34 seams
2 robots
1 nozzle
1 position on linear axis
*) Fujitsu
13. Seite 13Quantum inspired optimizationof roboticprocesses in manufacturing | 04.04.2019
FROM THE MANUFACTURING PROBLEM TO A QUANTUM INSPIRED SOLUTION.
1 2
Build model from
production process
Experimental implementation
3
Comparison of
the solutions
Distance Matrix
Quantum Annealing (QA) Digital Annealing (DA)Simulated Annealing (SA)
14. Seite 14Quantum inspired optimizationof roboticprocesses in manufacturing | 04.04.2019
QA Hardware is still in a development phase and requires further development for bringing concrete business value.
KEY TAKEAWAYS
• QA approach successfully foundthe optimal solution.
• Dueto hardware limitations only roundtrips with upto 5 seams could be solved.
RESULTS.
QUANTUM ANNEALING.
Performance
15. Seite 15Quantum inspired optimizationof roboticprocesses in manufacturing | 04.04.2019
RESULTS.
DIGITAL ANNEALING.
Processing Time
KEY TAKEAWAYS
• Roundtrips with upto 22 seams per Robot could be solved.
• Evaluation on 1024 different work distributions between 2 robots
• Computation time is 1-2 magnitudes faster than SA.
• Result quality is similar to SA and up to 7 % better than NN.
Seam Allocation
ProcessingTime
Simulated Annealing
Digital Annealing
Nearest Neighbor
Next Generation of Digital Annealer is capable totackle business relevant problems.
Seam Allocation
KEY TAKEAWAYS
• Notrivial solution for seam allocation.
• There are seams closer to Robot 1, but processed by Robot 2 and
vice versa.
Robot1 Robot2
Midplane
17. Seite 17Quantum inspired optimizationof roboticprocesses in manufacturing | 04.04.2019
QUANTUM COMPUTING TEAM @ BMW.
Murad Muradi
TP-322
Robotics / PVC Sealing
Oliver Wick
LT-7
Research / Scout
Stefan Benesch
FG-250
BigData
Bernhard Pflugfelder
FG-26
Big Data / AI Network
SelamWoldetsadick
FG-250
Big Data
Johannes Müller
FG-845
HPC
Marvin Erdmann
EE-541
CarSharing
Arpit Mehta
FG-250
Machine Learning
Markus Müller
LT-7
Research/DigAnnealer
UseCase
Owner
Core
Team
Khanh-Huy Tan
FG-130
IT Innovation
Thomas Hubregtsen
LT-3
Quantum Computing, AI
18. Seite 18Quantum inspired optimizationof roboticprocesses in manufacturing | 04.04.2019
RESULTS.
KEY TAKEAWAYS
• Dueto hardware limitations only roundtrips with upto 5 seams
could be solved.
• Found Solution was optimal.
Quantum Annealing
QA Hardware is still in a premature phase and requires
further development to getting a business value.
Digital Annealing
KEY TAKEAWAYS
• Roundtrips with upto 22 seams per Robot could be solved.
• Evaluation on 1024 different work distributions between 2 robots
• Computation time is 1-2 magnitudes faster than SA.
• Result quality is similar to SA and up to 7 % better than NN.
Next Generation of Digital Annealer is capabletotackle
business relevant problems.
Seam Allocation
ProcessingTime
Simulated Annealing
Digital Annealing
Nearest Neighbor