Post on 24-Sep-2020
3/9/2016
1
Complex Systems Analytics: a Promising Enabler for Sustainable Design and Manufacturing
Harrison M. Kimhmkim@illinois.edu
Associate ProfessorDepartment of Industrial & Enterprise Systems EngineeringUniversity of Illinois at Urbana‐Champaign
Overview
2
User‐Generated Contents Product Life CycleFlu Spread Pattern
Predictive Systems Design Analytics
100
50
02008 2009 2010 2011 2012
(Present)
FeatureCount
Time
CameraAppsMP3
Prediction
3/9/2016
2
Let’s envision ….
3
Manufacturers can predict when a product will reach its end of life and the condition of the returned product based on the product attributes and customer demographic data at the moment a consumer purchases a product.
Is it better to keep my phone as long as it lasts? “I am a green farmer. I would like to keep my harvester as long as I can.”
Let’s envision ….
4
Even at the early stages of product design, the manufacturer can predict which component will be reused, recycled, remanufactured, or replaced with a new component; and, in turn, modularize platforms (i.e., better design) for easy, profitable end‐of‐life recovery operations.
3/9/2016
3
Let’s envision ….
5
Manufacturers can design a variety of products that are targeted for the new product market, as well as the remanufactured product market in a simultaneous manner for a better market coverage and higher profit.
Research Contributions – Predictive Design and Analytics
6
Wind Farm Layout Design
Energy DistributionHybrid, Renewable Energy Generation
Green Profit DesignMultidisciplinary Design Optimization – Analytical Target Cascading (ATC)
Trend Mining Design
Kim, et al. 2003 Trans. ASME
Tucker & Kim 2008, 2009 Trans. ASMEMa & Kim 2014 Trans. ASME
Kwak & Kim 2010, 2012 Trans. ASMEKwak, Ma & Kim 2014 J Cleaner Prod.
Lu & Kim 2010 Trans. ASME Lu & Kim 2014 Eng. Opt. J.
Kannan, Shanbhag & Kim 2012 Opt. Mtd. & Software
3/9/2016
4
How does CDC detect influenza epidemic?
7
Visit
Patients with Influenza‐Like Illness (ILI) visit physicians.
Report
Physician visits are reported to CDC on regular basis.
CDC analyzes the collected data and publishes a report on ILI.
It takes _________ to detect influenza epidemics.It takes _________ to detect influenza epidemics.one week
How does Google detect influenza epidemics?
8
[Source] Ginsberg et al., 2009, “Detecting influenza epidemics using search engine query data,” Nature 457(19).
Search Submit
Can estimate current flu activity around the world in near real‐time. Can estimate current flu activity around the world in near real‐time.
For product design, can we utilize user‐generated data like this?For product design, can we utilize user‐generated data like this?
3/9/2016
5
Are these waste (or opportunities)?
9
Management of Used and End‐Of‐Life Electronics in 2009 (Source: EPA)
Ready for End‐of‐Life Management(million of units)
Disposed(million of units)
Collected for Recycling(million of units)
Rate of Collection for Recycling(by weight)
Computers 47.4 29.4 18 38%
Televisions 27.2 22.7 4.6 17%
Mobile Devices 141 129 11.7 8%
A product lives a life and causes environmental issues throughout the life (especially in usage phase).
10
Material Extraction
ManufacturingSales &
Customer UseProduct Design
Reverse Logistics
End‐of‐Life
Manufacturing Maintenance Usage (Fuel)Usage (Emission) End-of-Life0
50000
100000
150000
200000
250000
300000
10312.11(2.19%)
258182.5(54.78%)
94364.36(20.02%)
29838.24(6.33%)
GW
P 1
00
a [
kg C
O2
-eq
]
78637.09(16.68%)
• Natural resource depletion• Climate change• Toxic chemical releases• Waste and land management
⁞
3/9/2016
6
For example, a machine generates…
Kwak et al. (2012) “Life Cycle Assessment of Complex Heavy Duty Off-Road Equipment,” ASME ISFA.
Global Warming Potential (kg CO2e)
11,800x 9.4x 0.5x824 Tons CO2e
Manufacturing Maintenance Usage End-of-Life Total0
20
40
60
80
100E
I-99
sco
re [
%]
Fossil fuels Minerals Land use Acidification Ecotoxicity Ozone layer Radiation Climate change Resp. inorganics Resp. organics Carcinogens
11
A major obstacle is that there exists a time gap between design and end‐of‐life.
12
Age Distribution of EOL Products
Kwak, M., Behdad, S., Zhao, Y., Kim, H. M. and Thurston, D. "E‐waste Stream Analysis and Design Implications,“ Journal of Mechanical Design, Vol. 133, No. 10, 2011.
18 months
4 years
14 years
20 years (20,000 hrs)
Remember one week delay for flu case?
3/9/2016
7
A major obstacle is that there exists a time gap between design and end‐of‐life.
13
Kwak, M., Kim, H. M. and Thurston, D. "Formulating Second‐Hand Market Value as a Function of Product Specifications, Age, and Condition," Journal of Mechanical Design, 2011.
Market Value of EOL Products
Can we make recovery profitable?How to overcome the obsolescence ?Can we make recovery profitable?
How to overcome the obsolescence ?
Product Specification Trends (Laptop)
Green profit of reman changes over time!
A model for assessing the time‐varying advantages of remanufacturing can
• Compare the remanufactured and brand‐new versions of a product,• Clarify how the nature of the product influences the time‐varying value,• Provide a multi‐dimensional assessment tool (cost, impact, and net profit).
1 2 3 4 5 6 7 8 9 10-10
0
10
20
30
40
50
60
70
80
Tot
al S
avin
g O
ver
Bra
nd-N
ew (
%)
Product Age (year)
Manufacturing Cost Environmental Impact Net Profit
Physical DeteriorationTechnological Obsolescence
Val
ue o
f Rem
anuf
actu
ring
Kwak, M. and Kim, H. M., "Assessing Time‐Varying Advantages of Remanufacturing: A Model for Products with Physical and Technological Obsolescence," ICED15, Milan, Italy, 2015.
14
3/9/2016
8
15
For original equipment manufacturers (OEMs), producing both new and remanufactured products can be an effective strategy to achieve “green profit” (i.e., profits generated by an environmentally sustainable business).
Reman is composed of three sequential activities.
16
There is strong interdependence among the activities.
3/9/2016
9
Let’s see how these can be integrated for analytics.
17
Design Analytics
Life Cycle Design
Manufacturing Maintenance Usage (Fuel)Usage (Emission) End-of-Life0
50000
100000
150000
200000
250000
300000
10312.11(2.19%)
258182.5(54.78%)
94364.36(20.02%)
29838.24(6.33%)
GW
P 1
00a
[kg
CO
2-eq
]
78637.09(16.68%)
Predictive Design Analytics
Amount of diesel fuel (l)
Available historical data timepresent future
Build usage model Predict usage
real value
constant rate
PUMS
(e.g., 7‐year data) (e.g., 9.5‐year data)
User‐generated contents and advanced telematics help establish the link from design to predictive analytics.
18
Type Doors Cylinders HP RPM MPG Price Income Age Gender Choice
Standard 2 4 288 4900 31 45400 46816 40 Female Toyota
Standard 4 4 262 5100 30 41315 66590 54 Male Chevrolet
Standard 4 4 207 5000 35 37028 97504 39 Male Honda
Customer Choice & Review
[Source] http://deere.com
Customer Usage Pattern
3/9/2016
10
Mining customer trends
Tucker, C. and Kim, H. M. "Trend Mining for Predictive Product Design,” Journal of Mechanical Design, Vol. 133, No. 11, 2011.Tucker, C. and Kim, H.M., "Predicting Emerging Product Design Trend by Mining Publicly Available Customer Review Data,” ICED, 2011.Ma, J. and Kim, H. M. ”Predictive Usage Mining for Life Cycle Assessment" Transportation Research ‐ Part D, Vol. 38, pp. 125‐143, July 2015.
Customer review mining
20
Counting and Prediction
Review Text Retrieval
Product Feature Text Mining
3/9/2016
11
Customer review mining
21
Review Text Retrieval
Counting and Prediction
Eric Brill. A simple rule‐based part of speech tagger. In Proceedings of the Third Conference on Applied Natural Language Processing, pages 152–155, 1992.
Included memory is stingy.
The review is turned into a sequence with Part Of Speech* tags.
{included, VB}{memory, NN}{is, VB}{stingy, JJ}
{included, VB}{$feature, NN}{is, VB}{stingy, JJ}
Product Feature Text Mining
Customer review mining
22
Review Text Retrieval
Counting and Prediction
Unstructured Data
Text Tagging
Date Feature Count Type
Jan 31 Screen 5 Cons
Feb1 Camera 9 Pros
Feb1 Apps 15 Pros
Feb1 Keypad 4 Pros
Feb1 Weight 7 Pros
Feb1 MP3 3 Pros
Feb1 WiFi 8 Cons
Feb1 Color 4 Cons
Feb1 Screen 7 Cons
Feb2 Camera 8 Pros
Feb2 Apps 14 Pros
Feb2 Keypad 2 Pros
Structured Data
Product Feature Text Mining
3/9/2016
12
Customer review mining
23
Review Text Retrieval
Counting and Prediction
Product Feature Text Mining
100
50
0
FeatureCount
Time
CameraAppsMP3
Prediction
e.g., Holt‐Winters Forecasting
Level:
Trend:
Season:
1 1( ) (1 )( )t t t s t tL y I L T
1 1( ) (1 )t t t tT L L T
( ) (1 )t t t t sI y L I
Emerging
Static
Obsolete
Large‐scale sensor data from telematics system
• Companies such as Caterpillar (PRODUCT LinkTM) and John Deere (JD LinkTM) have developed telematics systems for their machinery.
• Operational data can be gathered in real time for various purposes: asset utilization monitoring, location tracking, fleet management, machine health prognostics, etc.
source : http://www.equipmentworld.com/24
3/9/2016
13
Engineering design scenario: combine harvesters
Estimate environmental impacts of product life cycle (e.g., 10 years or 20 years) with a usage model from sensor data
Amount of diesel fuel (l)
Available historical data timepresent future
Build usage model Predict usage
real value
constant rate
PUMS
(e.g., 7‐year data) (e.g., 9.5‐year data)
25
Ma, J. and Kim, H.M. “Predictive Usage Mining for Life Cycle Assessment," Transportation Research – Part D, Vol. 38, pp. 125‐143, 2015.
Predictive Usage Mining for Sustainability (PUMS)
26
3/9/2016
14
(Infrequent) Usage modeling
• Automatic segmentation‐ Detect periodic low‐activity periods and group them separately
‐Magnify important patterns and make predictions
‐ Combine the results and maintain the time stamp
Periodic very low‐activity periods
Original data Segmented data
predicted values
27
Usage modeling
• Time series models‐ PUMS‐ets
‐ PUMS‐arima
[Ref.] Hyndman et al. (2008)
autoregressive model, AR(p) integration, I(d)
forecast error
observed time series
Smoothing factor, 0 <α <1
moving average model, MA(q)observed time series
28
3/9/2016
15
Being green can be profitable?
29
Obstacles to making green profit
• OEM remanufacturer’s concerns:• Unbalance b/w supply of cores and demand for reman products
• Unproven environmental sustainability of remanufacturing
• Cannibalization of new product sales
Relevant literature:• Pricing: Focused on pricing from the economic perspective w/ little design
and process consideration. Guide et al. (2003), Ferrer et al. (2010), Vadde et al. (2011)
• Production planning: Pricing was separated from production planning. Mangun and Thurston (2002), Kwak and Kim (2010), Jayaraman (2006), Franke et al. (2006)
• Environmental assessment of remanufacturing: Mostly Life Cycle Assessments (LCAs) comparing new and reman products for the sake of product evaluation not decision‐making. Smith and Keoleian (2004), Goldey et al. (2010)
30
3/9/2016
16
Let’s consider economics together with environment.
For OEM remanufacturers, the new model considers joint pricing and production planning to develop an optimal portfolio of new and remanufactured products.
Two objectives:
Decision variables: • : buy‐back price and take‐back quantity of the end‐of‐life product
• : selling price and the production quantity of the new product
• : selling price and the production quantity of the reman product
,u up x
,n np x
,r rp x
1 2Maximize [total profit , total environmental saving ]f f
31
32
Transition matrix
Transition matrix represents the relationship between product design and remanufacturing operations in a matrix form, such that the impact of product design can be mathematically reflected in the production planning.
Input Outputi ij jj J iT Y
3/9/2016
17
33
Mathematical model (1/2)
The objective is to maximize the total profit from the sales of new and remanufactured products, while achieving environmental‐impact saving δ.
1
2 ,
3 ,
4
5
: ( )
: ( , )
: ( , )
g :
: ( ) { ( )}
k k k k
n l n l n rl L
r l r l n rl L
r k
M Nw k k n r i i j j i i d rk K i I j J i I
g X A s P k K
g Z Q d P P
g Z Q d P P
Z X
g e e X E Z e M e Y e N e Z
max. ( ) ( )M Nn n n r r i i k k j j i i d ri I k K j J i I
P C Z P Z c M P X c Y c N c Z
Profit from the new product Profit from the remanufactured product
Cost of recycling
Cost of takeback
Cost of production
Cost of spare parts
Cost of reassy
and sales
Take‐back availability of end‐of‐life products, given buy‐back price
Production quantity not exceeding the demand, given selling price
Upper limit for remanufacturing
Environmental‐impact saving exceeding the target δ
34
Mathematical model (2/2)
Constraints for the Input‐output flow balance
Variable conditions
1
2
3
: 0 corresponding to the end-of-life product ( )
: 0 corresponding to a part with external purchase availability
: 0 corres
k ij j ij J
i ij j ij J
ij j ij J
h X T Y M i k k K
h N T Y M i
h T Y M i
4
5
6
ponding to a part without external purchase availability
: 0 corresponding to the remanufactured product
: 0 part with external purchase availability
: 0 corre
ij j rj J
i
i
h T Y Z i
h N i
h M i
sponding to the remanufactured product
, , , , 0 and integer , ,
, , , 0 ,
k j n r i
k n r i
X Y Z Z N i I j J k K
P P P M i I k K
3/9/2016
18
35
Example: smartphone
Baseline case: New only. No takeback is conducted.
Target case: Both new and remanufactured products are offered.
Example: engine water pump
Suppose that an OEM manufacturer of engine water pumps is considering starting their own remanufacturing business.
Attribute New pump Competitor
Price $100 $50
Brand OEM Third‐party
Condition New Reman
Market share 45% 55%
Current market status (baseline)
Parameter Value
Available cores in the market 5,000 units
Attainable cores at pu=0 500 units
Buy‐back price for 100% take‐back $30
Assumptions on market conditions
36
3/9/2016
19
Customer preference in the market
Total market size: 3,000 units
Customers prefer a new product w/ lower price and OEM brand.
, , ,
, ,
Utility of product : [1 (1 )] ( )
where
0 remanufactured product 0 remanufactured by third-party
1 (new product) 1 (OEM)
j cond j cond price j price brand j brand
j cond j brand
j U y y y
if ify y
else else
, 1 criticalj price j jy p p
Attribute Utility factor (β) Ideal Critical
Price 0.8 $0 $100
Brand 0.2 OEM Third‐party
Condition 0.45 New Reman
Customer preference assumption
37
Bi‐objective problem: ε‐constraint approach
Using the ε‐constraint approach, the model can incorporate the trade‐offs between profit and environmental saving.
Lower bound for f2(f2 under max f1)
max f2
38
3/9/2016
20
Case study: Scenario 1 (new production only)
The company optimizes the selling price and quantity of the new pump. No remanufacturing is conducted.
39
Case study: Scenario 2 (remanufacturing only)
For this period, closed‐loop production is conducted. The company produces only remanufactured pumps. The selling price and production quantity for the remanufactured pump are optimized.
40
3/9/2016
21
Case study: Scenario 3 (new and reman together)
The company optimizes and sells both the new and remanufactured products.
41
The results shows an opportunity to increase profit and environmental saving at the same time.
Efficient frontier: Pareto‐optimal solutions
Profitable Green, Profitable
Green
Inferior
42
3/9/2016
22
Reman product should be cheaper.
43
Green profit of reman changes over time!
A model for assessing the time‐varying advantages of remanufacturing is developed. The model:
• Compares the remanufactured and brand‐new versions of a product,• Clarifies how the nature of the product influences the time‐varying value,• Provides a multi‐dimensional assessment tool (cost, impact, and net profit).
1 2 3 4 5 6 7 8 9 10-10
0
10
20
30
40
50
60
70
80
Tot
al S
avin
g O
ver
Bra
nd-N
ew (
%)
Product Age (year)
Manufacturing Cost Environmental Impact Net Profit
Physical DeteriorationTechnological Obsolescence
Val
ue o
f Rem
anuf
actu
ring
44
3/9/2016
23
Products w/ physical and technological obsolescence
Each part has its own lifecycle characteristics, in terms of cost, technological obsolescence, and physical deterioration.
45
Products w/ physical and technological obsolescence
Each part has its own lifecycle characteristics, in terms of cost, technological obsolescence, and physical deterioration.
Probability that a part is reusable both physically and technologicallycan be modeled as:
where
target (0)
0
( ) ( , )i i
i in
w t f n t
‐ w(t) : reusability of part i after t years of use‐ n : change in the part generation over t years‐ fi(n,t) : the probability that a total of n generations of part i will appear in [0, t]
the maximum change in part generation allowed for t years to satisfy customer requirement
46
3/9/2016
24
Cost advantage of a remanufactured product
Proposition 1. The cost advantage of remanufacturing over producing the equivalent brand‐new product is formulated as CNR‐RR(t).
• NR: Production of brand-new products w/ responsible take-back and recycling• RR: Remanufacturing w/ recycling; RS: Remanufacturing w/ part resale and recycling
Probability that a part is reusable both physically and technologically
Cost saving from reusing a part
target (0)
,target0
( ) ( ) ( , ) ( ) C ( ) ( )i i
new recond matlNR RR i i i i i
i I n
C t w t f n t C t t V t
1 2 3 4 5 6 7 8 9 100
50
100
150
200
250
300
350
400
450
Cos
t ($,
in p
rese
nt v
alue
at t
=0)
Timing of remanufacturing (product age, in year)
CNR(t) CRR(t) CRS(t) CNR-RR(t) CNR-RS(t)
47
Environmental advantage of a reman product
Proposition 2. The environmental advantage of a remanufactured product over its equivalent brand‐new is formulated as ENR‐RR(t).
Probability that a part is reusable both physically and technologically
Avoided impact by reusing a part
target (0)
,target0
( ) ( ) ( , ) ( ) ( ) ( )i i
new recond matlNR RR i i i i i
i I n
E t w t f n t E t E t E t
• NR: Production of brand-new products w/ responsible take-back and recycling• RR: Remanufacturing w/ recycling; RS: Remanufacturing w/ part resale and recycling
1 2 3 4 5 6 7 8 9 10-50
050
100150200250300350400450500
Env
iron
men
tal i
mpa
ct (
kg C
O2e
)
Timing of remanufacturing (product age, in year)
ENR(t) ERR(t) ERS(t) ENR-RR(t) ENR-RS(t)
48
3/9/2016
25
Net profit advantage of a reman product
Proposition 3. Let β be the price ratio of the remanufactured product to the equivalent brand‐new. Then, the advantage of remanufacturing from the net‐profit perspective is given as ΠRR‐NR(t).
target (0)
,target0
( ) ( ) ( )
( 1) ( ) ( , ) ( ) C ( ) ( )i i
RR NR R RR N NR
new recond matlN i i i i i
i I n
t P C P C
P w t f n t C t t V t
• NR: Production of brand-new products w/ responsible take-back and recycling• RR: Remanufacturing w/ recycling; RS: Remanufacturing w/ part resale and recycling
1 2 3 4 5 6 7 8 9 10-150-100-50
050
100150200250300350
Net
pro
fit (
$, in
pre
sent
val
ue a
t t=
0)
Timing of remanufacturing (product age, in year)
NR(t) RR(t) RS(t) RR-NR(t) RS-NR(t)
(β=0.7)
49
1 2 3 4 5 6 7 8 9 10-20%
0%
20%
40%
60%
80%
100% ENR-RR(t) ENR-RS(t) RR RS
Timing of remanufacturing (product age, in year)Env
iron
men
tal-
impa
ct r
atio
ove
r br
and-
new
0.5
0.6
0.7
0.8
0.9
1.0
1.1
Net profit advantage of a reman product
Corollary 1. The range of β where the remanufacture product becomes more profitable than the brand‐new is β≥β*, where β* is:
target (0)
*,target
0
1 (1/ ) ( ) ( ) C ( ) ( ) ( , )ii
new recond matlRR NR N i i i i i
i I n
P w t C t t V t f n t
Let t=4. Remanufacturing (RR) makes sense if consumers are willing to pay more than 81% of new price on a REMAN product.
Remanufacturing is more profitable only if consumers are willing to pay more than β*of new price.
If beta is known as 0.7, a reasonable strategy is to remanufacture (RR) until t=2.5 and to choose new production afterwards.
• NR: Production of brand-new products w/ responsible take-back and recycling• RR: Remanufacturing w/ recycling; RS: Remanufacturing w/ part resale and recycling 50
3/9/2016
26
CONCLUSION
51
Where is this research heading?
52
Consider cell phones’ fate when designing them.
Acknowledgement of support: NSF CAREER award, Engineering Design and Innovation Program grant, GOALI grant, CAT, Deere.
3/9/2016
27
Summary
53
Design Analytics
Life Cycle Design
Manufacturing Maintenance Usage (Fuel)Usage (Emission) End-of-Life0
50000
100000
150000
200000
250000
300000
10312.11(2.19%)
258182.5(54.78%)
94364.36(20.02%)
29838.24(6.33%)
GW
P 1
00a
[kg
CO
2-eq
]
78637.09(16.68%)
Predictive Design Analytics
Amount of diesel fuel (l)
Available historical data timepresent future
Build usage model Predict usage
real value
constant rate
PUMS
(e.g., 7‐year data) (e.g., 9.5‐year data)
Being green can be sustainable by carefully linking predictive analytics with design and manufacturing.
54
Well‐designed products enable more profitable, sustainable end‐of‐life product take‐back and recovery.
A quantitative model can close the loop of pre‐life, usage life and end‐of‐life of products for sustainability.
Predictive design analytics enables integrated approach to design, manufacturing, and sustainability.