1989. Renault 25. 120,000. 1998. Peugeot 306. 170,000 ... f d em an d. Peugeot
306. Peugeot 405. Toyota Echo ... Dashboard harness. Left interior AR. harness.
Mass Customization
Outline Introduction What is data mining? Mass customization Innovation and mass customization ISL projects Conclusion
Andrew Kusiak Industrial Engineering Program 2139 Seamans Center The University of Iowa Iowa City, Iowa 52242 - 1527
[email protected] http://www.icaen.uiowa.edu/~ankusiak Tel: 319 - 335-5934 Fax: 319 - 335-5669
Why Mass Customization?
History of Product Diversity Year
Car Model
No. of Models
1908
Ford T
1
1963
Renault 4
11
1971 1982
Renault 16 Renault 18
6,000 60,000
1989
Renault 25
120,000
1998
Peugeot 306
170,000
1
Product Demand vs. Diversity
Toyota Echo
100 90 80
% of demand
70 60
Peugeot 306 Peugeot 405
50 40 30 20 10 0 0
10
20
30
40
50
60
70
80
90
100
% of diversity
Industrial Experience (1)
Industrial Experience (2)
Engine CPE box
Fuse box UCE Air conditioning
A.B.S. harness Und. body shell A.B.S. harness Lateral opening harness Opening harness Front harness Engine harness Dashboard harness Left interior AR. harness Right interior AR. harness
Engine box
ABS box
2
Traditional Production Strategies
Make-to-stock: High inventory cost Short lead time Low pressure on process set-up reduction
Make-to-order: Low inventory cost Long lead time High pressure on process set-up reduction
Coping with Product Diversity Component standardization Modular design (within and across product families) Delayed product differentiation Product complexity management
Modern Production Strategy
Configure-to-order: Assemble-to-order: Manufacturing and service industry Pick-to-order (Components)
Configure-to-order Form standard configurations
Complexity Management
Data mining Optimization
(Standard configurations)
1 tractor type (Mass production)
I want my own tractor (One of a kind)
3
Package Development (1)
Package Development (2)
Sales Records for a Simple Tractor Data Mining
Package Development (3)
Package Development (4)
Transformation
4
What is Inside Product Configurator? (1)
What is Inside Product Configurator? (2) Data mining + Optimization Min dij y ij + ci x i i
i j
xi = 1 i S
for k = 1,… , q
k
xi
What is Inside Product Configurator? (3) More algorithms Random Design The selection process does not consider Information about customers’ behavior
y ij
• Product upgrades • Product downgrades
for i, j = 1,… , m
• Customer coverage
x i = 0, 1
for j = 1,… , m
y ij = 0, 1
for i, j = 1,… , m
• Costs
Partial Output (1)
Pattern-based Design
Componentheuristic Design
PatternHeuristic Design
Extracted rules from customers’ data and three assembly or marketing constraints
Demand generated from the components’ demand data and three assembly or marketing constraints using the PME
Demand generated with the extracted rules and three assembly or marketing constraints using the PME
5
Partial Output (2)
Future Developments (1) Product selected (and package formed) determined by: • Customer characteristics (Age, income level, …) • Environment characteristics (Interest rate, weather, …)
3000 2500
C1
2000
C2
# of Sales 1500
C3 C4
1000
C5 C7
500 C4
C6 C7
9065
9080
Cluster #
C1 9070
9175
9045
9225
9085
0
Contextual knowledge
Future Developments (2)
Future Developments (3)
Design process
Going beyond a single product type: • Standard components • Modules • Assembly-to-order
Manufacturing Process
Context
Component level
6
Future Developments (4)
Future Developments (5)
Synchronous delivery
Pre-assembly Assemble-toorder
Distant location
Nearby location Automotive supplier
How to solve the mass customization problem?
Car assembly plant
Car manufacturer
Association Rule Algorithm: Example Required Option
Engine
Cab
Front Axle
Wheels
Attachment 1
Attachment 2
Attachment 3
Customer 1
E1
C1
F2
W3
1
0
0
Customer 2
E2
C2
F2
W3
0
1
0
Customer 3
E3
C3
F2
W3
1
1
1
Customer 4
E4
C3
F1
W1
1
0
0
Customer 5
E2
C2
F2
W2
1
0
0
Customer 6
E2
C2
F1
W1
0
1
0
Customer 7
E3
C2
F1
W1
0
1
0
Customer 8
E1
C1
F3
W2
1
1
1
Customer 9
E1
C1
F4
W2
1
0
1
Customer 10
E1
C1
F3
W2
1
0
1
Data mining Association rule algorithm k-means algorithm Decision tree algorithm Clustering algorithm (see process decomposition)
Optional Attachment (Option)
7
Required Option
Association Rules Engine
Rule 1. Cab = C1 ==> Engine = E1; Support = 4/10, Confidence = 4/4 Rule 2. Engine = E2 ==> Cab = C2; Support = 3/10, Confidence = 3/3 Rule 3. Wheel = W1 ==> Front_Axle = F1; Support = 3/10, Confidence = 3/3 Rule 4. Wheel = W3 ==> Front_Axle = F2; Support = 3/10, Confidence = 3/3 Rule 5. Cab = C1 Wheel = W2 ==> Engine = E1; Support = 3/10, Confidence = 3/3
Rule Interpretation Rule 1. Cab = C1 ==> Engine = E1; Support = 4/10, Confidence = 4/4 Rule 1 suggests the Cab C1 and Engine E1 form a subassembly (E1, C1)
Wheels
Attachment 2
Attachment 3
E1
C1
F2
W3
1
0
0
Customer 2
E2
C2
F2
W3
0
1
0
Customer 3
E3
C3
F2
W3
1
1
1
Customer 4
E4
C3
F1
W1
1
0
0
Customer 5
E2
C2
F2
W2
1
0
0
Customer 6
E2
C2
F1
W1
0
1
0
Customer 7
E3
C2
F1
W1
0
1
0
Customer 8
E1
C1
F3
W2
1
1
1
Customer 9
E1
C1
F4
W2
1
0
1
Customer 10
E1
C1
F3
W2
1
0
1
Rule 1. Cab = C1 ==> Engine = E1; Support = 4/10, Confidence = 4/4
Clustering: Configuration SoldFeature Matrix F1 1
Configuration sold
Rule 2. Engine = E2 ==> Cab = C2; Support = 3/10, Confidence = 3/3 Rule 2 suggests Engine E2 should be assembled with Cab C2, (E2, C2)
Front Axle
Customer 1
Cab
Optional Attachment (Option) Attachment 1
F2 1
2 3
F6
1
1
1 1
5
7
Product feature F4 F5
1
4
6
F3
1 1
1 1 1 1
8
Clustered Matrix F2
F1
Solution
Product feature F4 F3 F6
F2
1 2
F1
F5
1
5
1
1
1 1
1
1 1 1
7
4
1
7
1
6
1
6
F6
1
Product feature
2
3
F5
1
1
4
1 1 1
1 1 1
Configuration sold
Configuration sold
1
F4
1
3
5
1
F3
1
F2
F4
1
1
1
5
1
1
F1
F5
3
1
1
6
1
1
2
1
4
1
F6
1
7
Cluster identification algorithm
1
min
Notation m = number of objects (configs sold) n = number of features p = number of clusters (desired configs) 1 if feature i belongs to xij = { cluster j 0 otherwise dij = distance between objects i and j
F3
n
n
dij xij
i =1 j =1
subject to: n
xij = 1 for all i = 1, ..., n j =1 n
xjj = p j =1
xij