You are on page 1of 15

JID: CAOR

ARTICLE IN PRESS [m5G;January 17, 2018;17:22]

Computers and Operations Research 0 0 0 (2018) 1–15

Contents lists available at ScienceDirect

Computers and Operations Research


journal homepage: www.elsevier.com/locate/cor

A sustainable aggregate production planning model for the chemical


process industry
Gerd J. Hahn a, Marcus Brandenburg b,c,∗
a
German Graduate School of Management & Law, Professorship of Operations Management and Process Innovation, Bildungscampus 2, 74076 Heilbronn,
Germany
b
School of Business, Flensburg University of Applied Sciences, Kanzleistraße 91-93, 24943 Flensburg, Germany
c
Department of Supply Chain Management, University of Kassel, Kleine Rosenstraße 1-3, 34117 Kassel, Germany

a r t i c l e i n f o a b s t r a c t

Article history: Process industries typically involve complex manufacturing operations and thus require adequate deci-
Received 30 March 2015 sion support for aggregate production planning (APP). In this paper, we focus on two relevant features
Revised 27 June 2017
of APP in process industry operations: (i) sustainable operations planning involving multiple alternative
Accepted 13 December 2017
production modes/routings with specific production-related carbon emission and the social dimension of
Available online xxx
varying operating rates, (ii) integrated campaign planning with the operational level in order to antic-
Keywords: ipate production mix/volume/routing decisions on campaign lead times and WIP inventories as well as
Sustainable production the impact of variability originating from a stochastic manufacturing environment. We focus on the issue
Aggregate planning of multi-level chemical production processes and highlight the mutual trade-offs along the triple bottom
Decision support system line concerning economic, environmental and social factors. To this end, production-related carbon emis-
Queuing networks sion and overtime working hours are considered as externalized factors as well as internalized ones in
Chemical process industry
terms of resulting costs. A hierarchical decision support tool is presented that combines a deterministic
linear programming model and an aggregate stochastic queuing network model. The approach is exem-
plified at a case example from the chemical industry to illustrate managerial insights and methodological
benefits of our approach.
© 2017 Elsevier Ltd. All rights reserved.

1. Introduction 2010; Zhou et al., 20 0 0) or process control (see Penkuhn et al.,


1997) entail substantial potential to improve the sustainability
The European chemical process industry is of high economic performance. Recent industry reports confirm the relevance and
relevance with sales of 558 bn EUR in 2012 being the fifth largest contribution of a sustainable approach to managing supply chain
industry in the EU28 countries (Cefic, 2014). However, the indus- (SC) operations in the chemical process industry (Cefic, 2014).
try is considerably challenged by sustainability issues involving In general, sustainable supply chain management (SSCM), i.e.,
ecological and social aspects such as natural resource consump- the “integration of sustainability concerns to the decisions of
tion, greenhouse gas (GHG) emissions, and anticipatory social the SC, in an attempt to improve and enhance the economic,
partnership (Cefic, 2014; Shah, 2005). Many of these issues arise environmental and social performances” (Boukherroub et al., 2015,
in strategic tasks such as product design and process engineering p. 181), has gained strong relevance for academics and practition-
(see Dietz et al., 2005; Jödicke et al., 1999), supply network con- ers (Govindan and Cheng, 2015). The corresponding triple bottom
figuration (see Hugo and Pistikopoulos, 2005; You et al., 2012) or line (TBL) of a profit-planet-people ecosystem is reflected in for-
the development and selection of eco-friendly technologies (see mal models and decision support systems for SSCM (Brandenburg
Chakraborty et al., 2004). Despite this, operational tasks such as et al., 2014; Taticchi et al., 2015). However, formal SSCM mod-
production planning and scheduling (see Adonyi et al., 2008; Ar- els are applied comparably seldom for supply chain operations
biza et al., 2008; Song et al., 2002; Vaklieva-Bancheva and Kirilova, planning in the chemical industry (Brandenburg et al., 2014). Ag-
gregate production planning (APP) models that support mid-term
production mix/volume and capacity management (Stadtler, 2005)

Corresponding author. represent a key tool to manage operational SC performance also
E-mail addresses: gerd.hahn@ggs.de (G.J. Hahn), marcus.brandenburg@hs- in the chemical industry (Papageorgiou and Pantelides, 1993).
flensburg.de, brandenb@uni-kassel.de (M. Brandenburg).

https://doi.org/10.1016/j.cor.2017.12.011
0305-0548/© 2017 Elsevier Ltd. All rights reserved.

Please cite this article as: G.J. Hahn, M. Brandenburg, A sustainable aggregate production planning model for the chemical process
industry, Computers and Operations Research (2018), https://doi.org/10.1016/j.cor.2017.12.011
JID: CAOR
ARTICLE IN PRESS [m5G;January 17, 2018;17:22]

2 G.J. Hahn, M. Brandenburg / Computers and Operations Research 000 (2018) 1–15

Consequently, the aforementioned ecological and social aspects the chemical process industry introduced by Papageorgiou and
need to be integrated into APP while accommodating for two Pantelides (1993).
specific requirements of chemical manufacturing which are further The remainder of this paper is structured as follows.
elaborated in the following. Section 2 comprises a review of related literature followed by
First, chemical production often operates in batch mode on the formal description of the decision support framework in
multi-purpose processing units with time- and quantity-based Section 3. In Section 4, we introduce the case example and present
storage constraints in a multi-stage manufacturing environment the findings of our numerical analysis. The paper ends with a
(Kallrath, 2002b).1 Many of such batch processes can be operated discussion of the results in Section 5 and concluding remarks in
on several alternative processing units and, vice versa, many Section 6. For a comprehensive description and lists of all planning
processing units can execute several processes. In order to route parameters we refer to Brandenburg and Hahn.
the material flow across different production stages, a routing is
formed, i.e., processing units are interconnected for the respective
processes before the production starts. As a consequence, finished 2. Literature review
products can be manufactured via alternative production routings
in which different processing units with specific carbon emission 2.1. Sustainable supply chain management
levels are combined. Selecting alternative production routings thus
is a key lever to manage the capacity utilization in an eco-efficient SSCM is a research area of growing interest that focuses not
way. Furthermore, the issue of capacity management for manufac- only on the economic perspective, but also includes the envi-
turing resources is closely related to social aspects, e.g. the amount ronmental and social dimension in SC decisions (Boukherroub
of overtime which usually results in additional working hours or et al., 2015; Seuring and Müller, 2008). In the following, main
extra shifts that are less preferred by employees (Yura, 1994) or SSCM characteristics are briefly summarized and explained. The
which are sometimes covered on short notice by contract workers forward orientation which SSCM inherits from traditional SCM
and backup facilities. To the best of our knowledge, there is no (Seuring and Müller, 2008) can be complemented by a reverse
comprehensive approach to managing the aforementioned aspects perspective of remanufacturing, reverse logistics, disassembly and
of eco-efficiency and social acceptability on an aggregate planning product recovery (Ilgin and Gupta, 2010). Strategic aspects of SSCM
level. cover the design of sustainable products, processes and networks
Second, chemical production processes are often operated in and furthermore the selection of green technology (Ilgin and
campaign mode, i.e., all resources are allocated to the repetitive Gupta, 2010; Seuring and Müller, 2008). Sustainable operations
production of a single product for a longer time period (several can be categorized using the functional structure of an enterprise
days or even weeks) thereby reducing setup costs and operations which includes purchasing, manufacturing and logistics as well as
complexity (Papageorgiou and Pantelides, 1993). The resulting management, finance and accounting functions (Jaehn, 2016). The
campaign planning problem thus combines aspects of mid-term organizational perspective of SSCM ranges from sustainable oper-
capacity planning and short-term batch scheduling (Li and Ier- ations within a single firm to an inter-organizational perspective
apetritou, 20 08; Shah, 20 05). This requires a vertically integrated across the SC when sustainability requirements are triggered by
approach to anticipate the impact of campaign scheduling on mid- downstream customers and propagated to upstream suppliers and
term capacity levels and aggregate lead times. Chemical processes ultimately to SSCM requirements that are generated outside the
feature considerable variability with respect to equipment avail- SC, e.g. by pressures and incentives of legal authorities or other
ability or the durations of setup and processing times (Ishii and stakeholders like NGOs (Seuring and Müller, 2008).
Muraki, 1996; Li and Ierapetritou, 2008). However, most campaign The overall objective of SSCM is to add economic, environmen-
planning models are based on deterministic assumptions thereby tal and social value (García-Serna et al., 2007). Traditional objec-
omitting the impact of operations variability (Kallrath, 2002a) tives of the economic dimension are related to cost and profit,
which is further amplified by campaign building and alternative inventory and makespan or earliness and tardiness (Méndez et al.,
production routings as discussed before. 2006). The environmental dimension comprises input-oriented
Motivated by these observed research gaps, an optimization ap- factors such as consumption of energy or natural resources and
proach for campaign planning under consideration of sustainability furthermore output-oriented ones that usually focus on waste and
criteria in a stochastic manufacturing environment is suggested in pollution or the emission of greenhouse gases (GHG), in particular
this paper. The proposed decision support tool builds on the foun- CO2 (Brandenburg et al., 2014).
dations of aggregate planning models to find a cost-efficient plan The people dimension of SSCM is strongly related to the work-
considering environmental aspects of production-related carbon force that operates the SC. In this regard, work conditions, work
emission and social implications of employees’ overtime. The TBL schedules and overtime work are identified as a crucial factor
criteria are either internalized in the economic cost perspective or of social performance (Northrup et al., 1979; Vogel et al., 2012).
externalized and thus measured independently. The APP approach This illustrates the importance of labor time as social factor in
is integrated with an aggregate stochastic queuing network (ASQN) production planning and scheduling.
model that considers the variability of equipment availability In contrast to conceptual frameworks and empirical studies, for-
and operations times as well as the induced variability due to mal models represent a comparably new stream in SSCM research
campaign building and the selection of alternative production that has strongly grown over the last few years (Brandenburg
routings. The proposed planning procedure is evaluated in a com- et al., 2014; Brandenburg and Rebs, 2015). Several reviews inform
prehensive numerical study that is based on a case example from about recent developments and future trends in this area (see e.g.
Brandenburg et al., 2014; Seuring, 2013), and innovative operations
research applications are described in a special section of Comput-
ers & Operations Research (see Govindan and Cheng, 2015, and the
1
According to Kallrath (2002b, p. 225), a production system includes ‘multi- papers cited therein). Regarding the model type, several reviews
purpose processing units’ if a unit can produce several products with mode- reveal that sustainability impacts of production systems are often
dependent production rates and, vice versa, a product can be produced in different
modes. Moreover, production is called ‘multi-stage’ if (mode- and production unit-
assessed by life cycle analyses (LCA) (Seuring, 2013) or determin-
dependent) recipe structures enable the production of intermediate products before istic multi-criteria models while stochastic approaches are rarely
the manufacture of the final product (Kallrath, 2002b, p. 225). applied (Brandenburg et al., 2014; Brandenburg and Rebs, 2015).

Please cite this article as: G.J. Hahn, M. Brandenburg, A sustainable aggregate production planning model for the chemical process
industry, Computers and Operations Research (2018), https://doi.org/10.1016/j.cor.2017.12.011
JID: CAOR
ARTICLE IN PRESS [m5G;January 17, 2018;17:22]

G.J. Hahn, M. Brandenburg / Computers and Operations Research 000 (2018) 1–15 3

Table 1
Formal SSCM models for the chemical process industry.

TBL dimensionb
a
Authors Planning task Econ. Environ. Social Modeling approachc

Infante et al. (2013) Sector anal. Various Various Various MCDM


Jawjit et al. (2015) Sector anal. n.a. Aggreg. n.a. LCA
Jawjit et al. (2010) Sector anal. n.a. GHG n.a. Linear eq.
Saidur and Mekhilef (2010) Sector anal. n.a. Energy n.a. Linear eq.
Zhao et al. (2014) Sector anal. n.a. Aggreg. n.a. SD
Dietz et al. (2005) Proc. eng. Cost Liquid & emission n.a. GA & LCA
Jödicke et al. (1999) Proc. eng. n.a. Aggreg. n.a. LCA
Chakraborty et al. (2004) Tech. & pol. Cost Waste n.a. MILP
Linninger et al. (20 0 0) Tech. & pol. Cost Waste & n.a. MILP
Brandenburg (2015) SC config. NPV CO2 e n.a. MILP
Hugo and Pistikopoulos (2005) SC config. NPV Aggreg. n.a. MILP & LCA
You et al. (2012) SC config. Cost CO2 e Employment MILP, LCA & IOA
Zhou et al. (20 0 0) SNP Profit Various Mkt. dem. MILP & AHP
Adonyi et al. (2008) PPS Cost Waste n.a. LP
Arbiza et al. (2008) PPS Cash flow Various n.a. GA & LCA
Berlin et al. (2007) PPS n.a. Waste n.a. Heuristics & LCA
Song et al. (2002) PPS Profit Aggreg. n.a. MILP
Vaklieva-Bancheva and Kirilova (2010) PPS n.a. Aggreg. n.a. MILP
Yue and You (2013) PPS Profit & Makespan Aggreg. n.a. MINLP & LCA
Penkuhn et al. (1997) Proc. Ctrl. Cost Waste & emission n.a. MINLP
a
Acronyms: PPS = production planning and scheduling, SNP = supply network planning
b
Acronyms: CO2 e = CO2 equivalent, GHG = greenhouse gases, NPV = net present value,
c
Acronyms: AHP = analytic hierarchy process, GA = genetic algorithm, IOA = input-output analysis, LCA = life cycle analysis, LP = linear program-
ming, MCDM = multi-criteria decision-making, MILP = mixed-integer linear programming, MINLP = mixed-integer non-linear programming, SD = system
dynamics

With regard to the sustainability goals, Seuring (2013) dis- dimensions represents another imbalance. Environmental aspects
tinguishes between complementary objectives that enable are considered in every SSCM model while the social dimension is
win-win situations between different TBL factors, tradeoffs reflected only in three studies and omitted in all operational plan-
between conflicting sustainability objectives and minimum sus- ning models. Ecologic indicators that measure waste and emissions
tainability performance requirements that need to be satisfied. are often applied as well as aggregate environmental performance
Boukherroub et al. (2015) find that environmental criteria are often metrics, e.g. the Eco-Indicators 95 & 99 (Hugo and Pistikopoulos,
internalized as economic cost objectives and that socio-ecological 2005; Jödicke et al., 1999). The social dimension is assessed by so-
factors are measured by aggregate indices. These approaches cietal metrics such as generated employment (Infante et al., 2013;
hardly achieve a comprehensive balance between the TBL dimen- You et al., 2012), work-related illness and death rates (Infante et al.,
sions. Consequently, Pareto analyses are useful to illustrate win- 2013) or market demand satisfaction (Zhou et al., 20 0 0). The eco-
win or tradeoff situations. Most reviews show that SSCM models nomic performance is most often related to cost or measured by
predominantly focus on environmental aspects while neglecting profit, net present value (NPV) or cash flow. MILP is the preferred
social ones (Brandenburg et al., 2014; Brandenburg and Rebs, modeling technique and is proven suitable for strategic decision
2015; Seuring, 2013). Especially models that reflect labor time or support (Brandenburg, 2015; Chakraborty et al., 2004; Hugo and
work schedule in production planning and scheduling are scant. Pistikopoulos, 2005; Linninger et al., 20 0 0; You et al., 2012; Zhou
Yura (1994) presents a scheduling problem with the objective of et al., 20 0 0) as well as for production planning and scheduling
satisfying workers’ preferences for days off and overtime and, thus, (Song et al., 2002; Vaklieva-Bancheva and Kirilova, 2010).
represents an interesting outcome of this scarce research area. Additional insights can be extracted from the paper sample.
Shah (2005) reinforces that SCs in the process industries should Ecologic considerations should represent an integral part of batch
be managed under consideration of ecological and social factors. In scheduling problems (Arbiza et al., 2008). This need is substanti-
the chemical process industry, many sustainability issues arise in ated by the observed relevance of product sequencing and product
product development or process engineering (García-Serna et al., recipe choice for the environmental performance (Adonyi et al.,
2007). For several decades, “environmental criteria were pushed 2008; Berlin et al., 2007; Vaklieva-Bancheva and Kirilova, 2010).
by (...) legislative frameworks [while] the implementation of social In this context, the consideration of uncertainty and variability is
concerns within the chemical process industry is not that evident” of particular importance in batch manufacturing (Linninger et al.,
(García-Serna et al., 2007, p. 9). Thus, regulatory pressures increase 20 0 0). Moreover, the adequacy of Pareto analysis to study win-win
the need of a systematic integration of environmental factors into opportunities and tradeoff requirements between TBL objectives
decision support systems for chemical production planning. How- is emphasized in many papers (Brandenburg, 2015; Dietz et al.,
ever, in contrast to, e.g., the electronics industry or the agricultural 20 05; Hugo and Pistikopoulos, 20 05; Song et al., 2002). These
sector, SSCM models are less often applied for SC planning in the observations and guidelines support preparing and conducting the
chemical process industry (Brandenburg et al., 2014). study at hand.
A sample of 20 formal SSCM models for the chemical process
industry as listed in Table 1 confirms these observations. The col-
lection shows that a majority of models (see upper part of Table 1) 2.2. Production planning
is applied for sustainability analyses of the whole industry sector
or for strategic tasks. In contrast, models that support operational Campaign planning and scheduling in the process industry.
tasks of production planning and scheduling or process control Campaign planning and batch scheduling belong to the major
(see lower part of Table 1) are scant. The reflection of the TBL SC tasks in the chemical process industry (Shah, 2005) and are
located at the mid- to short-term planning level (Stadtler, 2005).

Please cite this article as: G.J. Hahn, M. Brandenburg, A sustainable aggregate production planning model for the chemical process
industry, Computers and Operations Research (2018), https://doi.org/10.1016/j.cor.2017.12.011
JID: CAOR
ARTICLE IN PRESS [m5G;January 17, 2018;17:22]

4 G.J. Hahn, M. Brandenburg / Computers and Operations Research 000 (2018) 1–15

Méndez et al. (2006) develop a taxonomy of related planning APP is presented by Mirzapour Al-e hashem et al. (2013). This
problems and modeling approaches. The authors classify batch approach reflects the variability of cross-site transportation lead
scheduling problems according to in total 13 different criteria times and the uncertainty of long-term demands. It is among the
which include process topology, size and processing time of first APP models that reflects aspects of green logistics but in
batches, assignment and connectivity of equipment, inventory contrast to our presented approach it does not link mid-term APP
storage policies as well as time and resource constraints. The mod- to short-term production scheduling.
eling approaches are categorized with regard to representation Discrete-event simulation (DES) and clearing function-based
of time and event, material balances and optimization criteria of approaches have been mostly proposed to overcome the afore-
the objective function. Complexity issues of process scheduling mentioned shortcomings by anticipating the non-linear be-
problems are assessed by Honkomp et al. (20 0 0). Their study havior of stochastic manufacturing systems (Armbruster and
reveals that the problem complexity is influenced by the problem Uzsoy, 2012). More recently, Almeder et al. (2009) and
size, i.e. the number of products, processes and processing units, Gansterer et al. (2014) discuss simulation-optimization frame-
as well as by time- and quantity-related storage constraints and works for hierarchical production planning. For an overview
furthermore non-productive, sequence-dependent activities and of clearing function-based approaches, we refer the reader
flexible crewing. Modeling features and solution approaches are to Pahl et al. (2007) and Missbauer and Uzsoy (2011).
categorized and thoroughly discussed by Kallrath (2002b). Selçuk et al. (2008) and Asmundsson et al. (2009) provide
Decision support tools for offline and online campaign planning production planning approaches for multi-level and multi-product
and scheduling include exact algorithms, heuristics and hybrid environments by extending basic clearing functions formu-
methods (see, e.g., papers in Günther and van Beek, 2003; Gün- lations as described in Graves (1986) or Karmarkar (1987).
ther and Meyr, 2009). MILP is a prominent modeling technique Jansen et al. (2013) consider SC master planning and apply a local
to formulate production planning and scheduling problems from smoothing algorithm to manage aggregate order release targets.
chemical process industry (see, e.g., Kallrath, 2002a; Velez and Current approaches put only limited focus on the issue of
Maravelias, 2014, for an in-depth introduction). Campaign planning lot-sizing/batching and its implications on the performance of
for multi-stage batch processes represents a specific requirement stochastic manufacturing systems (Missbauer and Uzsoy, 2011).
in the chemical process industry (see, e.g., Berning et al., 2004; Correspondingly, Kang et al. (2014) investigate a multi-product
20 02; Brandenburg and Tölle, 20 09, for application examples) but single-stage manufacturing environment. Söhner and
which, in contrast to short-term batch scheduling, covers a time Schneeweiss (1995) and Lambrecht et al. (1998) propose queuing
horizon of several weeks or even up to one year. In addition to network-based models to find lead time-optimal lot-sizes in multi-
deterministic approaches, there is a large body of literature that level multi-product job shop environments. Building on these
is concerned with short-term process scheduling under uncer- ideas, an integrated approach combining APP with queuing net-
tainty of processing times, equipment availability or customer works has been presented in Hahn et al. (2016) in the context of
demand (Li and Ierapetritou, 2008; Méndez et al., 2006; Shah, manufacturing outsourcing for a discrete manufacturing environ-
2005). Although efficient algorithms are developed to solve the ment. However, a corresponding hierarchically integrated approach
resulting stochastic optimization problems, most of the papers for APP that is capable of dealing with the specific requirements
omit capturing the impact of variability on the performance of the of the process industry is missing in the literature so far.
manufacturing system regarding lead times or throughput.
Numerous benchmarking problems are formulated to character- 2.3. Summarized findings
ize particularities of chemical production planning and scheduling
and to stimulate the development and comparison of solution The brief review presented in this section provides a termino-
approaches. Burkardt and Hatzl (20 05, 20 06) numerically assess logical basis and an introduction into the research field. It shows
four different chemical production processes, inter alia a complex that a broad range of decision support models for SC planning in
one described by Papageorgiou and Pantelides (1993) and a bench- the chemical process industry exists. However, this research area
marking problem which was prepared first by Westenberger and still emerges and exhibits two major gaps which we intend to
Kallrath (1994) and later published by Kallrath (2002b). address with this paper.
Hierarchically integrated planning and managing variability. Hi- First, a model-based approach to sustainable production
erarchically integrated approaches to SC planning are required to planning in the chemical process industry that integrates socio-
accommodate the interdependencies between the strategic, tactical ecological factors and thus reflects the entire TBL concept is
and operational planning levels (Stadtler, 2005). The hierarchical missing. Most current approaches to mid-term campaign planning
planning paradigm for deterministic production and supply chain in the chemical process industry omit the value potential to
planning has been introduced by Hax and Meal (1975) in their optimize capacity utilization under consideration of environmental
seminal work and extensively studied since then (Stadtler, 2005). or social criteria. We close this gap by providing an APP approach
In their review of production planning approaches under un- that covers all TBL dimensions. In this approach, the profit dimen-
certainty, Mula et al. (2006) find that simulation is the most sion is modeled by traditional cost minimization. In addition, the
frequently used approach to deal with complex production pro- proposed approach can handle alternative production routings to
cesses with multiple sources of uncertainty. reduce carbon emissions thereby reflecting the planet dimension.
The tactical level of APP needs to manage the trade-off between Moreover, the proposed model optimizes overtime decisions that
capacity utilization and aggregate lead times while anticipating affect employees’ working times as an objective for the people
the consequences of the mid-term production mix/volume and dimension. We compare two options to model the three TBL
capacity decisions on the operational level in stochastic manufac- dimensions in the objective function. One option is to internalize
turing systems (Pahl et al., 2007). However, most of the extant APP socio-ecologic factors into cost as single objective and the other
models assume constant throughput irrespective of the workload is to formulate a multi-criteria objective function that comprises
in the system and operational batching decisions. The same is not only cost, but also carbon emissions and overtime explicitly as
true concerning hierarchically integrated planning approaches for externalized objectives.
the process industries that also omit the non-linear behavior of Second, a hierarchically integrated approach is needed that
stochastic manufacturing environments (see, e.g., Das and Sen- is capable of anticipating the impact of mid-term decisions
gupta, 2009; Rutten, 1993). A stochastic model for green multi-site concerning production mix/volume and routing selection on op-

Please cite this article as: G.J. Hahn, M. Brandenburg, A sustainable aggregate production planning model for the chemical process
industry, Computers and Operations Research (2018), https://doi.org/10.1016/j.cor.2017.12.011
JID: CAOR
ARTICLE IN PRESS [m5G;January 17, 2018;17:22]

G.J. Hahn, M. Brandenburg / Computers and Operations Research 000 (2018) 1–15 5

erational performance in a stochastic manufacturing environment. order to allocate the capacity requirements to the respective time
However, variability due to stochastic operating times as well bucket and processing unit according to the targeted due date and
as equipment failures and induced system variability resulting the sequence of processing units in the production routing.
from the aggregation of several batches into larger production It should be noted that subsequent production steps are syn-
campaigns are important factors to be considered especially in chronized with regard to input and output quantities while the
the chemical process industry. Consequently, we close this gap synchronization of starting times of linked batches remains a
by providing a hierarchically integrated approach for aggregate detailed scheduling task that should be supported by other algo-
campaign planning in the chemical process industry that captures rithms (see e.g. Rager et al., 2015; Ribas et al., 2010). The APP and
the performance implications of manufacturing systems with the ASQN models are explained in greater detail in the following
multiple sources of stochasticity. subsection. A list of symbols is summarized in Section A.1 of the
appendix.
3. Decision support framework
3.2. Mathematical modeling
3.1. Overview and approach
The deterministic linear APP model covers a planning period
We develop a sustainable APP approach for the process indus- of one year and minimizes total relevant costs since customer
try that incorporates alternative production routings and supports demands must be fulfilled. We focus on the issue of multi-level
campaign planning in a stochastic multi-level manufacturing production and campaign planning at one manufacturing site.
environment. Social and ecological factors comprise overtime to However, the current approach could be easily extended towards
represent unfavorable working times from an employee’s point a manufacturing network with several sites. The decision variables
of view and production-related energy consumption that creates are the amount of overtime Ot per period t, the inventory Ipt and
carbon emissions. Hence, the suggested model complies with the backlog levels Bpt per (intermediate or finished) product p at the
TBL paradigm of sustainability. The socio-ecological aspects can end of period t, and the production quantities Xmzt per production
either be externalized and measured by independent indicators, routing m starting in period z and finishing in period t.
in particular overtime hours and carbon emission, or internalized Consequently, the objective function in (1) covers overtime
into the economic objective to reduce total costs. costs (with cost rate co), inventory and backlog cost per product
Stochasticities are reflected by variable durations of operations (cip , cbp ), production-related energy (cem ) and conversion (cxm )
processes and equipment failures. For this purpose, we deter- costs as well as WIP costs (cwm ) per production routing. WIP costs
mine aggregate campaign lead times according to production are calculated from aggregate lead times (ltmz ) per production
mix/volume and routing selection decisions under consideration of routing m and starting period z which are determined in the
stochastic setup and processing times as well as availability of pro- respective ASQN models.
duction equipment. Consequently, our approach allows for manag-
ing the trade-offs between capacity costs of overtime and interme- 
T 
T

diate and finished product inventories/backlogs, routing-dependent


min COST = co · Ot + (ci p · Ipt + cb p · B pt )
t=1 p∈P t=1
conversion and energy costs, and WIP costs according to aggregate
campaign schedules. More precisely, increasing campaign sizes 
T 
T

improves capacity utilization and reduces capacity costs with the + (cem + cxm ) · Xmzt
m∈M z=1 t=z
consequence that aggregate lead times prolong and WIP costs rise.
Modeling the aforementioned optimization problem in an inte- 
T 
T

grated approach would require a stochastic non-linear mathemat- + cwm · ltmz · Xmzt (1)
ical program. Building on Hahn et al. (2016), we thus decompose m∈M z=1 t=z

the problem hierarchically into a deterministic linear program for We examine a multi-level production process with alternative
APP as the top-level and multiple ASQN models (one per APP time production routings. Customer demands dpt for product p in
bucket) as the anticipated base-level to capture the non-linear period t (measured in tons) are considered for both intermediate
behavior of stochastic manufacturing systems. In doing this, our and finished products. Equation (2) covers the mass balances
approach follows the hierarchical planning paradigm using two for intermediate and finished products p. Since process industry
planning levels that are coordinated iteratively via instructions and manufacturing allows various production routings m to produce
feedback (Schneeweiss, 2003). More precisely, the APP model acts a specific product p, our approach determines the production
as a coordinating instance to capture inter-temporal dependencies. quantities Xmzt per production routing m with corresponding
The time-discrete APP model at the top-level covers a mid-term campaigns starting in period z and finishing in period t. The
planning period (typically a full seasonal cycle (Stadtler, 2005)) binary parameter γ mzt indicates (=1) which campaign size and
and is used to derive optimal capacity levels with respect to corresponding aggregate schedule have been determined by the
overtime as well as inventory/backlog levels of intermediate and ASQN model. The parameter β mp represents the conversion rate
finished products. Moreover, we obtain the production mix and from intermediate product p into production routing m for the
the corresponding production volumes per routing. respective superordinate product. In (3), we ensure that only
For each APP time bucket, we translate the production volumes customer orders are backlogged.
into inter-arrival times of batches which serve as instructions for
the respective (time-continuous) ASQN model. A separate ASQN  
t
I p,t−1 − B p,t−1 + γmzt · Xmzt
model per time bucket is used to comply with the steady state
m∈M p z=τ
assumption of the underlying queuing network approach given the
dynamic planning environment. From the ASQN models, we obtain  
T
− βmp · Xmt v − Ipt + B pt = d pt
campaign sizes and aggregate schedules per production routing
m∈M p v=t
as well as required capacity buffers to accommodate failures of
processing units and capacity losses due to setups which in turn
∀ p ∈ P; t = 1 . . . T (2)
serve as inputs for the APP model. For this purpose, the contin-
uous aggregate campaign schedules need to be discretized and in B pt − B p,t−1 ≤ d pt ∀ p ∈ P; t = 1 . . . T (3)

Please cite this article as: G.J. Hahn, M. Brandenburg, A sustainable aggregate production planning model for the chemical process
industry, Computers and Operations Research (2018), https://doi.org/10.1016/j.cor.2017.12.011
JID: CAOR
ARTICLE IN PRESS [m5G;January 17, 2018;17:22]

6 G.J. Hahn, M. Brandenburg / Computers and Operations Research 000 (2018) 1–15

The capacity constraint for each processing unit u and period t The objective function consists of three elements: first, the
is covered in (4). On the RHS of the equation (4), we have capacity waiting time until the campaign is closed depends on the routing-
supply given regular working time capX and additional overtime specific average inter-arrival times ym and average batch sizes bqm .
capacity Ot in period t. The parameter α denotes the availability of Expected inter-arrival times of batches per production routing are
processing unit u. The LHS of the equation (4) calculates capacity approximated from the aggregate production volumes in the APP
demand using the production quantity Xmzv and the capacity re- model and the corresponding gross capacity (in production hours)
quirements kxmru of production routing m on processing unit u for considering average batch sizes. The waiting times are weighted
operation r. The binary lead time offset coefficient δ mruzt is derived according to the relative importance of the respective production
from the ASQN models and indicates (=1) whether production routing using the corresponding inter-arrival rates λm for pro-
routing m requires capacity on processing unit u for operation r duction routing m. Second, the term E(wqu ) covers the queuing
in period t given the campaign size and the aggregate schedule time between consecutive manufacturing steps at processing
(starting period z and finishing period t) derived in the ASQN unit u and is calculated using an adapted version of the Kraemer
model. The parameter stut sum captures the cumulative capacity loss and Langenbach–Belz approximation (Kraemer and Langenbach-
due to setups at processing unit u in period t which is derived Belz, 1976) (for further details see Lambrecht et al., 1998). The last
from the ASQN models. element of the objective function covers setup (st) and processing
(pt) times according to campaign size and production routing;
 
t 
T
δmruzt · kxmru · Xmzv + stutsum ≤ αu · (capX + Ot ) the inter-arrival rates are used again for weighting the different
m∈M r∈Rm z=τ v=t production routings. The binary coefficient ωmru indicates (=1)
whether operation r on processing unit u is part of the production
∀u ∈ U; t = 1 . . . T (4)
routing m.
The constraint on storage capacity for product p is captured in Optimal campaign sizes per production routing m are de-
(5). Maximum overtime is respected due to (6). In (7), we require termined using a dedicated optimization routine described in
that only a fraction ν p of cumulative demand for product p can be Lambrecht et al. (1998) that considers given minimum and
backlogged beyond the planning horizon T. maximum campaign sizes. The obtained lead times need to be
converted into aggregate campaign schedules which serve as
I pt ≤ Imax
p ∀ p ∈ P; t = 1 . . . T (5)
feedback to the APP model. For this purpose, we apply a simple
forward scheduling approach along the production routing to
Ot ≤ Omax t = 1 . . . T (6) obtain aggregate schedules for the campaigns assuming earliest
possible starting times of the campaigns. More precisely, the

T
campaigns are discretized into time buckets of the APP model
B pT ≤ ν p · d pt ∀p ∈ P (7)
to obtain information about material availability (the binary
t=1
parameter γ ) and capacity consumption (the binary parameter δ ).
Initial inventories and backlogs are handed over in (8) and (9). Given production volumes and optimal campaign sizes per
Equation (10) covers production quantities that have started in production routing, we can determine the number of setups and
period τ before the planning period and will finish in period t = 1 subsequently derive the necessary capacity buffer for setups stut sum

or later. Non-negativity constraints are summarized in (11). to be considered in the APP model. The APP model and the
ASQN models are solved iteratively until both planning levels are
I p0 = I0p ∀p ∈ P (8)
coordinated. We assume coordination if the objective value of
the APP model only changes within a certain range between two
B p0 = B0p ∀p ∈ P (9) consecutive iterations of the algorithm. To ensure coordination, we
gradually reduce the interval between minimum and maximum
0
Xmzt = Xmzt ∀m ∈ M; z = τ . . . 0, t = z . . . T (10) campaign sizes in the ASQN models. A detailed description of the
iterative solution procedure can be found in Hahn et al. (2016).
I pt , B pt , Xmzt , Ot ≥ 0 ∀ p ∈ P; m ∈ M; z = τ . . . T , t = 0 . . . T (11) Having outlined the internalized approach, we need to define
the objective function of our approach for an externalized view
For the ASQN model, we consider a multi-product multi- on the TBL. For this purpose, we extend the mere focus on the
processing unit facility and determine optimal campaign sizes Qm profit dimension measured by costs of the optimized aggregate
per production routing m. A production routing contains an production plan (COST, see (1)), and additionally take into account
arbitrary but fixed sequence of processing units with operation- the planet dimension indicated by the amount of energy-related
specific setup and processing times. The objective function of the carbon emissions (EMIS, see (13)) and the people dimension
ASQN model in (12) minimizes average weighted campaign lead reflected by the overtime hours (OVER, see (14)). The parameter
time (further details see below). The lead times of the production eem denotes the emissions per ton for production on routing m.
routings are weighted in the objective function based on their
proportionate production volumes and the corresponding inter- 
T 
T
EMIS = eem · Xmzt (13)
arrival rates λm for production routing m (considering the average
m∈M z=1 t=z
batch size bqm ). By minimizing campaign lead times in each ASQN
model, we manage the cost of WIP inventories (invoking Little’s 
T
OVER = Ot (14)
Law) from an aggregate planning perspective (see Eq. (1)).
  t=1

λm · bqm Qm · bqm − 1 · ym  In the objective function (15), these three dimensions are
min  · + E (wqu ) weighted by non-negative parameters θ ∈ [0, 1] for each factor. For
m∈M λm · bqm 2 · bqm
Qp
m∈M u∈U
 the Pareto analysis, the weights sum up to 1. We normalize on the
 λm · bqm · ωmru value of the corresponding optimal internalized solution (marked
r∈Rm
+   with a superscript ∗ ) in order to reduce the impact of different
u∈u m∈M m∈M r∈Rm λm · bqm · ωmru
  scales of the three factors.
 λm · bqm · ωmru   COST EMIS
·  · st mr + Qm · bqm · pt mr (12) min θ COST · + θ CE ·
COST∗ EMIS∗
r∈Rm r∈Rm λm · bqm · ωmru
Please cite this article as: G.J. Hahn, M. Brandenburg, A sustainable aggregate production planning model for the chemical process
industry, Computers and Operations Research (2018), https://doi.org/10.1016/j.cor.2017.12.011
JID: CAOR
ARTICLE IN PRESS [m5G;January 17, 2018;17:22]

G.J. Hahn, M. Brandenburg / Computers and Operations Research 000 (2018) 1–15 7

Fig. 1. State-task-network of the chemical processes based on Papageorgiou and Pantelides (1993, p. 30) and Burkardt and Hatzl (2006, p. 1178).

  OVER 4. Numerical analysis


+ 1 − θ COST − θ EMIS · (15)
OVER∗
4.1. Case example

The proposed planning approach is numerically analyzed at a


case example for campaign planning and scheduling at a chemicals
manufacturer (see Fig. 1) which is introduced by Papageorgiou and
Pantelides (1993). We resort to this case example to avoid prob-
lems with regard to data confidentiality. This considered case
In the numerical study, optimal internalized solutions make example encompasses numerous characteristics of planning and
use of overtime (i.e. OVER∗ > 0). Hence, dividing by zero in (15) is scheduling in the process industry (see, e.g., Honkomp et al., 20 0 0;
not an issue. It should be noted that OVER∗ could be replaced Kallrath, 2002b; Méndez et al., 2006, for a taxonomy) including
by a positive parameter OVER > 0 if OVER∗ = 0, i.e. if the optimal batch processes with variable sizes and fixed, unit-independent
internalized solution does not make use of overtime. Alternatively, processing times that are operated on multi-purpose production
the term (1 − θ COST − θ EMIS ) · OVER
OVER
∗ could be removed from the units with full connectivity in multi-stage production.
objective function (15) and OVER = 0 could be added as a new The processing equipment consists of 8 multi-purpose units (u1,
constraint of the model in order to ensure that overtime remains . . . , u8) with different capacities on which 13 processes are oper-
unused in any solution of the Pareto analysis.

Please cite this article as: G.J. Hahn, M. Brandenburg, A sustainable aggregate production planning model for the chemical process
industry, Computers and Operations Research (2018), https://doi.org/10.1016/j.cor.2017.12.011
JID: CAOR
ARTICLE IN PRESS [m5G;January 17, 2018;17:22]

8 G.J. Hahn, M. Brandenburg / Computers and Operations Research 000 (2018) 1–15

ated to manufacture three finished products (A, B, C) from three Overtime Production
Intermediate
stage Final
raw materials and ten intermediate products. The process-unit-
Inventory/backlog (tons) Overtime (hrs)
assignment results in a total of 23 different tasks, each of them
representing a unique production mode. Process durations are 1,400 1,303 400
task-dependent, i.e., they depend on the particular process and on 1,200 1,174
the chosen processing unit. Storage constraints apply with regard 300
1,000
to quantity (unlimited (UIS), finite (FIS) or no inventory storage 818
(NIS)) and time (unlimited (UW), finite (FW) or zero wait (ZW) 800
660 200
storage durations). Taking into account that some buffer capacities 600 576
are limited and that zero wait conditions apply for some processes, 459
400 375
323
the tasks can be aggregated to in total five different stages, each 100
comprising two or three processes. As illustrated in Fig. 1, in total 200 129
55
28 different production routings have to be considered. 0 0
0 0
Fig. 1 illustrates the chemical processes of the case example as 1 2 3 4 5 6 7 8 9 10 11 12
a state-task-network (STN), in which ovals represent states and the Period
numbers in parentheses inform about initial and maximum inven-
tory levels. Furthermore, rectangles represent tasks for the con- Fig. 2. Inventory/backlog quantities and overtime hours per period.
version of states, contain data for process duration, minimum and
maximum batch size (numbers in parentheses) and inform about Net primary Production Intermediate
the production unit (u1, . . . , u8) on which the task is operated. demand stage Final

Dotted lines encircle the tasks that are grouped to the same stage. Production volume (tons) Demand (tons)
To test the proposed planning approach at this case example, fi- 2,500 1,200
nancial, operational and environmental figures have to be defined. 2,033 2,020
The sustainability factors are reflected by unit-specific differences 2,000
900
in carbon emissions and by overtime costs that reflect workers’
1,473
preferences to work during regular shifts. To ensure scientific 1,500
1,148 600
rigor, realistic magnitudes of the additional numerical factors are
1,000 882
derived from related literature. A comprehensive description and
lists of all planning parameters is given by Brandenburg and Hahn. 575 568 300
500 452
Under consideration of minimum batch sizes and maximum 189
campaign sizes, the available capacities have to be balanced with 0 0 0
0 0
customer demands over a planning horizon of 12 standard months
1 2 3 4 5 6 7 8 9 10 11 12
that consist of four weeks covering a calender year and a full Period
seasonal cycle. The weekly base capacity of 80 hours reflects that
the production equipment is operated over five days with two Fig. 3. Production volumes per stage and time bucket (starting period view).
shifts of eight hours per day and can be extended by overtime to
as much as 168 hours per week.
In the case example, we focus on the variability of setup and Investigating the corresponding production levels (see Fig. 3),
processing times, and we also consider stochastic equipment avail- we largely see fluctuating production levels which are the con-
ability. Variability in seasonally-adjusted demand or concerning sequence of demand seasonality and comparably large campaign
yields of manufacturing processes could be subject of further sizes to fully utilize available capacity. Moreover, overtime is
investigations which are beyond the scope of this paper. systematically considered for filling up campaigns in order to
optimize capacity utilization. Given the average lead time of
campaigns of around 0.9–1.6 time buckets, we clearly see the
4.2. Results effect on inventory coverage and the implications for net demand
as depicted in Fig. 3. A second observation is that the approach
Base scenario. As a first step, we investigate the base scenario uses decoupling of production stages in order to manage interme-
using the cost-based internalized approach. We obtain an objective diate and final products differently. More precisely, intermediate
function value of 3,222.5k currency units (CU) with the following products are made to stock in advance to benefit from larger
split along the cost categories (in k CU): Inventory 54.8 (2%), over- campaign sizes (although there is also a considerable share of
time 319.8 (10%), backlog 0 (0%), energy 691.9 (21%), conversion primary customer demands at the intermediate level) while final
2,087.6 (65%), WIP 68.4 (2%). We observe that routing-dependent products are mostly made to order avoiding preproduction.
conversion and energy costs account for more than 86% of the Analyzing the decisions on production routing selection and
relevant costs. This underlines the relevance of routing selection volume allocation allows insight into the drivers of energy, con-
decisions for aggregate planning in the chemical process industry. version, and WIP costs. We find that the optimal production plan
Although inventory/backlog and overtime costs only account for utilizes the cost-efficient routings (lowest energy and conversion
12% of total costs, the underlying decisions reveal some interesting cost per ton) per production stage in a range of 68% and 100%.
results with respect to mid-term capacity management (see Fig. 2). Switching to less cost-efficient routings is due to capacity restric-
One can see that overtime (beyond regular production hours) is tions and campaign sizes since the routings partially share the
only used in distinct situations (periods 1 and 5) in order to same production equipment. In contrast to the production stages
considerably increase the available capacity by additional 112% and 1, 2, and 4, the stages 3 and 5 involve more than two alternative
43%, respectively. While there are no backlogs at the final pro- routings. However, this additional flexibility is used only to a very
duction stage and only comparably small inventories, the model limited extent. Even for stage 5, which offers 18 different routing
largely builds up inventories at the intermediate stage to exploit options, the production plan suggests only two preferred routings
large campaign sizes. Inventory built-up corresponds to demand to manufacture 79% of the total production volume and another
seasonality with a peak in the middle of the planning period. four routings for the remaining 21% of the volume.

Please cite this article as: G.J. Hahn, M. Brandenburg, A sustainable aggregate production planning model for the chemical process
industry, Computers and Operations Research (2018), https://doi.org/10.1016/j.cor.2017.12.011
JID: CAOR
ARTICLE IN PRESS [m5G;January 17, 2018;17:22]

G.J. Hahn, M. Brandenburg / Computers and Operations Research 000 (2018) 1–15 9

Cost-efficient Non-cost efficient

# of routing Lead times*


Routing mix alternatives (periods)
(in Percent) used / feasible Min / Avg / Max

1 83% 17% 2/2 0.6 / 1.3 / 1.8

2 68% 32% 2/2 0.4 / 0.9 / 2.2

Production
3 77% 23% 2/4 0.4 / 1.0 / 1.2
stages

4 100% 0% 2/2 0.5 / 1.4 / 2.2

5 79% 21% 6 / 18 0.6 / 1.6 / 1.8

* Trimmed 90% interval

Fig. 4. Volume split and aggregate lead times per production stage.

Table 2 and periods. The results are summarized in Table 3. We observe


Parameter ranges of the sensitivity analysis.
that increasing demand seasonality, equipment unavailability, and
Parameter Range processing time variability lead to a consolidation of production
Demand seasonality 0.1, 0.2, 0.3, 0.4, 0.5 volumes on the most cost-efficient routings while the results for
Equipment unavailability (1 − α ) 0.00, 0.02, 0.04, 0.06, 0.08 increasing setup time variability point into the opposite direction.
SCV process time, scaling factor 0.5, 1.0, 1.5, 2.0, 2.5 From these results, we conclude that higher levels of variation can
SCV setup time, scaling factor 0.5, 1.0, 1.5, 2.0, 2.5 be compensated by other capacity-related measures such as in-
Overtime cost (co), in CU 512, 576, 640, 704, 768
ventory/backlog and overtime. However, the impact of high setup
Energy cost (ce), scaling factor 0.50, 0.75, 1.00, 1.25, 1.50
Backlog cost (cb), scaling factor 0.60, 0.75, 1.00, 1.50, 3.00 time variability can only be offset by splitting volumes across
different less cost-efficient production routings. Corresponding
effects are obtained concerning the lead times. We find longer
campaign lead times (except for the case of increasing setup time
Although WIP costs only account for 2% of total costs, the variability) which is a consequence of additional (less) campaigns
underlying lead times have considerable implications on aggre- on the major production routings. The peculiarity of decreasing
gate planning with respect to capacity allocation and due date campaign lead times for higher levels of demand seasonality can
reliability towards the customer. While the average weighted be explained by the fact that high levels of variation in both
lead times across the production stages already span a range of directions allow for capacity leveling using preproduction. This in
0.9–1.6 periods, we even observe substantial fluctuations in the turn results in shorter average lead times.
lead times per production stage over the planning period, e.g., Investigating the effects of changing cost levels, we find that
0.6–1.8 for stage 1 or 0.5–2.2 for stage 4. These fluctuations can increasing cost rates only have minor or no effects on the loading
result in excess inventories or unnecessary backlogs if wrongly of production routings, but such cost increases result in a substan-
approximated in the planning approach. These findings clearly tial increase in average lead times. Summarizing our findings, we
support our approach to determining lead times endogenously. conclude that the variation of operational parameters influences
The results are summarized in Fig. 4. the loading of production routings and lead times while cost rates
Sensitivity analysis. As the second part of our numerical inves- only impact on lead times.
tigations, we conduct sensitivity analyses on cost and non-cost Pareto analysis. The first two analyses of our numerical study
parameters. For this purpose, we vary demand seasonality, equip- consider the internalized (cost-based) approach. Following the TBL
ment unavailability, and the squared coefficients of variation (SCV) concept, we now focus on the externalized perspective that in-
of setup and processing times and inspect different levels of corporates carbon emissions (EMIS) and overtime (OVER) working
overtime and energy cost. For the non-cost parameters, we thus hours into the optimization model. Using the objective function
focus on the key demand-side and operations parameters that in (15), we conduct a Pareto analysis in order to illustrate the
are subject to uncertainty and stochasticity as well as the cost potential to leverage win-win opportunities between conflicting
parameters that relate most closely to the social and ecological TBL objectives and, thus, to exemplify that socio-ecologic factors
dimension of the TBL. We also included backlog cost to further may be improved without major losses of economic performance.
study its impact since we observed no backlogs in the base case as The weights are selected from the interval of 0–1 in steps of
presented above. The parameter ranges are summarized in Table 2 0.1 and sum up to 1. A minimum cost weight of 0.1 is assumed
with the values from the base scenario highlighted in boldface. to ensure that the fundamental trade-off in mid-term planning
The backlog cost correspond to fill rates in the range of 97.5% and between capacity utilization and WIP inventories is resolved. In
99.5% in steps of 0.5%. total, 54 optimization runs are executed and carefully evaluated
In the analysis, we focus on capacity efficiency and aggregate for the Pareto analysis. The results are visualized in Fig. 5(a) with
lead time as the two fundamental operations metrics at the mid- the three dimensions normalized on the values of the optimal
term planning level. Two indicators are used for this purpose. First, internalized solution.
the fraction of the production volume which is manufactured on Summarizing the results that are listed in Table 4, we observe
production routings with the highest cost-efficiency, and second, that OVER varies in a larger range of 0–1.90 compared to EMIS
the weighted average lead time across all production routings

Please cite this article as: G.J. Hahn, M. Brandenburg, A sustainable aggregate production planning model for the chemical process
industry, Computers and Operations Research (2018), https://doi.org/10.1016/j.cor.2017.12.011
JID: CAOR
ARTICLE IN PRESS [m5G;January 17, 2018;17:22]

10 G.J. Hahn, M. Brandenburg / Computers and Operations Research 000 (2018) 1–15

Table 3
Parameter ranges of the sensitivity analysis.

Volume on cost-efficient routings Weighted lead time

Parameter Range (in Percent) Direction of effect Range (in Periods) Direction of effect

Demand seasonality 78–97 ↑ 0.97–1.29 


Equipment unavailability 85–100 ↑ 1.16–1.29 
SCV process time 85–95 ↑ 1.14–1.29 ↑
SCV setup time 84–94 ↓ 1.04–1.29 ↓
Overtime cost 85–96 1.12–1.29 ↑
Backlog cost 85–95  1.12–1.29 ↑
Energy cost 69–95 →a 0.85–1.29 ↑a
a
Corrected for one outlier instance

(a) Three dimensional Pareto cube. (b) Pareto analysis of emission and overtime.

(c) Pareto analysis of cost and overtime. (d) Pareto analysis of cost and emission.
Fig. 5. Pareto analysis of the TBL criteria.

Table 4 putting more emphasis on reductions of the two non-financial


Minimum, average, and maximum factors of the Pareto analysis.
sustainability criteria OVER and EMIS. Examining the instances
Factor in greater detail, we find that environmental improvements are
Dimension Min Avg Max achieved in 45 of the 54 optimization runs while the social
performance is improved in every second optimization run. Simul-
COST 1.00 1.16 1.50
EMIS 0.94 0.97 1.06
taneous improvements of both the ecological and social bottom
OVER 0.00 1.02 1.90 line are obtained in 18 optimization runs. From these findings,
we conclude that the multiple trade-offs in the TBL need to be
reflected carefully. Especially in this regard, our proposed plan-
with a range of 0.94–1.06. Consequently, OVER reacts considerably ning approach provides appropriate decision support and thus is a
more sensitive to changes in management priorities than EMIS. valuable tool to manage production processes in a sustainable way.
Furthermore, we learn that costs can increase by up to 50% when

Please cite this article as: G.J. Hahn, M. Brandenburg, A sustainable aggregate production planning model for the chemical process
industry, Computers and Operations Research (2018), https://doi.org/10.1016/j.cor.2017.12.011
JID: CAOR
ARTICLE IN PRESS [m5G;January 17, 2018;17:22]

G.J. Hahn, M. Brandenburg / Computers and Operations Research 000 (2018) 1–15 11

The results of the Pareto analysis, as illustrated in Fig. 5 and into halves without compromising on the economic efficiency
listed in Table A.5 in the appendix of the paper, reveal insights into (COST increase of 2%); however, this improvement is obtained at
complementarity of and conflicts between the TBL objectives.2 the expense of strongly increasing carbon emissions by 6% (see
They show that it is not possible to improve all three factors optimization run OR53). The dashed line in Fig. 5(c) illustrates
simultaneously in comparison to the base case. It is to mention that additional OVER reduction can only be obtained at a COST
that the base case is obtained by a cost-focused optimization and increase of at least 9% (see optimization run OR35), but going
thus represents a lower bound for COST in the Pareto analysis. along with emission reduction of 3%. Additional OVER reduction is
Hence, a win-win-win solution that optimizes all three factors as only obtained when COST is tremendously increased by 50% (see
conceptualized by Seuring and Müller (2008) would be detected if optimization run OR12). These observations highlight that consid-
both non-financial sustainability factors are improved without cost erable self-financing social improvements can only be obtained at
increase. It can be seen from Table A.5 that such results are not the expense of higher pollution.
obtained in any of the 54 optimization runs. However, the results Third, we assess the coherence of economic and ecologic fac-
exemplify the possibility to achieve moderate carbon emission tors and set the weight for OVER to zero. The solid line in Fig. 5(d)
reductions of 2% and strong overtime reductions of more than one illustrates the possibility to achieve EMIS reductions of 3% while
third at comparably low cost increases of about 3% (optimization keeping the COST efficiency nearly constant (see optimization
run OR31). In this optimization run, all three sustainability criteria run OR54). However, this improvement is achieved by increasing
are weighted almost equally in the objective function. Further OVER by one third. Furthermore, comparable EMIS reductions of
reductions of one of the two non-financial factors without com- at least 3% obtained by small COST increases of at most 5% always
promising on the other one are only obtained at a cost increase result in additional overtime requirements (OVER increases by at
of 9% (see optimization run OR35) or more (see optimization runs least 15%, see optimization runs OR39, OR45, OR48, and OR51).
OR4, OR12, OR15, OR16, OR22, OR24). The dashed lines in Fig. 5(d) indicate that the lowest EMIS level
A comparison of the gaps to ideal, i.e., the Euclidean distance can be achieved either at comparable moderate COST increases
between the ideal point (1.0 0, 0.94, 0.0 0) and the triple of values of 6% (optimization run OR19) or at strong COST increases of 30%
for COST, EMIS and OVER obtained by the respective optimization (optimization run OR10). Each of these optimization runs increases
run, reveals two interesting insights. First, very strong overtime OVER by about two thirds. These insights suggest that (nearly)
reductions of more than 80% with considerably lowered carbon self-financing EMIS reduction deteriorates the social performance.
emissions (3% reduction) are enabled by slight cost increases of However, the lowest EMIS can be achieved while reducing OVER
less than 10% (see optimization run OR35). In the optimization by one third at comparably moderate COST increases of at most
run OR35, COST and OVER are equally weighted while EMIS is 15% (see optimization runs OR17 or OR24), if positive weights are
omitted. Second, stronger reductions of both non-financial factors assigned to all three criteria in the objective function.
are only obtained for stronger cost increases of more than 50% Two major conclusions are drawn from these Pareto analyses:
(see optimization runs OR12) resulting from strongly weighted first, the two-dimensional Pareto analyses exemplify that one fac-
OVER and nearly equally weighted COST and EMIS factors. tor can be improved without considerably deteriorating a second
The results of two-dimensional Pareto analyses as illustrated in one if two factors are reflected in the objective function. However,
Fig. 5(b)–(d) inform about the complementarity of or the conflicts in such cases the third factor is considerably deteriorated. Hence,
between two TBL criteria. Observations and interpretations of it is recommended to assign balanced weights to all criteria in
these results are complemented by information on the third factor the objective function. In doing this, considerable reductions of
as listed in Table A.5. First, we focus on the interplay of the two both non-financial factors can be obtained at low or moderate
non-financial sustainability criteria EMIS and OVER and assign a cost increases (as exemplified by optimization runs OR23, OR24
weight of 0.1 to COST. The solid lines in Fig. 5(b) indicates that and OR31). Second, a comprehensive three-dimensional Pareto
major reductions of both OVER (at least 15%) and EMIS (at least analysis can be conducted to find a solution that is closest to
5%) are possible (see optimization runs OR4 and OR8). However, the ideal point. This approach also leads to improvements of
these improvements are obtained at the expense of comparably both non-financial criteria without greater economic losses (as
strong increases of COST (at least 22%) and with highly imbalanced illustrated by optimization run OR35).
weights for the two non-financial factors. Optimization run OR24
exemplifies that similar improvements of both non-financial fac- 4.3. Implementation issues
tors can be obtained at a comparably moderate COST increase of
10% when the weights for all three criteria are better balanced. The Aggregate planning is performed using IBM ILOG OPL with
dashed line in Fig. 5(b) shows that OVER can be reduced by more CPLEX v12.6.0.0 for the APP model and a steepest descent al-
than one third while keeping the EMIS level nearly unchanged gorithm implemented in C++ for the ASQN model. The iterative
(see optimization run OR1). This improvement is obtained at a less solution procedure was run on an Intel Core i5 machine with
strong COST increase of about 10% with a high weight for OVER 2.6 GHz per core and 8GB RAM applying a threshold value of 0.5%
and a weight of zero for EMIS. Improvements of both non-financial to check for coordination, i.e., the algorithm terminates as soon
sustainability criteria can be obtained at COST increases of at as the relative deviation of two consecutive objective function
most 10% (see optimization runs OR22, OR31, OR35, OR37). These values falls below the given threshold value. For the base case, the
insights suggest that the two non-financial factors of the TBL can algorithm terminates after six iterations and requires 183 seconds.
be improved simultaneously at moderate cost increases. While the execution of the APP and ASQN models requires only a
Second, we elaborate on the socio-economic interface of sus- few seconds, the data consolidation and transformation between
tainability and set the EMIS weight to zero. The solid line in the calculation steps accounts for most of the computation time.
Fig. 5(c) indicates the possibility to cut the OVER requirements In the sensitivity analysis, the number of iterations is restricted
to 20 which implies that we gradually reduce the range of mini-
mum and maximum campaign sizes in the ASQN models after it-
2
It is to be noted that - as explained on p. Section 3.2 in context to the itera- eration 10 to ensure convergence of the algorithm. On average, the
tive APP-ASQN coordination - the planning approach generates optimized solutions
but it does not necessarily solve the problem to optimality. Thus, Fig. 5 depicts the
algorithm requires 15.8 iterations to reach a sufficient level of con-
optimized solutions obtained from the different optimization runs but it does not vergence with a minimum of 3 and a maximum of 20 iterations.
show Pareto frontiers. For the Pareto analyses, we impose the additional requirement that

Please cite this article as: G.J. Hahn, M. Brandenburg, A sustainable aggregate production planning model for the chemical process
industry, Computers and Operations Research (2018), https://doi.org/10.1016/j.cor.2017.12.011
JID: CAOR
ARTICLE IN PRESS [m5G;January 17, 2018;17:22]

12 G.J. Hahn, M. Brandenburg / Computers and Operations Research 000 (2018) 1–15

e.g. a frozen planning horizon, have to be taken into account.


Moreover, the links to mid-term master planning, purchasing
planning and distribution planning need to be considered.

5.2. Research contribution

The research contribution of the study at hand is related to


the two observed major gaps in related research (see Section 2.3)
which justify the need for the conducted study. A brief discussion
of findings addresses these gaps and helps positioning the paper
Fig. 6. Software modules covering the SC planning matrix (Meyr et al., 2002, p. 99 at hand into extant literature.
and Stadtler, 2005, p. 579). First, the presented model-based approach to sustainable
production planning in the chemical process industry integrates
socio-ecological factors and, thus, represents a holistic approach
solutions need to be internally consistent with the Pareto principle.
that reflects the entire TBL concept. Hence, it helps bridging the
If this requirement is violated, we further reduce the threshold
gaps of holistic SSCM models for chemical production planning (as
value for convergence and conduct additional iterations until
illustrated by Table 1 in Section 2.1) and of formal models that
both criteria are met. On average, the algorithm stops after 16.1
reflect social factors (as detected by, e.g., Brandenburg et al., 2014;
iterations with a minimum of 4 and a maximum of 20 iterations.
Brandenburg and Rebs, 2015; Seuring, 2013). The study at hand
reflects the social dimension of sustainability by the overtime
5. Critical reflection
work of workforce that operates the SC. This approach is in line
with findings of Northrup et al. (1979) and Vogel et al. (2012). The
5.1. Generalized findings and practical implications
presented modeling approach is suitable to reflect socio-ecologic
criteria as internalized factors as well as externalized ones and,
The numerical study analyses a specific problem instance for
thus, complements models in which sustainability criteria are
a particular case example and results will probably differ when
internalized as cost objectives or measured by aggregate indices
problem settings such as demand or capacity availability change
as revealed by Boukherroub et al. (2015). Win-win-situations
or when a different case example is assessed. Thus, obtaining
and tradeoffs between the three TBL objectives as conceptualized
universally valid findings or general insights is hardly possible.
and verbally described by, e.g., Seuring (2013) or Seuring and
However, the results of the presented numerical analysis can be
Müller (2008) are numerically analyzed in the paper at hand.
interpreted in a general way to derive managerial insights and
Second, the approach at hand represents a hierarchically inte-
practical implications for planners and other decision-makers.
grated one which is capable of anticipating the impact of mid-term
The case example illustrates that considering the variability
decisions concerning production mix/volume and routing selec-
of operational parameters in aggregate production planning for a
tion on operational performance in a stochastic manufacturing
mid-term horizon helps achieving better results. Reducing the pa-
environment. More specifically, we endogenize the determination
rameter variability can considerably effect the production cost, as
of campaign lead times (and corresponding WIP inventories) and
shown in the sensitivity analysis. This finding may motivate prac-
consider the impact of variability originating from various sources.
titioners to strive for stable production processes and operational
In doing this, we go beyond extant approaches for short-term
activities which, thus, are better predictable.
process scheduling under uncertainty (see, e.g. Li and Ierapetritou,
The Pareto analysis exemplifies that win-win potential can be
2008; Méndez et al., 2006). From a methodological point of view,
realized to a certain extent even if sustainability objectives are
we provide an alternative approach to solving stochastic non-linear
conflicting. As a consequence, it is recommended to formulate
production planning problems in the chemical process industry –
socio-ecologic criteria explicitly as externalized criteria in an
in contrast to simulation-optimization and clearing function-based
objective function rather than internalizing them into a single
approaches (see Missbauer and Uzsoy, 2011) – by combining
cost objective. Conducting a Pareto analysis under consideration
deterministic linear programming with stochastic modeling based
of the distance to the ideal point may be useful in case that the
on queuing theory.
determination of adequate weights for each sustainability objective
is difficult.
The described solution approach for sustainable aggregate 6. Conclusion and outlook
production planning can extend advanced planning systems (APS)
which represent state-of-the-art IT tools for chemical production A novel decision support framework for sustainable planning
planning as well. The modules of such APS as conceptualized by of production campaigns in the chemical process industry is
Meyr et al. (2002) and described by Stadtler (2005) are depicted in presented. In contrast to previous work in this field, we pro-
Fig. 6. Possibilities and potentials of APS extensions are discussed vide a holistic TBL approach that explicitly accommodates the
in, e.g., Günther and van Beek (2003) or Günther and Meyr (2009). socio-ecological dimensions. Our approach sheds light on the
Following the framework depicted in Fig. 6, the approach de- benefit of selecting alternative production routings which can be
scribed in the paper at hand can be implemented as an additional used to improve capacity utilization while reducing production-
component of the module for mid-term production planning. related carbon emissions or unwanted overtime working hours.
Considering operational variability in manufacturing systems and As a second contribution, we examined the interdependencies
anticipating the implications at the mid-term planning level could of campaign planning and mid-term APP. For this purpose, a
help to resolve the limitations of current APS due to their de- queuing network-based approaches was introduced and combined
terministic approach (Hahn and Packowski, 2015; Stadtler, 2005). with a classical APP model to anticipate the impact of mid-term
As a consequence, the interface to short-term scheduling is of production mix/volume and routing selection decisions on the
particular importance. Results of the optimization runs need to be operational level in a stochastic manufacturing environment. This
propagated to short-term scheduling. Vice-versa, constraints and allows for a proactive approach to manage variability which is an
frame conditions that result from existing short-term schedules, important concern in the chemical industry.

Please cite this article as: G.J. Hahn, M. Brandenburg, A sustainable aggregate production planning model for the chemical process
industry, Computers and Operations Research (2018), https://doi.org/10.1016/j.cor.2017.12.011
JID: CAOR
ARTICLE IN PRESS [m5G;January 17, 2018;17:22]

G.J. Hahn, M. Brandenburg / Computers and Operations Research 000 (2018) 1–15 13

The planning approach presented in this paper is numerically gave when the authors presented first results of the study at hand
assessed at a case example from chemical process industry. The at a scientific conference.
managerial insights from the numerical study are threefold and
can be summarized as follows. First, the decoupling of inter- Appendix A
mediate and final production stages allows to pursue different
strategies with respect to capacity utilization and campaign sizing. A1. Parameters and variables
The theoretical flexibility of alternative production routings is used
only to a limited extent. Considerable fluctuations of campaign Sets and Indices
sizes and corresponding schedules over the planning period and p∈P Set of products p
across production stages support our approach of determining m ∈ Mp Set of alternative production routings m for product p
lead times endogenously. Second, the sensitivity analyses reveal r ∈ Rm Set of operations r in production routing m
u∈U Set of processing units u
that cost and non-cost parameters have different implications t, z Time periods
on capacity loading and aggregate campaign schedules at the
mid-term level. Third, even if social and environmental factors can Parameters
be internalized as cost-related targets, it is beneficial to reflect αu Availability of processing unit u
βmp Direct demand coefficient of product p for production routing m
these criteria explicitly in the objective function. This provides a
γmzt Binary lead time offset coefficient for production routing m for
more comprehensive view on related trade-offs and supports the production starting in period z and finishing in period t
decision-maker with additional information regarding robustness δmruzt Binary lead time offset coefficient for production routing m with
of solutions given different managerial priorities. operation r on processing unit u in period t with production
starting in period z
There are several avenues for further research building on
νp Maximum faction of cumulative demand that can be backlogged
our approach. First, the developed decision support framework beyond the planning horizon
could be tested and further validated in settings with more com- λm Arrival rate of orders for production routing m (1/ym )
plex material flows, e.g. the case example of Westenberger and τ Maximum offset period
Kallrath (1994) or a real-life large scale application. Second, the ωmru Binary parameter indicating whether operation r of production
routing m is performed at processing unit u
functional focus on campaign planning in a single plant could
bqm Average batch size on production routing m
be extended to include cross-functional aspects such as synchro- capX Standard working time
nization of manufacturing and distribution processes or to cover co Cost of overtime
a network of several production facilities. Third, the mid-term ci p Cost of inventory for product p
campaign planning model can be linked to short-term campaign cem Cost of energy for production routing m
cxm Cost of conversion for production routing m
scheduling procedures that take into account sequence depen- cwm Cost of holding WIP for production routing m
dency of production processes. Fourth, various environmental d pt Demand for product p in period t
factors or different causes of carbon emissions could be evaluated Imax
p Maximum allowable inventory of product p
in greater detail. In this context, production leveling to reduce kxmru Capacity requirement for production routing m and operation r at
processing unit u
energy consumption could be considered as an optimization ob-
ltmz Lead time for production routing m starting in period z
jective. Overall, it can be concluded that the presented approach Omax Maximum allowable overtime usage
enriches the prevailing research area of sustainable production ptmr Average processing time of operation r for production mode m
planning and stimulates future studies in this emerging field. stmr Average setup time of operation r for production mode m
sum
stut Cumulative setup time losses on processing unit u in period t
T Planning horizon
Acknowledgment
wqu Average queuing time before processing unit u
ym Average inter-arrival time of batches in production mode m
The authors would like to thank the guest editors of this special
issue and three anonymous reviewers for their encouraging com- Decision variables
B pt Backlog of product p at the end of period t
ments and constructive criticism that helped improving the qual-
Ot Amount of overtime used in time period t
ity of the manuscript. Moreover, the authors would like to express I pt Inventory of product p at the end of period t
their sincere thanks to Rainer Leisten for the fruitful remarks he Qm Campaign size for production routing m
Xmzt Quantity for production routing m starting in period z and
finishing in period t

Please cite this article as: G.J. Hahn, M. Brandenburg, A sustainable aggregate production planning model for the chemical process
industry, Computers and Operations Research (2018), https://doi.org/10.1016/j.cor.2017.12.011
JID: CAOR
ARTICLE IN PRESS [m5G;January 17, 2018;17:22]

14 G.J. Hahn, M. Brandenburg / Computers and Operations Research 000 (2018) 1–15

A2. Results of the Pareto analysis Armbruster, D., Uzsoy, R., 2012. Continuous dynamic models, clearing functions,
and discrete-event simulation in aggregate production planning. In: Mirchan-
dani, P.B. (Ed.), TutORials in Operations Research: New directions in informatics,
Table A.5 contains the results of the Pareto analysis. optimization, logistics, and production. INFORMS, Hanover, pp. 103–126.
Asmundsson, J., Rardin, R.L., Turkseven, C.H., Uzsoy, R., 2009. Production planning
with resources subject to congestion. Nav. Res. Logist. 56, 142–157.
Table A.5 Berlin, J., Sonesson, U., Tillmann, A.-M., 2007. A life cycle based method to minimise
environmental impact of dairy production through product sequencing. J. Clean.
Results of the Pareto analysis.
Prod. 15, 347–356.
# θ COST θ EMIS θ OVER COST
COST∗
EMIS
EMIS∗
OVER
OVER∗
Gap to ideala Berning, G., Brandenburg, M., Gürsoy, K., Kussi, J.S., Mehta, V., Tölle, F.-J., 2004. In-
tegrated collaborative planning and supply chain optimization for the chemical
OR01 0.1 0.0 0.9 1.11 1.01 0.63 0.642 process industry (i) methodology. Comput. Chem. Eng. 28, 913–927.
OR02 0.1 0.1 0.8 1.04 0.98 0.92 0.920 Berning, G., Brandenburg, M., Gürsoy, K., Mehta, V., Tölle, F.-J., 2002. An integrated
OR03 0.1 0.2 0.7 1.32 0.95 1.27 1.310 system solution for supply chain optimization in the chemical process industry.
OR04 0.1 0.3 0.6 1.22 0.95 0.63 0.666 OR Spectrum 24, 371–401.
OR05 0.1 0.4 0.5 1.32 0.95 1.27 1.309 Boukherroub, T., Ruiz, A., Guinet, A., Fondrevelle, J., 2015. An integrated approach
OR06 0.1 0.5 0.4 1.23 0.95 1.27 1.291 for sustainable supply chain planning. Comput. Oper. Res. 54, 180–194.
OR07 0.1 0.6 0.3 1.50 0.95 1.27 1.365 Brandenburg, M., 2015. Low carbon supply chain configuration for a new product a
goal programming approach. Int. J. Prod. Res. 53, 6588–6610.
OR08 0.1 0.7 0.2 1.24 0.94 0.84 0.871
Brandenburg, M., Govindan, K., Sarkis, J., Seuring, S., 2014. Quantitative models
OR09 0.1 0.8 0.1 1.12 0.95 1.27 1.276
for sustainable supply chain management: developments and directions. Eur. J.
OR10 0.1 0.9 0.0 1.30 0.94 1.66 1.682
Oper. Res. 233, 299–312.
OR11 0.2 0.0 0.8 1.17 0.99 0.68 0.701 Brandenburg, M., Hahn, G. J. Sustainable aggregate production planning in the
OR12 0.2 0.1 0.7 1.50 0.96 0.00 0.493 chemical process industry - a benchmark problem and dataset. Data Brief Sub-
OR13 0.2 0.2 0.6 1.21 0.97 1.33 1.350 mitted for publication.
OR14 0.2 0.3 0.5 1.23 0.95 1.37 1.385 Brandenburg, M., Rebs, T., 2015. Sustainable supply chain management: a modeling
OR15 0.2 0.4 0.4 1.15 0.95 0.63 0.648 perspective. Ann. Oper. Res. 229, 213–252.
OR16 0.2 0.5 0.3 1.18 0.95 0.63 0.655 Brandenburg, M., Tölle, F.-J., 2009. MILP-based campaign scheduling in a specialty
OR17 0.2 0.6 0.2 1.12 0.94 0.65 0.657 chemicals plant: a case study. OR Spectrum 31, 141–166.
OR18 0.2 0.7 0.1 1.13 0.95 1.42 1.430 Burkardt, R., Hatzl, J., 2005. Review, extensions and computational comparison of
OR19 0.2 0.8 0.0 1.06 0.94 1.62 1.623 milp formulations for scheduling of batch processes. Comput. Chem. Eng. 29,
OR20 0.3 0.0 0.7 1.20 1.03 0.41 0.461 1752–1769.
Burkardt, R., Hatzl, J., 2006. A complex time based construction heuristic for batch
OR21 0.3 0.1 0.6 1.48 1.05 0.63 0.800
scheduling problems in the chemical industry. Eur. J. Oper. Res. 174, 1162–1183.
OR22 0.3 0.2 0.5 1.08 0.95 0.63 0.634
Cefic, 2014. Teaming up for a sustainable europe: the chemical industry’s commit-
OR23 0.3 0.3 0.4 1.15 0.95 0.87 0.882
ment to sustainability, 2013/2014 report. Brussels, Cefic - The European Chemi-
OR24 0.3 0.4 0.3 1.15 0.94 0.63 0.648 cal Industry Council.
OR25 0.3 0.5 0.2 1.09 0.97 0.68 0.684 Chakraborty, A., Malcolma, A., Colberg, R.D., Linninger, A.A., 2004. Optimal waste
OR26 0.3 0.6 0.1 1.10 1.00 0.90 0.907 reduction and investment planning under uncertainty. Comput. Chem. Eng. 28,
OR27 0.3 0.7 0.0 1.25 0.95 1.84 1.854 1145–1156.
OR28 0.4 0.0 0.6 1.07 0.98 0.68 0.683 Das, K., Sengupta, S., 2009. A hierarchical process industry production–distribution
OR29 0.4 0.1 0.5 1.19 0.98 1.33 1.348 planning model. Int. J. Prod. Econ. 117, 402–419.
OR30 0.4 0.2 0.4 1.13 0.98 1.33 1.341 Dietz, A., Pibouleau, L.G., Azzaro Pantel, C., Domenech, S., 2005. A framework for
OR31 0.4 0.3 0.3 1.03 0.98 0.63 0.632 multiproduct batch plant design with environmental consideration: application
OR32 0.4 0.4 0.2 1.19 0.97 1.43 1.447 to protein production. Industr. Eng. Chem. Res. 44, 2191–2206.
OR33 0.4 0.5 0.1 1.33 0.97 1.23 1.272 Gansterer, M., Almeder, C., Hartl, R.F., 2014. Simulation-based optimization methods
OR34 0.4 0.6 0.0 1.10 0.94 1.90 1.901 for setting production planning parameters. Int. J. Prod. Econ. 151, 206–213.
García-Serna, J., Pérez-Barrigón, L., Cocero, M., 2007. New trends for design towards
OR35 0.5 0.0 0.5 1.09 0.97 0.17 0.191
sustainability in chemical engineering: green engineering. Chem. Eng. J. 133,
OR36 0.5 0.1 0.4 1.26 1.00 0.48 0.547
7–30.
OR37 0.5 0.2 0.3 1.10 0.96 0.67 0.675 Govindan, K., Cheng, T., 2015. Sustainable supply chain management: advances in
OR38 0.5 0.3 0.2 1.01 1.05 0.44 0.451 operations research perspective. Comput. Oper. Res. 54, 177–179.
OR39 0.5 0.4 0.1 1.02 0.95 1.21 1.212 Graves, S.C., 1986. A tactical planning model for a job shop. Oper. Res. 34, 522–
OR40 0.5 0.5 0.0 1.09 0.94 1.58 1.580 533.
OR41 0.6 0.0 0.4 1.25 1.01 0.68 0.724 Günther, H.-O., van Beek, P. (Eds.), 2003. Advanced Planning and Scheduling Solu-
OR42 0.6 0.1 0.3 1.11 0.99 1.27 1.276 tions in Process Industry. Springer.
OR43 0.6 0.2 0.2 1.21 0.99 1.12 1.144 Günther, H.-O., Meyr, H. (Eds.), 2009. Supply Chain Planning - Quantitative Decision
OR44 0.6 0.3 0.1 1.08 0.96 0.90 0.908 Support and Advanced Planning Solutions. Springer.
OR45 0.6 0.4 0.0 1.05 0.96 1.66 1.663 Hahn, G., Packowski, J., 2015. A perspective on applications of in-memory analytics
OR46 0.7 0.0 0.3 1.10 0.98 0.71 0.722 in supply chain management. Decis. Support Syst. 76, 45–52.
OR47 0.7 0.1 0.2 1.00 0.98 1.27 1.271 Hahn, G.J., Sens, T., Decouttere, C., Vandaele, N.J., 2016. A multi-criteria approach to
robust outsourcing decision-making in stochastic manufacturing systems. Com-
OR48 0.7 0.2 0.1 1.03 0.97 1.15 1.148
put. Industr. Eng. 98, 275–288.
OR49 0.7 0.3 0.0 1.12 0.98 1.57 1.570
Mirzapour Al-e hashem, S.M.J., Baboli, A., Sazvar, Z., 2013. A stochastic aggregate
OR50 0.8 0.0 0.2 1.07 1.01 0.92 0.925
production planning model in a green supply chain: considering flexible lead
OR51 0.8 0.1 0.1 1.04 0.97 1.44 1.439 times, nonlinear purchase and shortage cost functions. Eur. J. Oper. Res. 230,
OR52 0.8 0.2 0.0 1.11 0.98 1.49 1.494 26–41.
OR53 0.9 0.0 0.1 1.02 1.06 0.42 0.437 Hax, A., Meal, H., 1975. Hierarchical integration of production planning and schedul-
OR54 0.9 0.1 0.0 1.01 0.97 1.32 1.317 ing. In: Geisler, M. (Ed.), Studies in Management Science. Elsevier, pp. 53–
69.
Note: θ OVER = 1 − θ COST − θ EMIS . Honkomp, S., Lombardo, S., Rosen, O., Pekny, J., 20 0 0. The curse of reality: why pro-
a
Gap to ideal measures the Euclidean distance to the ideal point. cess scheduling optimization problems are difficult in practice. Comput. Chem.
Eng. 24, 323–328.
Hugo, A., Pistikopoulos, E., 2005. Environmentally conscious long-range planning
and design of supply chain networks. J. Clean. Prod. 13, 1471–1491.
Ilgin, M.A., Gupta, S.M., 2010. Environmentally conscious manufacturing and product
recovery (ECMPRO): a review of the state of the art. J. Environ. Manage. 91,
References 563–591.
Infante, C.E.D.d.C., Molica de Mendoça, F., Purcidonio, P.M., Valle, R., 2013. Triple
Adonyi, R., Biros, G., Holczinger, T., Friedler, F., 2008. Effective scheduling of a bottom line analysis of oil and gas industry with multicriteria decision making.
large-scale paint production system. J. Clean. Prod. 16, 225–232. J. Clean. Prod. 52, 289–300.
Almeder, C., Preusser, M., Hartl, R.F., 2009. Simulation and optimization of supply Ishii, N., Muraki, M., 1996. A process-variability-based online scheduling system in
chains: alternative or complementary approaches? OR Spectrum 31, 95–119. multiproduct batch process. Comput. Chem. Eng. 20, 217–234.
Arbiza, M.J., Bonfill, A., Guillén, G., Mele, F.D., Espunã, L.P., 2008. Metaheuristic Jaehn, F., 2016. Sustainable operations. Eur. J. Oper. Res. 253, 243–264.
multiobjective optimisation approach for the scheduling of multiproduct batch Jansen, M.M., de KokTon, G., Fransoo, J.C., 2013. Lead time anticipation in supply
chemical plants. J. Clean. Prod. 16, 233–244. chain operations planning. OR Spectrum 35, 251–290.

Please cite this article as: G.J. Hahn, M. Brandenburg, A sustainable aggregate production planning model for the chemical process
industry, Computers and Operations Research (2018), https://doi.org/10.1016/j.cor.2017.12.011
JID: CAOR
ARTICLE IN PRESS [m5G;January 17, 2018;17:22]

G.J. Hahn, M. Brandenburg / Computers and Operations Research 000 (2018) 1–15 15

Jawjit, W., Kroeze, C., Rattanapan, S., 2010. Greenhouse gas emissions from rubber Saidur, R., Mekhilef, S., 2010. Energy use, energy savings and emission analysis in
industry in thailand. J. Clean. Prod. 18, 403–411. the malaysian rubber producing industries. Appl. Energy 87, 2746–2758.
Jawjit, W., Pavasant, P., Kroeze, C., 2015. Evaluating environmental performance of Schneeweiss, C., 2003. Distributed decision making: a unified approach. Eur. J. Oper.
concentrated latex production in Thailand. J. Clean. Prod. 98, 8491. Res. 150, 237–252.
Jödicke, G., Zenklusen, O., Weidenhaupt, A., Hungerbühler, K., 1999. Developing en- Selçuk, B., Fransoo, J.C., De Kok, A.G., 2008. Work-in-process clearing in supply chain
vironmentally-sound processes in the chemical industry: a case study on phar- operations planning. IIE Trans. 40, 206–220.
maceutical intermediates. J. Clean. Prod. 7, 159–166. Seuring, S., 2013. A review of modeling approaches for sustainable supply chain
Kallrath, J., 2002a. Combined strategic and operational planning: an milp success management. Decis. Support Syst. 54, 1513–1520.
story in chemical industry. OR Spectrum 24, 315–341. Seuring, S., Müller, M., 2008. From a literature review to a conceptual framework
Kallrath, J., 2002b. Planning and scheduling in the process industry. OR Spectrum for sustainable supply chain management. J. Clean. Prod. 16, 1699–1710.
24, 219–250. Shah, N., 2005. Process industry supply chains: advances and challenges. Comput.
Kang, Y., Albey, E., Hwang, S., Uzsoy, R., 2014. The impact of lot-sizing in multiple Chem. Eng. 29, 1225–1235.
product environments with congestion. J. Manuf. Syst. 33, 436–444. Söhner, V., Schneeweiss, C., 1995. Hierarchically integrated lot size optimization. Eur.
Karmarkar, U.S., 1987. Lot sizes, lead times and in-process inventories. Manage. Sci. J. Oper. Res. 86, 73–90.
33, 409–418. Song, J., Park, H., Lee, D.-Y., Park, S., 2002. Scheduling of actual size refinery pro-
Kraemer, W., Langenbach-Belz, M., 1976. Approximate Formulae for the Delay in the cesses considering environmental impacts with multiobjective optimization. In-
Queueing System GI/G/1. International Teletraffic Congress. dustr. Eng. Chem. Res. 41, 4794–4806.
Lambrecht, M.R., Ivens, P.L., Vandaele, N.J., 1998. Aclips: a capacity and lead time Stadtler, H., 2005. Supply chain management and advanced planning – basics,
integrated procedure for scheduling. Manage. Sci. 44 (11), 1548–1561. overview and challenges. Eur. J. Oper. Res. 163, 575–588.
Li, Z., Ierapetritou, M., 2008. Process scheduling under uncertainty: review and chal- Taticchi, P., Garengo, P., Nudurupati, S., Tonelli, F., Pasqualino, R., 2015. A review of
lenges. Comput. Chem. Eng. 32, 715–727. decision-support tools and performance measurement and sustainable supply
Linninger, A.A., Chakraborty, A., Colberg, R.A., 20 0 0. Planning of waste reduction chain management. Int. J. Prod. Res. 53, 6473–6494.
strategies under uncertainty. Comput. Chem. Eng. 24, 1043–1048. Vaklieva-Bancheva, N.G., Kirilova, E.G., 2010. Cleaner manufacture of multipurpose
Méndez, C.A., Cerdá, J., Grossmann, I.E., Harjunkoski, I., Fahl, M., 2006. batch chemical and biochemical plants. scheduling and optimal choice of pro-
State-of-the-art review of optimization methods for short-term scheduling of duction recipes. J. Clean. Prod. 18, 1300–1310.
batch processes. Comput. Chem. Eng. 30, 913–946. Velez, S., Maravelias, C.T., 2014. Advances in mixed-integer programming methods
Meyr, H., Wagner, M., Rohde, J., 2002. Structure of advanced planning systems. In: for chemical production scheduling. Annu. Rev. Chem. Biomol. Eng. 5, 97–121.
Stadtler, H., Kilger, C. (Eds.), Supply Chain Management and Advanced Planning- Vogel, M., Braungardt, T., Meyer, W., Schneider, W., 2012. The effects of shift work
Concepts, Models Software and Case Studies. Springer, pp. 99–104. on physical and mental health. J. Neural Transm. 119, 1121–1132.
Missbauer, H., Uzsoy, R., 2011. Optimization models of production planning prob- Westenberger, H., Kallrath, J., 1994. Formulation of a job shop problem in process
lems. In: Kempf, K.G., Keskinocak, P., Uzsoy, R. (Eds.), Planning Production and industry. Technical Report. Bayer AG, Leverkusen, Germany and BASF AG, Lud-
Inventories in the Extended Enterprise. Springer, Boston, pp. 437–507. wigshafen, Germany. Unpublished working paper.
Mula, J., Poler, R., Garcia-Sabater, J.P., Lario, F.C., 2006. Models for production plan- You, F., Tao, L., Graziano, D., Snyder, S., 2012. Optimal design of sustainable cellu-
ning under uncertainty: a review. Int. J. Prod. Econ. 103, 271–285. losic biofuel supply chains: multiobjective optimization coupled with life cycle
Northrup, H.R., Wilson, J.T., Rose, K.M., 1979. The twelve-hour shift in the petroleum assessment and input-output analysis. Am. Inst. Chem. Eng. J. 58, 1157–1180.
and chemical industries. Industr. Labor Relat. Rev. 32, 312–326. Yue, D., You, F., 2013. Sustainable scheduling of batch processes under economic and
Pahl, J., Voß, S., Woodruff, D.L., 2007. Production planning with load dependent lead environmental criteria with minlp models and algorithms. Comput. Chem. Eng.
times: an update of research. Ann. Oper. Res. 153, 297–345. 54, 44–59.
Papageorgiou, L., Pantelides, C., 1993. A hierarchical approach for campaign planning Yura, K., 1994. Production scheduling to satisfy workers preferences for days off and
of multi-purpose batch plants. Comput. Chem. Eng. 17, 27–32. overtime under due-date constraints. Int. J. Prod. Econ. 33, 265–270.
Penkuhn, T., Spengler, T., Püchert, H., Rentz, O., 1997. Environmental integrated pro- Zhao, Y., Xu, J., Xie, X., Yu, H., 2014. An integrated environmental impact assess-
duction planning for the ammonia synthesis. Eur. J. Oper. Res. 97, 327–336. ment of corn-based polyols compared with petroleum-based polyols produc-
Rager, M., Gahm, C., Denz, F., 2015. Energy-oriented scheduling based on evolution- tion. J. Clean. Prod. 68, 272–278.
ary algorithms. Comput. Oper. Res. 54, 218–231. Zhou, Z.Y., Cheng, S.W., Hua, B., 20 0 0. Supply chain optimization of continuous
Ribas, I., Leisten, R., Framiñan, J., 2010. Review and classification of hybrid flow shop process industries with sustainability considerations. Comput. Chem. Eng. 24,
scheduling problems from a production system and a solutions procedure per- 1151–1158.
spective. Comput. Oper. Res. 37, 1439–1454.
Rutten, W., 1993. Hierarchical mathematical programming for operational planning
in a process industry. Eur. J. Oper. Res. 64, 363–369.

Please cite this article as: G.J. Hahn, M. Brandenburg, A sustainable aggregate production planning model for the chemical process
industry, Computers and Operations Research (2018), https://doi.org/10.1016/j.cor.2017.12.011

You might also like