Вы находитесь на странице: 1из 6

Available online at www.sciencedirect.

com

ScienceDirect
Procedia CIRP 57 (2016) 527 – 532

49th CIRP Conference on Manufacturing Systems (CIRP-CMS 2016)

A research on optimization method for integrating component selection


and production scheduling under mass customization
Shota Suginouchia,* , Toshiya Kaiharaa , Daisuke Kokuryoa , Swee Kuika
a 1-1 Rokkodai, Nada, Kobe 657-8501, Japan
∗ Corresponding author. Tel.: +81-78-803-6250; fax: +81-78-803-6250. E-mail address: suginouchi@kaede.cs.kobe-u.ac.jp

Abstract

This paper proposes a method which is to achieve an optimal production scheduling and component selection simultaneously using an autonomous
distributed technique under mass customization environment. The proposal method decides a production scheduling and component selection
to aim minimizing the delay of acceptable due date. The proposal method uses a combinatorial auction for modelling negotiation between
manufacturer and customers toward an optimal plan. Numerical examples are then discussed to demonstrate its suitability in obtaining an optimal
plan under negotiations between manufacturer and customers.
©c2016
2016Published by Elsevier
The Authors. B.V. This
Published by is an openB.V.
Elsevier access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Peer-review under responsibility of Scientific committee of the 49th CIRP Conference on Manufacturing Systems (CIRP-CMS 2016).
Peer-review under responsibility of the scientific committee of the 49th CIRP Conference on Manufacturing Systems
Keywords: Scheduling, Parts allocation planning, Combinatorial auction.

1. Introduction parts for customers’ demands to meet their different character-


istic, such as lightweight, high durability and low price. How-
The aim of this paper is to establish value co-creation manu- ever, these aspects may cause production inefficiency (e.g. in-
facturing system with manufacturer and customers under mass creasing number of set up, increased complications of inventory
customized environment[1]. Mass customized system can re- planning). The purpose of this study is to find optimal variety
spond variety of customers’ demands, and achieve high pro- of parts for manufacturer and customer based on the proposed
duction efficiency with low cost production. To implement a negotiation.
mass customized system, manufacturers are required to achieve This study is the first step to actualize the manufacturing sys-
an optimal production planning, inventory planning, production tem for tailor-made shoes by a reactive 3D printer[2].
scheduling, component selection and so on. When a manufac-
turer optimizes the plans individually, manufacturer may not
be able to obtain an optimal solution due to complexity of the 2. Model formulation
targeted scenario. As the first step, this study focuses on pro-
duction scheduling and component selection, and proposes a 2.1. Model
method integrating production scheduling and component se-
lection to realize of efficient production. Component selec- The model formulation is developed to analyze the negotia-
tion chooses parts to manufacture items under limited material tion between manufacturer and customers for deciding variety
stock, and is dealing variety of customers’ demands. In this pa- of parts.
per, ”customer’s demand” is defined as the brevity of lead time Fig. 1 shows the production flow from raw materials to fin-
to suit for customer preference. ished products. A product consists of P parts (Part 1, Part 2,
The proposed method uses a combinatorial auction in which · · · , Part P). Manufacturer makes P parts in the first process
a social negotiation between manufacturers and customers is and assembles them to manufacture shoes in the second pro-
examined in order to obtain an optimal plan. cess as shown on Fig. 1. Part p (p = 1, · · · , P) has T variations
In this study, the determination of customer’s acceptable due which are called type t (t = 1, · · · , T ). A customer has satisfac-
date depends on each part of composing shoes. Customer may tion value UT Pi,p,t for each type t. Each customer may choose
choose preferred parts which have unique characteristics to pro- the type t of each part p in accordance with the customer’s de-
duce shoes. Manufacturer is required to prepare wide variety of mand. The characteristics differ according to type t, for exam-

2212-8271 © 2016 Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Peer-review under responsibility of the scientific committee of the 49th CIRP Conference on Manufacturing Systems
doi:10.1016/j.procir.2016.11.091
528 Shota Suginouchi et al. / Procedia CIRP 57 (2016) 527 – 532

WƌŽĚƵĐƚ In this study, manufacturer hosts auction. A process to make


customer’s order is a bidder.
ƐƐĞŵďůŝŶŐ
;^ĞĐŽŶĚƉƌŽĐĞƐƐͿ 2.3. Notation
ϭƉŝĞĐĞ ϭƉŝĞĐĞ The notations used in this model are shown as follows:
WĂƌƚϭ WĂƌƚ
i : Order number (i = 1, . . . , I)
WĂƌƚ WĂƌƚ WĂƌƚ WĂƌƚ
ϭ͕ϭ 䞉䞉䞉 ϭ͕ 䞉䞉䞉 ͕ϭ 䞉䞉䞉 ͕ j : Process number of order i ( j = 1, · · · , Ji )
k : Bid number of bidder (i, j) (k = 1, · · · , Ki, j )
WƌŽĐĞƐƐŝŶŐ p : Part (p = 1, . . . , P)
;&ŝƌƐƚƉƌŽĐĞƐƐͿ t : Type (t = 1, . . . , T ) ∈ RP
l : Material (l = 1, . . . , L)
DĂƚĞƌŝĂůϭ
ARi : Arriving time of order i
DĂƚĞƌŝĂů ϭ͕ϭ  ΀Ő΁
ACi,t : Acceptable due date of order i using type t
䞉䞉䞉

S DPi : Standard due date of order i


DĂƚĞƌŝĂůϭ͕  ′ ΀Ő΁ DPi,t : Number of time slot of the delay of due date of order i
manufacturing type t
Fig. 1. Production flow pti, j,k : Processing time of bid k by bidder (i, j)
ple, lightweight, high durability and low price. The customer sti, j,k : Starting time of processing of bid k by bidder (i, j)
evaluates the satisfaction value of type t of part p. yi, j,k,l : Number of material l required for bid k of bidder (i, j)
Customer who gave an order i (i = 1, · · · , I) has the ac- Y j,l,q : Number of q th ordering quantity of material l to man-
ceptable due date DPi,t which is determined by customer i’s ufacture part j
satisfaction value UT i,t of selected type t ∈ RP . A relationship QT q : Arriving time of q th ordering
between the acceptable due date DPi,t and the satisfaction value PNi, j,k : Processed type by bid k by bidder (i, j) (PNi, j,k =
UT i,t is given as an experimental conditions. 1, · · · , T )
In the first process, manufacturer can process different parts UT Pi,p,t : The customer i’s satisfaction value of type t of part
p and p which compose the same order i, at the same time. The p
order can move to the second process on completion of all parts UT i,t : The sum of customer i’s satisfaction value of type t of
which compose the order. 
each part (UT i,t = Pp=1 UT Pi,p,t )
Manufacturer requires at most L kinds of material l (l =
CO : Set-up time
1, · · · , L) to process part p. Processing time and required quan-
Ci,i , j,k,k : A dependent variable that is 1 if the type of pro-
tity of materials are determined in advance as an experimental
cessed part by bid k of bidder (i, j) is same as the bid k
condition and vary by type t . Manufacturer has 1 processing
bidder (i , j) (PNi, j,k = PNi , j,k ) and otherwise 0
machine per part and 1 assembling machine. A set-up change
occurs, when a machine changes manufactured type t. Ui, j,k,q : A dependent variable that is 1 if starting time of pro-
Materials are limited to stock on hand at 0[Time Slot]. Ma- cessing of sti, j,k is earlier than the qth arriving time of ma-
terials arrive at manufacturer in batches afterward. The fixed terials and otherwise 0
interval and fixed quantity ordering system for the inventory COM : The probability to change processed type of part,
planning is assumed. when the acceptable delivery delay occurred at previous
Manufacturer’s criteria is to minimize the total delays from auction
the acceptable due date DPi,t . Each customer’s criterion is to pick : Number of bids per bidder
minimize the delay of the acceptable due date DPi,t . Ite : Number of iteration of auction
BVi, j,k : Bid value of bid k by bidder (i, j)
2.2. Combinatorial auction xi, j,k : A decision variable in bid k by bidder (i, j)
LR : The probability to bring starting time of processing sti, j,k
The combinatorial auction [3] is known as social contract forward the time slot, when a bidder re-makes a bid
based approach and used to develop optimal scheduling and LOOP : The probability to bring starting time of process-
stock allocation planning. The algorithm is shown as follows: ing sti, j,k forward or backward additional time slot, when a
STEP1 Initially set count = 1. Auctioneer sets the number of bidder re-makes a bid
iteration as Ite, and shows the information of goods to the
bidders. 2.4. Bid determination problem

STEP2 The bidders make bids (Bid determination problem). 2.4.1. Definition of a bid
STEP3 The auctioneer decides winner bids which makes ob- In this study, a bidder is process j of order i. Bidder (i, j)
jective function maximum or minimum (Winner determi- makes pick bids an auction. Bid (i, j, k) consists of starting time
nation problem). sti, j,k , number of required material yi, j,k,l to process j of order i
and processed type PNi, j,k . If the bidder (i, j) is second process
STEP4 If count = Ite, then finish. Otherwise, count = count+ of order i ( j = assem), number of required material yi,assem,k,l
1 and return to STEP2. are 0 for all material l.
Shota Suginouchi et al. / Procedia CIRP 57 (2016) 527 – 532 529

2.4.2. Bid value bearing bidding number k of process j of order i becomes the
Equation (1) expresses bid value BVi, j,k of bid k by bidder winner. When this value is 0, the same becomes the loser.
(i, j) . Objective function is expressed by equation (4), which is in-
tended for minimizing the delay of due date.
Equation (5) guarantees that only a bid per bidder becomes
⎧ winner. Equation (6) guarantees that a product can be assem-


⎨max{0, (sti, j,k + pti, j,k − DPi,t )} (if j = assem)
BVi, j,k =⎪ (1) bled after the completion of parts. Equation (7) is constraint

⎩0 (otherwise) equation related to the rights on processing machine under con-
sidering set-up time at a time slot. Ci,i , j,k,k is a dependent vari-
able relating to set-up as shown in equation (11).
2.4.3. Bid constraints
Each bidder avoid to make the bid including infeasible com- ⎧
bination. Then bidding rules in conjunction with the bid are set ⎪

⎨CO (if PNi, j,k = PNi , j,k )
Ci,i , j,k,k = ⎪
⎪ (11)
as shown in equations (2), (3). ⎩0 (otherwise)

Equation (8) is constraint equation related to the rights on as-


ARi ≤ sti, j,k (2) sembling machine at a time slot. Equation (9) is constraint
max min sti , j,k + pti , j,k ≤ sti,assem,k (3) equation related to the inventory upper bound at a time. Equa-
{∀ j| jassem} ∀k
tion (10) defines the sum of customer i’s satisfaction value of
type t of each part UT i,t . A relationship between the accept-
Equation (2) expresses the starting time to process lower able due date DPi,t and the satisfaction value UT i,t can be given
bound for first process. Equation (3) expresses the starting time as an experimental condition. When the winner of first process
to assemble lower bound for second process. xi, j,k ( j  assem) is determined, the acceptable due date DPi,t is
determined.
2.5. Winner determination problem
3. Bidding rule
Objective function and constraints are expressed as follows:
3.1. Initial bid

I K
 i,assem On winner determination problem discussed in detail in sec-
min BVi,assem, j × xi,assem,k (4) tion 2.5, constraints have to be considered. If each bidder do
i=1 k=1 not consider other bidder’s bids, sometimes feasible solution of

Ki, j winner determination problem is not obtained. Because auc-
s.t. xi, j,k = 1 (∀i, ∀ j) (5) tioneer establishes an upper bound of number of bids per bid-
k=1 der to obtain the solution in a short time. Therefore, all bidders

Ki, j K
i,assem make an initial bid in accordance with the dispatching rules.
( sti, j,k × xi, j,k ) + pti, j ≤ sti,assem,k × xi,assem,k The detail of the dispatching rules is as follows:
k=1 k =1
({∀i, ∀ j| j  assem}) (6) STEP1 Arrange orders i in ascending order of arriving time
xi, j,k + xi , j,k ≤ 1 ARi .
({∀i, ∀i , ∀ j, ∀k, ∀k |i  i ∧ j  assem∧ STEP2 Arrange orders i in descending order of standard due
sti, j,k ≤ sti , j,k ∧ sti , j,k ≤ sti, j,k + pti, j + Ci,i , j,k,k }) (7) date S DPi among orders which have same standard due
date S DPi .
xi,assem,k + xi ,assem,k ≤ 1 STEP3 Giving orders for each machine in accordance with
({∀i, ∀i , ∀k, ∀k |i  i ∧ sti,assem,k ≤ sti ,assem,k ∧ the queue which is obtained in STEP1. If the situation of
sti ,assem,k ≤ sti,assem,k + pti,assem,k }) (8) material lacking arises, a machine waits for the arrival of
 materials. If orders have not arrived yet, a machine waits

I 
Ki, j 
q
for the arrival of orders.
yi, j,k,l × Ui, j,k,q × xi, j,k ≤ Yi,l,q
i=1 k=1 q=1
STEP4 Each bidder makes a bid in accordance with the sched-
(∀ j, ∀l, ∀q ∈ {1, 2, · · · , Q}) (9) ule which is obtained in STEP3. Processed type is the

P 
Ki, j largest satisfaction value type for the bidder.
UT i,t = UT Pi,p,PNi, j,k × xi, j,k
p=1 k=1 3.2. Remaking bid by loser
({∀i, ∀ j| j  assem}) (10)
Each bidder makes a bid which is same combination as win-
ner bid in previous auction to prevent a change for worse. To
The decision variable is xi, j,k . When this value is 1, the bid improve solution, a bidder (i, j) re-makes a bid k which did not
530 Shota Suginouchi et al. / Procedia CIRP 57 (2016) 527 – 532

win in previous auction. rand is an uniform random variable in • Initial inventory of material l : 70
the interval [0,1). • Ordering quantity of material l from the second time : 20
• Standard delivery time S PDi : 305̃00[TS]
STEP1 The bidder sets manufacturing type t, starting time of • Set-up time CO : 3[TS]
processing sti, j,k , processing time pti, j,k , and required quan- • Processing time pti, j,k ( j  assem) : 2 5[TS]
tity yi, j,l of material l the same value as the bid of bidder • Assembling time pti,assem,k : 1[TS]
(i, j) which won at previous auction. • The probability to bring starting time of processing sti, j,k
STEP2 If rand < COM , j  assem , and the acceptance de- forward the time slot LR : 0.75
livery and the acceptable capacity delay occurs at previous • The probability to starting time of processing sti, j,k forward
auction, go to STEP3-1. Otherwise go to STEP3-2. or backward additional a time slot LOOP :0.8
• The probability to change processed type of part, when the
STEP3-1 The bidder changes manufacturing type t into type acceptable delivery delay occurs COM : 0.03
t which is processed right before processing order i in the • Number of iteration Ite : 1000
schedule which is obtained at last auction. Go to STEP5 • Number of trials : 50
STEP3-2 If rand < LR , go to STEP4-1. Otherwise, go to
STEP4-2.
Table 1. Processing time and required material
STEP4-1 The bidder brings starting time of processing sti, j,k Processing time of Required quantity
forward a time slot. sti, j,k = sti, j,k − 1. If the bid does not Type t type t of part p [TS] of material l
satisfy bid constraints as shown in equation (2) or (3), the p=1 p=2 p=3 l=1 l=2 l=3
bidder suspends to bring starting time of processing sti, j,k 1 4 4 2 5 2 0
forward, and go to STEP4-2. If rand < LOOP , the bidder 2 3 3 4 4 2 1
3 2 3 3 4 1 1
brings starting time of processing sti, j,k forward additional
4 3 3 5 3 2 2
a time slot. Go to STEP4-1. Otherwise, the bidder finishes
5 4 2 2 3 2 2
making bid. Go to STEP5. 6 4 2 4 2 2 3
7 3 3 4 2 1 3
STEP4-2 The bidder brings starting time of processing sti, j,k
8 2 3 3 1 1 4
backward a time slot. sti, j,k = sti, j,k + 1. It rand < LOOP,
9 4 4 2 1 2 4
the bidder brings starting time of processing sti, j,k back- 10 5 5 3 0 2 5
ward additional a time slot. Go to STEP4-2. Otherwise,
the bidder finishes making bid. Go to STEP 5.
STEP5 If there is the bid k (k = 1, · · · , k − 1) of bidder (i, j) Table 1 shows processing time and required quantity of ma-
which is same combination as the bid k of bidder (i, j), terials to process part 1, part 2 and part 3 for each type t. Pro-
bidder re-makes the bid k. Go to STEP1. cessing time of type t is different from a part p. Required
quantity of material l of type t is common in all parts p. This
4. Computational experiments study uses CPLEX 12.6[4] and Inter(R) Xeon(R) CPU E5 1650
3.50GHz 16.0GB memory computer to solve bid determination
4.1. Evaluation criteria problem.

The criteria of the computation experiments are as follows:


4.3. Definition of the acceptable due date and the satisfaction
value
• OF [TS] : The sum of acceptable capacity DPi , t
• NOD : Number of order that delay of due date occurs
Customers evaluate the satisfaction value for selected type t
• IC [TS] : The sum of idle time and set-up time
of each parts. The satisfaction value of the largest satisfaction
• MT [TS] : Makespan
type t , is 10. The satisfaction value of type (t+Δt) and (t−Δt) is
• AU p : Average satisfaction value of part p for customers
(10−Δt). The satisfaction value UT i,t is between 3 to 30 because
• CT [sec] : Calculation times
the number of part varieties P is 3 and number of type varieties
T is 10. Fig. 2 shows the example of customer’s satisfaction
4.2. Experimental conditions
value for each type. In this case, when type 3 is selected, the
customer’s satisfaction becomes the largest.
The experiments are performed with the following condi-
This study defines the relationship of the acceptable due date
tions:
and the satisfaction value as shown in equations (12), (13), (14).
• Number of order : 100
• Number of part varieties P : 3
• Number of type varieties T : 10 UT i,t
• Number of product varieties : 1000 DPi,t = S DPi × (12)
⎧ 30
• Number of material varieties : 3 ⎪

⎨S DPi (if UT i,t ≥ 20)
• Arrival time of orders ARi : 0[TS] DPi,t = ⎪
⎪ (13)
⎩S DPi − S DPi,t × 20−UT i,t
(otherwise)
• Arrival time of materials : 0,100,130,160, ... [TS] 20
Shota Suginouchi et al. / Procedia CIRP 57 (2016) 527 – 532 531

Utility Table 3. Results using proposal method


ϭϬ OF NOD IC MT
ϴ Linear Ave. 625.7 38.7 1015.8 540.9
S.D. 252.2 8.2 26.7 9.2
ϲ Max. 1320 55 1066 560
ϰ Min. 115 16 947 516
Saturation Ave. 655.9 40.4 1018.3 542.5
Ϯ
S.D. 227.3 6.9 19.6 7.3
Ϭ Max. 1507 56 1064 563
ϭ Ϯ ϯ ϰ ϱ ϲ ϳ ϴ ϵ ϭϬ Type t Min. 167 24 957 520
Un- Ave. 691.1 41.4 1025.0 544.3
compromising S.D. 161.6 4.8 17.1 5.4
Fig. 2. Example of customer’s satisfaction value
Max. 1335 52 1075 563
Min. 426 32 983 533



⎨S DPi (if UT i,t = 30)
AU1 AU2 AU3 CT
DPi,t = ⎪
⎪ (14) Linear Ave. 9.96 9.99 9.99 3515.6
⎩0 (otherwise) S.D. 0.05 0.03 0.02 183.5
Max. 10 10 10 4632.5
Min. 9.80 9.88 9.88 3417.1
Equation (12) is linear type. Equation (13) is saturation type. Saturation Ave. 9.97 9.99 9.99 3578.9
Equation (14) is uncompromising type. When a customer’s S.D. 0.03 0.02 0.03 234.1
type is uncompromising type, the customer does not change the Max. 10 10 10 4541.2
using type t of each part p . Min. 9.86 9.88 9.83 3375.8
In all type, the customers order the most suitable type for all Un- Ave. 10 10 10 3612.0
parts at initial bid. compromising S.D. 0 0 0 119.9
Max. 10 10 10 4024.4
Min. 10 10 10 3491.4
4.4. Initial solution

To discuss efficacy of proposal method, Table 2 shows the


initial solution obtained from the dispatching rule which is
shown in section 3.1.

Table 2. Initial solution


OF NOD IC MT is 38.7 on average 100 trials, and 46.1% of the initial solution
3757 84 1294 620 and 93.5% of the solution of uncompromising type.
AU1 AU2 AU3 CT When customers are saturation type, objective function
10 10 10 1.6
value is 655.9 on average 100 trials, and 17.5% of the initial
solution and 94.9% of the uncompromising type customers.
Number of delayed order NOD is 40.4 on average 100 trials,
and 48.1% of the initial solution and 97.6% of the solution of
4.5. Experimental results uncompromising type.
Because some customers changed required type t of each
This paper uses three different datasets for the analysis us- part p. However, there is little difference between the sum of
ing the proposed method. Dataset 1 is that all customers are idle time and set-up time of linear or saturation type and one of
linear type. Dataset 2 is that all customers are saturation type. uncompromising type. In the experimental condition, there is
Dataset 3 is that all customers are uncompromising type. Other tight limitation about material upper bound. It is not effective
conditions are common in the three datasets. to change required type t. Average satisfaction value AU p is
Uncompromising customer is specific optimization of com- near 10. It seem that it is more effective to change required type
ponent selection and production scheduling. Component selec- t under slack limitation about material upper bound. In that
tion is the first stage. A customer chooses the largest satisfac- case, the customers delayed those acceptable due date, changes
tion type. Production scheduling is the second stage. Combina- required type t.
torial auction is used to plan production schedule. Many customers require the type t which is processed in a
Table 3 shows the results obtained from simulated experi- short time. As a result, number of set-up reduces, makespan
mental works. MT shortens, number of time slot of the acceptance delay of
Objective function value OF of the initial solution is 3757. due date IC reduces, number of type t variety reduces and high
Number of delayed order NOD of the initial solution is 84. efficient production is realized.
When customers are uncompromising type, objective function The purpose of proposed method is to obtain the optimal
value OF is 691.1 on average 100 trials, and number of delayed component selection and production scheduling. Manufacturer
order NOD is 41.4 on average 100 trials. When customers are can obtain meeting point of negotiation between manufacturer
linear type, objective function value OF is 625.7 on average 100 and customers by applying proposal method as a by-product,
trials, and 16.7% of the initial solution and 90.5% of the un- and use the information about the meeting point to develop
compromising type customers. Number of delayed order NOD products.
532 Shota Suginouchi et al. / Procedia CIRP 57 (2016) 527 – 532

5. Conclusion

This paper examined an optimal component selection and


production scheduling under mass customized system. This
paper proposed a method for integrating all of them using the
combinatorial auction. The results obtained based on simulated
experiments showed that the objective function value can be
achieved at 16.7% of the initial solution that is for their ac-
ceptable due date with linear condition. However, there has the
slack limitation for material upper bound and it is not effective
strategy to change required type t.
The next step of this research is to integrate inventory plan-
ning to the proposed method.

Acknowledgement

This research was supported by the Cabinet Office, Govern-


ment of Japan through the Cross-ministerial Strategic Innova-
tion Promotion Program(SIP).

References

[1] B. Joseph Pine II. Mass Customization. Harvard Business School Press.
1992.
[2] Toshiya Kaihara, et al.. Research on Innovative Design/Manufacturing
Methodology of Tailor-made Rubber Products and Socio-Economic Value
Co-Creation with Reactive 3D printer FY2014 Annual Report. 2014.
[3] Sven de Cries and Rakesh V. Vohra. Combinatorial auctions: A Survey.
INFORMS Journal on Computing. 2003.
[4] IBM Web page. http://www-01.ibm.com/software/commerce/
optimization/cplex-opitimizer

Вам также может понравиться