Manufacturing System

Flow Analysis
Ronald G. Askin
Systems & Industrial Engineering
The University of Arizona
Tucson, AZ 85721
[email protected]

October 12, 2005

How Many IEs Does It Take to
Change a Light Bulb?
 n?
• One to Work Sample to Detect Burned out Bulbs
• One to Flowchart the Process
• One to Schedule the Maintenance
• One to Supervise the Maintenance Task
• One to Implement a Process Improvement Plan/Kaizen
Event
• One to Determine Optimal Lumens for Replacement Bulb
• One to do an Economic Analysis of Buying Longer Life
Bulbs
• …
 Overview of Session
• The Modern (Lean) Factory
• WIP vs. Flowtime & Throughput (Little’s Law)
• Transfer Batches vs. Process Batches (Lot-streaming)
• Cross-Training (Balancing and Buckets)
• Performance Evaluation – Open & Closed Cells
1. Factory Flow Thru Cell System

Gears Chassis Assembly

Shafts Cards Frame

 Flow in a Cell

J. T. Black, Design of the Factory with a Future, 1991

 Cell Independence (Burbidge)
• Dedicated Team of (Compatible) Workers
• Dedicated Set of Machines
• Specified Set of Parts/Products
• Dedicated Space for Operations
• Common Goal and Evaluation
• Independence of Success
• Ideally 7-10 Members
 2. Little’s Law: Defining Rule for Flow

L = λW
(N = XT)

WIP = Prod. Rate x Flow Time

 Theoretical Profile!

Capacity
Deterministic
Production

Probabilistic (Exponential)
N=XT

WIP
 Empirical Profile
Little's Law and Chaos

12

10
Remember
N = XT
8
Throughput

6
Deterministic

Exponential
4
Empirical

0
0 10 20 30 40 50 60 70 80 90
WIP

10 stages, µ = 1
 Questions?
• What happens when we release jobs to a
busy shop floor?
• What happens when we reduce variability?
 Typical Scenario:
High Utilization, So Jobs are Late,
Therefore Release More Jobs Early

L=λW (or N=XT)

1. λ high implies ∆λ small;
2. Since L increases, W increases;
3. As W (lead time) increases, tempted to release
jobs even earlier
4. Congestion and interference reduce throughput
 Reducing Variability

General Arrivals (λ) and Service (S)

E (ThroughputTime) = E (Wq ) + E ( S )
ρ (1 + C
2 2
s ) ⋅ (C 2
a +ρ C
2 2
s )
E (Wq ) ≈
1 + ρ 2 Cs2 [ 2λ (1 − ρ )]
ρ = λ ⋅ E (S )
(ρ = X ÷ Capacity)
Question: How Far Is the Blue (Random) Line
from the Purple (Deterministic) Line?
• ρ = 0.8,
• Exponential Arrivals vs. Fixed Interarrivals
• Random Service vs. Standardized Service
What happens if we release jobs at fixed intervals?
What happens with reliable processes & standard tasks?
 3. Transfer vs. Process Batches

• Lot-Streaming – Dividing the process batch

into multiple transfer batches for concurrent
processing at successive stages
 Simple Illustration
Machine

2
• Three stages 3

• Batch size = 20 20 80 120 Time

• Unit proc. times = 1, 3, 2 a. One Transfer Batch

Machine

• No setup 1

10 40 70 90 Time

b. Two Transfer Batches

Machine
1

3 ...

014 20 61 63 Time

c. Single Unit Transfer Batches

 MH vs Thruput Time Tradeoff
MH Loads vs. Cycle Time

25

20

15

MH Loads

10

0
0 20 40 60 80 100 120 140
Cycle Time
 Basic Rules (L Sublots, Q units)

1. Consistent, equal sublots good (not optimal)

(p2 qi = p1 qi+1 is optimal for adjacent WSs)

2. Decreasing marginal benefit:

2 sublots 50% of max gain

Q
T = Q ⋅ pb + ⋅ pi
L i ≠b
3. Protect bottleneck (avoid sublot setup loss)
 4. Cross-Training
• Ensure Redundancy
• Consider Job Enrichment as Motivator
• Task Frequency Sufficient for Proficiency
• Lead Experts for Each Task
• Cover all Responsibilities
• Pay per Skill Breadth and Depth
• Worker Flexibility vs. WIP Safety Stock
a. Dynamic Rebalancing
1
4 min 3 min

6 min

8 min 3 min

a. Two Workers Total Time = 24

1
4 min 3 min

6 min

8 min 3 min

b. Three Workers
Part Flow
Worker Flow (Orbit)
Workstation
 b. Bucket Brigades (TSS) & Variants

• BB
Assumes Task Continuity
Ordered Workers
Slowest to Fastest
Effective in Picking
Buffers can be added
•Champion Strategy
(For low machine ρ)
•Leapfrog Strategy
(Less worker movement)
 5. Performance Evaluation
N=XT

• Find X & T given N & Capacity

• Find T and needed N for desired X given Capacity
• Find T, X Tradeoff
 Open System (Receive and Release)

Random
 Basic Poisson Process Estimate

1. Compute Effective Arrival 2. Evaluate Each

Rates at Workstations Workstation
(M/M/1)
m
λ 'j = λ j + λk' ⋅ pkj ρ
P(0) = 1-ρ
k =1

5/day 4 L = ρ/(1-ρ
ρ)
6
(A)
4 λ
W = L/λ
5
2

5
6/day (B)
 System & Product Measures
3. Aggregate Across
Workstations

m
W= v j ⋅W j
j =1

WB= W + .67W +W
 Closed System (CONWIP)
External Demand
 Basic Performance Evaluation - Closed
Consider a Closed System with N Jobs:
X = Production rate, T = Throughput time

M
C= cj Total Servers or Max Active Jobs
j =1

M
P= tj Total Job Processing time
j =1

min(C , N )
T ≥ P so N = XT → X ≤
P
 Performance Evaluation Extension

• Assume WIP Evenly Spread Out

N −1
1+ ⋅P , Exponential Processing Time
M
T=
N
⋅P , Constant, Synchronous Processing with N ≥ M
M

As Always, N=XT
Very Optimistic Model! No Starvation when N ≥ M
 References and Extensions
1. Askin, R. & J. Goldberg, Design and
Analysis of Lean Production
Systems, Wiley& Sons, 2002
2. Askin, R. & C. Standridge, Modeling
and Analysis of Manufacturing
Systems, Wiley & Sons, 1993
3. Black, J. T., Design of the Factory
with a Future, McGraw Hill,
1991
4. Harmon, R & L. Peterson,
Reinventing the Factory, Free
Press, 1989
5. Hopp, W. and M. Spearman,
Factory Physics, McGraw Hill,
2000.