1. A new approach to tolerance improvement through real-time selective assembly
Arne Thesen and Akachai Jantayavichit
Department of Industrial Engineering University of Wisconsin-Madison University Ave, Madison, WI 53706, U.S.A.
Arne Thesen and Akachai Jantayavichit

2. Research Objective Pins Bushings Assembly Station
To develop and evaluate efficient algorithms for tolerance improvement of assembly parts through selective assembly
Selective Assembly Process
Tol. 10-3
Tol. 10-3
Pins
Bushings
Tol. 10-4
Assembly Station
5x10- 4
Arne Thesen and Akachai Jantayavichit

3. Previous research focuses on batch process
Arne Thesen and Akachai Jantayavichit

4. Example: An artificial heart valve
Must avoid leakage
Arne Thesen and Akachai Jantayavichit

5. Arne Thesen and Akachai Jantayavichit
Previous work
1985 Boyer and A statistical selective assembly method Nasemetz (SSA) for a real time process
1990 Malmquist OMID heuristic: determine the batch size
1992 Pugh SSA for a batch process
1993 Robin and Multiple Regression modeling Mazharsolook
1996 Zhang and Set theory and probability method
Fang
Coullard et al. Matching theory
1999 Chan and Linn Balanced probability and unequal tolerance zone
1999 Thesen and Evaluate scroll compressor shells for Jantayavichit real time process
Arne Thesen and Akachai Jantayavichit

6. Arne Thesen and Akachai Jantayavichit
The production system
Arne Thesen and Akachai Jantayavichit

7. Arne Thesen and Akachai Jantayavichit
We focus on real-time applications in high-speed assembly systems (6 sec cycle times)
Arne Thesen and Akachai Jantayavichit

8. Arne Thesen and Akachai Jantayavichit
The compressor
Arne Thesen and Akachai Jantayavichit

9. Arne Thesen and Akachai Jantayavichit
Example: A SCROLL COMPRESSOR Needs close tolerances to maintain high pressure
Arne Thesen and Akachai Jantayavichit

10. Arne Thesen and Akachai Jantayavichit
A high-speed selective assembly station Note: This is presently a three-operator manual operation
Arne Thesen and Akachai Jantayavichit

11. Tolerance improvement
Worst-case gap without selective assembly is 6s
PIN BUSHING
Arne Thesen and Akachai Jantayavichit

12. Tolerance improvement
Classify components by size into tolerance classes
Worst-case gap using 2 classes and matching identical classes is 3s
Resulting system is unstable
PIN BUSHING
Arne Thesen and Akachai Jantayavichit

13. Tolerance improvement with direct matchings s s
Arne Thesen and Akachai Jantayavichit

14. Tolerance improvement Direct matching
Buffer utilization
Yield
8 tolerance classes
Arne Thesen and Akachai Jantayavichit

15. Tolerance improvement Direct matching
For reject rates less than 1%
Very large buffer capacities needed
Very high buffer utilizations expected
Arne Thesen and Akachai Jantayavichit

16. Tolerance improvement with neighbor search
Allow matching with component in neighbor class
Worst case using 8 classes is 1.5s
Arne Thesen and Akachai Jantayavichit

17. Tolerance improvement Allowing matches with neighbor class
Arne Thesen and Akachai Jantayavichit

18. Neighbor search Matching strategies
Random
Direct match first
Least likely first
Most likely first
Arne Thesen and Akachai Jantayavichit

19. Arne Thesen and Akachai Jantayavichit
Performance Analysis
Performance measure: Yield
Assuming that
All system states can be enumerated
Decisions in a given state are always made the same way
Then we can compute steady state probability for
being in each state
making any state transition
Decision rules for state space with 100,000 can be easily evaluated
Simulation will be used for large models
Arne Thesen and Akachai Jantayavichit

20. Neighbor search: Yield
Arne Thesen and Akachai Jantayavichit

21. Neighbor search: Buffer utilization
Arne Thesen and Akachai Jantayavichit

22. Arne Thesen and Akachai Jantayavichit
Neighbor search, least likely first Buffer Capacity = 48, Discard upon deadlock
Arne Thesen and Akachai Jantayavichit

23. Designing a real-time assembly station
Establish required level of tolerance reduction
From this set number of tolerance classes
Establish algorithm for selecting components
Neighborhood
Decide how to deal with deadlock
Discard
Return
Specify buffer capacity
Arne Thesen and Akachai Jantayavichit

24. Recommendations Use neighbor search with most unlikely first
Arne Thesen and Akachai Jantayavichit

25. Arne Thesen and Akachai Jantayavichit
CONCLUSION
Significant tolerance improvement is possible.
Must use neighborhood matching rule.
Results only valid for identical distributions.
Extensions to unequal distributions under way.
Arne Thesen and Akachai Jantayavichit

26. Thank you Any Questions ?
Arne Thesen and Akachai Jantayavichit