Professional Documents
Culture Documents
Projectfinal Invent1
Projectfinal Invent1
B. Objective
According to Claassen and Gerdessen (2016) about Lot-sizing and scheduling in
food processing industry, they are considered as the most complicated characteristic that
the planners handle with them carefully. In practical situation, lot-sizing and scheduling
problems are conducted by separately in successive hierarchical phases (Claasen and
Beek (1993)). However, the planning process often has to be revised and probably
reschedule everyday, because of the separation between the generated schedule with
practical production demand. Furthermore, the lot-sizing and scheduling models have a
lot of difference. So, the application of MILP models helps the planners to minimize the
total cost of changeover and holding cost in lot-sizing and scheduling.
In this project, the MILP models are divided into two models which in respect of
particular issues. Firstly, it is the optimization of production cost with non-triangular
setups related to lot-sizing and scheduling problems. Then, the second model has the
same purpose of the first model however, it takes perishable characteristic into
consideration. In addition, the non-triangular setups and product decay are the significant
factors that affect directly to the result of this optimizing work as well as the
manufacturing plans. In Problem statement section, it will give the readers a detailed view
about them.
2. Problem statement:
Lot-sizing and scheduling problems in Food Processing Industry (FPI) are heavily
influenced by decay and non-triangular factors. In a distinctive issue on lot-sizing and
scheduling the research of Clark et al. (2011) pointed out that the need for more
realistic and practical variants of models for simultaneous lotsizing and scheduling. So
the main purpose of this project is how to cover two characteristics in models for
simultaneous lot-sizing and scheduling , MILP model that handles these characteristics
First of all, sequence-depend setup time, in general, can be defined as the time required
to prepare the necessary resource (e.g., machines, people) to perform a task (e.g., job,
operation). the sequence-dependent type, setup time depends on both the task and its
preceding task. Secondly, Non-triangular is as important factor in this problem, it can be
desmontrated “triangular” setup times occur when it is never worse to setup from
product p to r directly than to setup via a third product q, so that triangular inequality
s(p,r) ≤ s(p,q) + s(q,r) always holds (as shown on the left side of Figure 2). However, in
the animal feed and other industries, the contamination of a product r by a previous
product p just beforehand can be often avoided by producing enough of an
intermediate product q so that it absorbs p’s contamination. For this to save time, the
triangular inequality must not hold in this case, ie, the sum of the setup times s(p,q)
from p and s(q,r) to r must be short enough so that s(p,q) + s(q,r) < s(p,r). With
Sequence-dependent setups and non-triangular setups. There is a complicating problem
with respect to sequence-dependent setup costs and times, commonly referred to as
the assumption of the triangular setup conditions (Almada-Lobo et al., 2008; Clark et al.,
2011; Clark et al., 2014; Gupta and Magnusson, 2005)). Menezes et al. (2011) confirm
that nontriangular setups may occur in FPI due to contamination between production
lots. Clark et al. (2014) mention that contamination is a particularly concern for FPI. The
writer refer for instance to the severe (i.e. lethal) impact of copper in concentrates for
sheep due to an ineffective sequencing of production lots.
By adding another product k between the process of two subsequent products i and j,
we can reduce changeover costs and times consumed by processing conditions of
different product variants (e.g. several heating and/or cooling levels) and other product
specific requirements (e.g. flavours, addition of specific additives, the danger of
contamination between subsequent production runs. The aftermath of applying
triangular setup conditions models is: there might be result in non-consistent solutions
from a scheduling orientation. It is common in FPI cases that after production, the
quality or value of perishable food products degrades in no time. Breaking down
production in lot-sizing requires smaller production quantities. As a result, the pressure
rises as individual production requires higher manufacturing frequency, which level up
the diffculty of sequencing. This paper investigates implementing the characteristics into
models for simultaneous lot-sizing and scheduling under tight capacity constraints. We
present a MILP model that includes the identified characteristics.
After adding non-triangular setups and product decay in small scale examples, the
process shows significant changes in the optimal production schedules. We compare
two presented model formulations with a known approach from literature. The
remainder of the paper is as follows.
In literature section, we embed the model in existing approaches from the literature.
Model formular Section presents two MILP models for the problem under
consideration. Numerical illustrations and benchmark Section provides small scale
numerical examples to demonstrate the impact of non-triangular setups and product
decay. Besides, we also conduct the complexity of the model. Section of analys provides
numerical results for medium-size instances, including a comparison with a
straightforward MP-based heuristic. Concluding remarks and suggestions for further
research are given in last section.
A. Small bucket
A common instance for SB approaches is the Discrete Lot-sizing and Scheduling Problem
(DLSP). The essential DLSP includes (sequence-independent) setup costs and setup
carry-over at zero setup time (Fleischmann, 1990). formation of setup carry-over implies
that setup states of a machine are carried over between period boundaries. Porkka et al.
(2003) compare models with and without setup carry-overs. They demonstrate that
substantial savings in costs and production time can be attained by fundamentally
different production plans enforced by carry-overs. Comparable results are found by Sox
and Gao (1999). Nonetheless, in the basic DLSP, setup states are not preserved over idle
time. Sequence-dependent setup costs and times are neither considered in the DLSP.
Many expansion of the (basic) DLSP have been described in literature. We refer to Drexl
and Kimms (1997) and Salomon et al. (1991) for a broader view on variants of the DLSP.
Fleischmann (1994) considers the multi-item single machine DLSP with
sequencedependent setup costs. An artificial product is introduced to deputized
idleness of the machine. Salomon et al. (1997) continue the latter study and reformulate
a DLSP that captures sequence-dependent setup times (DLSPSD). The triangular setup
conditions are claimed to hold. Nonetheless, machine idleness is represented by an
artificial product. Jordan and Drexl (1998) present a comparable model in which idleness
is stated by an artificial product too. It should be mentioned that if idleness is
represented by an artificial product, the changeover matrix must fulfil in strict
conditions to cope with sequence-dependent setup times. In other way, the setup state
of the machine is not accurately carried over across the barriers of idleness. A recent
approach is due to Guimarães et al. (2014). The authors reach a classification framework
to survey and classify the main modelling approaches for the integration of sequencing
decisions in discrete time lot-sizing and scheduling models.
Wolsey (1997) elongate the work of Constantino (1996) for issue with sequence
independent setups for the formulations of small bucket with sequence-dependent
setup times and costs. Idleness is not represented by an artificial product. Nonetheless,
the conditions of triangular setup are hold as assumption. We will refer to Wolsey’s
model as the general small bucket model (GSB)
B. Big bucket
Opposed to small bucket models, the planning horizon of a big bucket (BB) model is
usually separated into longer periods of equal length. Time intervals in a BB model may
represent a time slot of one or more than week in practice (Drexl and Kimms, 1997.
There can be multiple products generated in each cycle. Releasing the "all-or-nothing"
development concept of (most) SB models means that the option of deciding
continuous lot sizes is included in a BB model.
The Capacitated Lot-Sizing Problem (CLSP) is a typical example of a big bucket model. It
is closely related to the (small bucket) DLSP. Decision variables, parameters and
objective function are comparable in both problems (Drexl and Kimms, 1997).
Nonetheless, the CLSP does not consist of sequence-dependent setup costs and times.
As a result, there is no included configuration carry-over between time limits either.
Suerie and Stadtler (2003) used the simple problem of plant position to obtain a tight
new model formulation with sequence-independent setup costs and times for setup
carryover in the CLSP.
Sox and Gao (1999) introduce the Generalized Capacitated Lot-sizing Problem (GCLP).
The GCLP applies less binary variables for including setup carry-over in the CLSP with
sequence-independent setup costs and no setup times. Sequence-independent setup
times may be included. Perhaps at the cost of extra work to compute. As suggested by
Eppen and Martin (1987), the authors also apply the network reformulation method and
compare the behaviour of a series of models. The findings show that the introduction of
setup carry-over has a huge impact on both cost and lot sizes.
Notation a propensity to integrate features of small bucket models into big bucket
models in simultaneous lot-sizing and scheduling. We refer to suggested variants of the
CLSP for clarification (Almada-Lobo et al., 2007; Almada-Lobo et al., 2008;
Gopalakrishnan, 2000; Gopalakrishnan et al., 1995; Guimarães et al., 2013; 2013; Gupta
and Magnusson, 2005; Haase and Kimms, 2000; Menezes et al., 2011), variants of hybrid
BB and SB models like the General Lot-sizing and Scheduling Problem (GLSP) (Ferreira et
al., 2012; Ferreira et al., 2009; Fleischmann and Meyr, 1997. Guimarães et al., 2013,
2014; Meyr, 2000; Transchel et al., 2011), and variants of block planning approaches,
originally introduced by Gunther et al. (2006). The literature review on extensions of
capacitated lot-sizing by Quadt and Kuhn (2008) and Guimarães et al. (2014) confirm the
trend in which (hybrid) BB approaches are preferred to SB models. For a complete
overview we also refer to (Claassen and Hendrix, 2014).
It is still a challenge to integrate sequence-dependent setup costs and times for
simultaneous lot-sizing and scheduling.
1. Non-triangular setup:
Non-triangular setups may occur in FPI due to contamination between
production lots. Clark et al. (2014) mention that contamination is a particularly
concern for FPI. The authors refer for instance to the severe (i.e. lethal) impact of
copper in concentrates for sheep due to an ineffective sequencing of production
lots. Due to processing conditions of different product variants (e.g. several
heating and/or cooling levels) and other product specific requirements (e.g.
flavours, addition of specific additives, the danger of contamination between
subsequent production runs), changeover costs and times between two
subsequent products i and j may become substantially less by processing
another product k between i and j. As a consequence, applying models that
assume triangular setup conditions may generate non-consistent solutions from
a scheduling point of view.
B. Limitation:
- the computational effort for the SB model is substantial. We performed
exploratory research with a straightforward implementation of a Relax-and-Fix
construction heuristic. Numerical tests show that the quality of R&F solutions is
promising at manageable computational effort.
- For further research, the improvement computationa performance by adding
valid inequalities (VI’s) a priori to the initial formulation. Preliminary research
revealed that adding the inequalities Y W Y W V j t j t j t j t j t j t , , , , 1 , 1 , 1
tightened the linear relaxation and provided substantially better
heuristic solutions in an R&F framework. The inequalities are based on feasibility
conditions within any production scheme and are added before calling the MP-
solver.
II. Model formulation and description:
In this section, two SB models are demonstrated as the impact of (i) relaxing the
triangular setup conditions and (ii) taking product decay into account. Section Outline of
the lot-sizing and scheduling problem is used as an illustration for the problem and then
modeled in Section Notation and model formulationas a lot-sizing and scheduling
problem with non-triangular setups. In Section Modelling product decay, we expand the
model used to focus on the product decay of inventory by adding an age-dependent
factor in the inventory-holdings costs.
Variables
𝐼𝑖,𝑡 inventory level of item i at the end of period t
1 if product is produced in period t
𝑌𝑖,𝑡 ={
0 otherwise
𝑉𝑗,(𝑡−𝑎𝑖,𝑗+𝜏) ≥ 𝑌𝑖,(𝑡−𝑎𝑖,𝑗−1) + 𝑌𝑗,𝑡 + 𝑊𝑗,𝑡 For all (𝑖, 𝑗) ∈ 𝐴 ≔ {(𝑖, 𝑗)|𝑎𝑖,𝑗 > 0} (6)
𝑡−1 𝑡 = (𝑎𝑖,𝑗 + 2), . . 𝑇, 𝜏 = 0, . . (𝑎𝑖,𝑗 − 1)
− ∑ ∑ 𝑌𝑘,𝑙 − 1
𝑘;𝑘≠1≠𝑗 𝑙=𝑡−𝑎𝑖,𝑗
• The objective function (2) minimizes the total amount of changeover and inventory-
holding costs.
• Constraint (3) represents the inventory equilibrium equations and makes sure that
demand di,t for item i in period t is satisfied without backlogging.
• Along with equations (4), equations (9) and (10) assure that in each time gap, the
machine is either producing item i at maximum capacity.
• Constraint (5) guarantees that the sufficient time between two subsequent
production bundles, i and j, is restrained for the changeover.
• For constructive adjustment over time, inequality (6) accomplishes the setup
variables to be non-zero between two following batches i and j. The term in (6)
represents the manufacture of an item other than i and j within the time interval.
• Constraints (7) assure that a setup cannot take place after idleness. As a result, a
possibility for an idle period will take effect after a system becomes valuable.
• Inequalities (8) help avoid the situation when idleness in period t before
manufacturing item i is followed by the production (or preliminary idleness) of
another item j.
• Finally, restrictions (9) and (10) define the integrality and non-negativity
requirements.
When a conventional objective function like (2) is applied for decomposable items, commonly
used linear holding costs in lot-sizing models may interrupt a critical equilibrium between
changeover costs on the one hand and inventory-holding costs on the other hand. Product
decay affects the remaining shelf life of products. An age-dependent component in the
inventory-holding costs contains this aspect.
Product decay of inventory can integrate into an SB type model like (2)-(10) in the following
way. Let additional parameters represent the perishability rate of item i. Next, we reconsider
the inventory variables as the inventory level of item i at the end of period t, produced in period
q. Now, the objective function (2) is replenished by (2b) in which (t-q) represents product age.
minimize cost;
subject to {
Constraint (3) represents the inventory balance equations and assures demand d[i][t] for item i in
period t is fulfilled without backlogging:
forall (i in items, t in periods)
I[i][t-1] + Y[i][t] - d[i][t] == I[i][t]; //constraint 3
Equations (4), together with (9) and (10), guarantee that in each time interval the machine is
either producing item i at full capacity, changing over, or idle before manufacturing the next
batch of an item:
forall (t in periods)
sum(i in items)Y[i][t] + sum(j in items)V[j][t] + sum (j in
items)W[j][t] == 1; //constraint 4
Constraint (5) assures that between two subsequent production batches i and j, sufficient
time (ai,j) is reserved for changeover:
forall (i in items,j in items: a[i][j]>0)
forall (z in 1..a[i][j])
forall (t in 1..nbperiods-z)
Y[i][t] - sum(k in items, s in (t+1)..(t+z-1): k!=i &&
k!=j)Y[k][s] + Y[j][t+z] <=1 ; //constraint 5
For positive change over time (ai,j>0), inequality (6) enforces the setup variables to be non-zero
between two subsequent batches i and j; if item j is produced in period t or the machine is idle in
period t before manufacturing item j in period t’ (t’>t), setup variables V[j][t-a[i][j]+z]
should get a value of one for z = 0..a[i][j]-1. The term sum(k in items, l in t-a[i][j]..t-1:
k!=i && k!=j)Y[k][l] in (6) represents the production of another item than i and j within time
interval [t-a[i][j], t-1]:
forall (i in items,j in items: a[i][j]>0)
forall (t in a[i][j]+2..nbperiods)
forall (z in 0..a[i][j]-1)
V[j][t-a[i][j]+z] >= Y[i][t-a[i][j]-1] + Y[j][t] + W[j][t]
- sum(k in items, l in t-a[i][j]..t-1: k!=i && k!=j)Y[k][l] - 1; //constraint 6
Constraints (7) assure that a setup cannot take place after idleness. Hence, a possible idle period
will take place after a setup is counted for:
forall (t in 1..nbperiods-1)
sum(j in items)W[j][t] + sum(j in items)V[j][t+1] <= 1; //constraint 7
Inequalities (8) prevent that idleness in period t before manufacturing item i is followed by the
production (or preliminary idleness) of another item j:
forall (i in items, t in 1..nbperiods-1)
W[i][t] + sum(j in items: j!=i)Y[j][t+1] + sum(j in items:
j!=i)W[j][t+1] <= 1; //constraint 8
Finally, restrictions (9) and (10) define the integrality and non-negativity requirements:
forall (i in items, j in items, t in periods){
Y[i][t] >= 0 && Y[i][t] <=1;
V[i][t] >= 0 && V[i][t] <=1;
W[i][t] >= 0;
I[i][t] >= 0;
} //constraint 9+10
}
Now, objective function (2) is replaced by (2b) in which (t-q) represents product age:
dexpr float cost = sum(j in items, t in periods) S[j] * V[j][t] + sum(i in items, t
in periods, q in periods: q <= t)(h[i] * pr^(t-q) * I[i][t][q]);
minimize cost;
subject to {
Replacing (3) by (3a)-(3e) describes the age dynamics of the inventory levels:
forall (i in items)
I[i][1][1] == Y[i][1] - d[i][1]; //constraint 3a
forall (i in items)
I[i][2][2] == sum(z in 1..2)(Y[i][z]-d[i][z]) - I[i][2][1]; //constraint
3c
forall (t in periods)
sum(i in items)Y[i][t] + sum(j in items)V[j][t] + sum (j in
items)W[j][t] == 1;
forall (t in 1..nbperiods-1)
sum(j in items)W[j][t] + sum(j in items)V[j][t+1] <= 1;
As you can see, at the end of the code we have to exclude the two constraints 9 and 10 out of
the code. Because there is a mistake in this paper, the “On production planning and scheduling
in food processing industry: Modelling non-triangular setups and product decay” published in
21 June 2016, which leads to the negative inventory in the result. The mistake is shown in the
picture below:
In order to make the code running, we decided to exclue those two constraints. Therefore, the
results of model with product decay will have negative inventory at some period.
i \ t 0 1 2 3 4 5 6 7 8 9 10 11 12
1 0 0 0 0 0 1 0 0 0 0 0 2 0
2 0 0 0 0 0 0 1 0 0 0 0 0 1
3 0 0 0 0 0 0 0 0 1 0 0 0 1
Table 1: Daily demand
Holding cost h[i] Setup cost S[i] Product decay rate pr[i]
1 11 100 1.2
2 30 100 1.2
3 35 100 1.2
Table 2
The triangular setup conditions refer to the changeover matrix A (with respect to setup times),
the changeover matrix S (with respect to setup costs), or both:
aij aik + akj and/or sij sik + skj for all items i, j, k
i \
j 1 2 3
1 0 2 1
2 1 0 1
3 1 1 0
i \
j 1 2 3
1 0 2 1
2 3 0 3
3 1 1 0
2. Result analysis
This section provides small scale numerical examples to demonstrate the impact of non-
triangular setups and product decay:
❖ Triangular setup and no product decay (using table 3 and pr=1):
❖ Non-triangular setup and no product decay (using table 4 and pr=1):
V. Sensitivity analysis
This paper studied of the impact of two features (i) non-triangular setup condition and (ii)
perishability. Therefore, in the sensitivity analysis we will use the changeover time matrix in
Table 4, the product decay rate of pr=1.2 and change the holding cost to see the differences:
❖ Holding cost h[i]=[11 30 35]:
❖ Decrease the holding cost of item 3 because the Inventory level of item 1 is more than the two
other items (2&3) and we have h[i]=[5 30 40]:
Conclusion: Compare the total cost of 3 cases that we use to analyze the sensitivity, we can see
the total cost is reduced be reduce the holding cost of the item that have the most inventory
level.