Servitization in the
Capital Goods Industry
Geert-Jan van Houtum
Prof. of Maintenance, Reliability, and Quality
Eindhoven University of Technology
Key words:
System availability
Size capital goods industry Netherlands
Import:
117 billion Euro
Export:
Size capital goods industry Netherlands
Import:
117 billion Euro
Source:
Statistisch Jaarboek, 2009
Export:
117 billion Euro
Needs and
requirements
Design
Production Exploitation
Disposal
Big influence on System
Availability and TCO
One may look at:
β’ Modular design
β’ More reliable components
Life Cycle
β’
The maximum
system availability is
determined in the
design phase
β’
A large portion of
the TCO is made in
Long term trends
β’
Maintenance is outsourced to third party or OEM (pooling
resources, pooling data, remote monitoring)
β’
More extreme: One sells function plus availability
β’
Feedback to design (better systems, higher sustainability)
β’
Maintenance of complex systems becomes too complicated
for users themselves
β’
Users require higher system availabilities (less downtime)
Challenges for the OEM\third party?
Challenges for the OEM\third party?
β’
Definition of the service portfolio
Service supply chain: Example
Supply
spare parts
Central
Stockpoint
Local
Stockpoint
Customers with
contracts
Reg. repl.:
1-2 weeks
Emergency Shipm.:
1-2 days
Reg. repl.:
1-2 weeks
Local
Stockpoint
Lateral
Shipments:
A few hours
Customers with
contracts
Direct
sales
Challenges for the OEM\third party?
β’
Definition of the service portfolio
β’
Design of the service supply chain
β’
Design of the service processes
β’
Pricing
β’
Organization
β’
New business models
PURPOSE OF THIS TALK
Showing how we can contribute
via OM research
CONTENTS
1.
Introduction
2.
Creation of differentiation while keeping the
portfolio effect
3.
Effect of product design: Redundancy decision
4.
New opportunity: Remote monitoring data
2. Creation of differentiation while
keeping the portfolio effect
Spare parts network for high-tech
equipment
HUB
Local Warehouse
4hr response area
2hr response area
Legend
customer (region)
Spare parts network for high-tech
equipment
HUB
Local Warehouse
4hr response area
2hr response area
Legend
customer (region)
Spare parts network
Key features:
Lateral transshipments
Multiple customer classes
β’
Central Warehouse (CW): Infinite stock
β’
Multiple Local Warehouses (LWβs)
β’
Poisson demand processes
β’
Cost factors:
β’ Cost to fulfill a demand at point x from LW i or from the CW
β’ Unit replenishment costs
β’ Penalty costs for violating the maximum response time
constraints
Planning problems
Tactical planning problem:
Decision on
base stock levels
:
Given in this study
Operational planning problem:
We consider two
allocation rules
ο
Static
:
Common in practice
ο
Dynamic
:
Static Allocation (SA-rule):
β’
Fulfill demand from nearest LW with positive on-hand stock
β’
But, fulfill from CW if cheaper
β’
Easy to compute, easy to execute
(Markov model)
β’
Weak differentiation between customer classes
Dynamic Allocation (DA-rule):
β’
Estimate near-future effect when fulfilling from one of the
LWβs with positive on-hand stock
(Use of Appr. Dyn. Progr.)
β’
Select LW or CW with lowest βdirect + near-future costsβ
β’
Requires computer support to be applied
The two allocation rules
Large instances:
ο§
Relative savings of DA-rule compared to
SA-rule:
7.9%
Conclusion:
β’
Savings relative to the static rule are significant
β’
Dynamic rule gives a better way to differentiate
between customer classes
β’
You get implicitly a kind of dynamic, reserved stock levels for
high-priority customers
3. Effect of product design:
Redundancy decision
Setting
Production site
Spare parts
stock
Regular
replenishments
Emergency
supply procedure
(in case of stockout)
Model (cont.)
Per machine:
β’ Multiple critical components
β’ Serial structure
Component
i
Component
2
Component
2
Component
1
Component
m
Component
m
Model (cont.)
Three possible policies per component
1. No redundancy
2. No redundancy, apply emergency supply procedure
when on-hand stock drops to 1
3. Redundancy
Optimization problem
Min. TCO
Approach
β’
Generation of the
efficient frontier
for TCO
and system availability (via Lagr. Rel.)
β’
One curve for case with policy 2
β’
One curve for case without policy 2
β’
One gets an order for which components
Analysis per component
Policy 1
Policy 2
Policy 3
ο¬
Costs
Efficient frontier
With policy 2
Without
policy 2
Efficient frontier
With policy 2
included
Without
policy 2
Notice:
The optimal design depends strongly
on the required availability
So:
Multiple customer classes => multiple
designs
4. New opportunity:
Remote monitoring data
Monitoring data
β’
Condition data:
parameters
which are
directly or
indirectly related
with the health
state of Module
X
β’
Failure data:
failure time
Sample Data: Collected at central level,
for one critical unit
MACHINE
NUMBER
TIME
STAMP
VALUE
MACHINE
TYPE
SITE
ID
CUSTOMER
CONTINENT
CUSTOMER
COUNTRY
CUSTOMER
NUMBER
PARAM
ID
M1297
17-Dec-09
-8.856
T0010
1288
Asia
South Korea
188
3756
M2572
22-Oct-09
-8.9597
T0005
665
Asia
Singapore
2046
990
M2488
30-Jul-09
-3.9977
T0083
755
Other
Other
OT01
981
M0822
14-Jul-09
-4.0141
T0016
1284
Asia
South Korea
188
960
M1621
08-May-09
-3.8854
T0010
1294
Asia
South Korea
1146
957
M1647
23-Oct-09
-3.9167
T0001
277
North
America
USA
196
966
M0003
21-Jul-09
-3.873
T0010
1291
Asia
South Korea
188
990
M0004
21-Feb-09
-3.8264
T0010
1291
Asia
South Korea
188
966
M2862
27-Aug-09
-3.7398
T0004
629
Asia
Taiwan
222
993
M2631
06-Jan-09
-8.551
T0004
801
Europe
France
192
972
M1141
10-Aug-09
-6.8885
T0011
1290
Asia
South Korea
1146
966
M3241
22-Apr-09
-8.551
T0010
629
Asia
Taiwan
222
963
M0051
05-Sep-09
-8.9597
T0008
1178
Asia
Taiwan
386
996
M1171
28-Feb-09
-3.9977
T0006
629
Asia
Taiwan
222
987
M1171
12-Aug-09
-6.8885
T0006
629
Asia
Taiwan
222
990
Imperfect warnings
Demand signals produced by the prediction model are
imperfect
:
β’
Prediction model can produce false signals (
false
positives
)
β’
Exact time of the failure is uncertain
β’
Prediction model may also produce
false negatives
β Sudden, unpredicted
failures
which cannot be
detected in advance by the monitoring system
Research topic
Value of the imperfect warnings
for
spare parts supply
Remark: We do
not
look at preventive replacements.
Setting:
β’ Single stockpoint (a
local warehouse
, gets
replenishments from a central warehouse)
β’ Single item
β’ Imperfect warnings =
Imperfect Advance Demand
Information (ADI)
PAGE 32 / School of Industrial 10/15/2014 Engineering
Imperfect ADI
t
ο΄
[
]
ο΄
l
ο΄
u
L
Demand
Supply
p
p
:
probability that a
signal will ever
become a demand
realization =
reliability
(false positives)
[
ο΄
l
,
ο΄
u
] :
prediction
interval
for the
demand lead time
(timing)
q
:
ratio of predicted
demand to total
demand
=
sensitivity
demand lead time
supply lead time
Approach
3 scenarios:
1.
Optimal cost
without ADI
(benchmark)
2.
Optimal cost
with imperfect ADI
, but
no returns
allowed
3.
Optimal cost
with imperfect ADI
, and
returns
allowed
Case study at an OEM
β’
4 parts that OEM supplies its customers all over the world.
β’
p
,
q
, [
ο΄
l
,
ο΄
u
]
: obtained from prediction model in use
β’
c
em
: transportation cost + high downtime cost
c
em
= 75000 Euro
β’
c
r
: transportation cost + pipeline holding cost
β’
L
: 2 weeks
Case study at an OEM
β’
Part P (
p
an
d
q
are low
and information is timely)
β’ value of information is high when returning excess inventory is allowed.
β’ benefit of returns
β’
Policy with ADI and no returns: Local warehouse carries no stock + a spare
part is shipped to the local warehouse only if a warning is issued
β’
Part T (
p
and
q
are high and information is timely),
β’ value of information is high even for without return case.
Part
(β¬/unit/week)
h
[
π
π, π
π]
(week)
(unit/week)
π
π
π
π
π(β¬/week)
π
π΅ππ¨π«π°π
π¨π«π°π΅ππΉπππππ(β¬/week)
(β¬/week)
π
π¨π«π°π·πͺπΉ
π¨π«π°π΅ππΉππππππ·πͺπΉ
π¨π«π°P
2720
[2,8]
0.0188
0.42
0.44 5500 1406.01
1399.68
956.06
0.45%
32.00%
T
112
[8,16]
0.0600
0.90
0.90
325
248.13
139.78
131.80
43.67%
46.88%
X
152
[0,4]
0.0019
0.45
0.43
400
145.29
138.83
134.37
4.45%
7.52%
W
646
[0,1]
0.0036
0.90
0.50 1400 273.28
273.28
260.00
0.00%
4.86%
Case study at an OEM
β’
Parts X and W (
L
>
ο΄
l
),
β’ value of information is low even when return is allowed =>
negative impact
of timing of information on the value of information
Part
(β¬/unit/week)
h
[
π
π, π
π]
(week)
(unit/week)
π
π
π
π
π(β¬/week)
π
π΅ππ¨π«π°π
π¨π«π°π΅ππΉπππππ(β¬/week)
(β¬/week)
π
π¨π«π°π·πͺπΉ
π¨π«π°π΅ππΉππππππ·πͺπΉ
π¨π«π°P
2720
[2,8]
0.0188
0.42
0.44 5500 1406.01
1399.68
956.06
0.45%
32.00%
T
112
[8,16]
0.0600
0.90
0.90
325
248.13
139.78
131.80
43.67%
46.88%
X
152
[0,4]
0.0019
0.45
0.43
400
145.29
138.83
134.37
4.45%
7.52%
W
646
[0,1]
0.0036
0.90
0.50 1400 273.28
273.28
260.00
0.00%
4.86%
Topic:
Servitization in the capital goods industry
Purpose of this talk:
Showing how can we contribute with OM research
1.
Creation of differentiation while keeping the portfolio
effect
2.
Effect of product design: Redundancy decision
3.
New opportunity: Remote monitoring data
/ School of Industrial Engineering 10/15/2014 PAGE 40
β’
Design of service networks and processes
β’
Creation of differentiation while keeping the portfolio effect
β’
Location strategy
β’
For spare parts, service engineers, tools, back office,β¦
β’
Where to decompose: For the offered services and
between service provider and user
β’
Effect of product design decisions in general
β’
Redundancy decision
,
choice of components/suppliers,β¦
β’
Commonality across multiple machine types
β’
Level of modularity
β’
How to deal with modifications
β’
Design per type of service contract?
β’
Exploiting new technologies in general
β’
Remote monitoring data
β’
Use of 3D printers
β’
Sharing of resources/data by multiple companies
β’
Service costs and pricing
β’
Sustainability issues:
β’
Re-use of systems and components
β’
CO2 and other emissions
β’
β¦