A Model-Based Decision of Production Planning in The Biopharmaceutical Industry

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 14

Computers & Industrial Engineering 129 (2019) 354–367

Contents lists available at ScienceDirect

Computers & Industrial Engineering


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

A model-based decision support framework for the optimisation of T


production planning in the biopharmaceutical industry
Miguel Vieira, Tânia Pinto-Varela , Ana P. Barbosa-Póvoa

CEG–IST, Instituto Superior Técnico, Universidade de Lisboa, Lisboa, Portugal

ARTICLE INFO ABSTRACT

Keywords: The enterprise management of current industrial processes has becoming increasingly complex, imposing novel
Decision support challenges to the implementation of efficient decision support tools. Manual-based planning/scheduling systems
Optimisation model have been replaced by integrated computational tools for core business processes, often in real-time, to enable a
Production and maintenance planning collaborative platform for the control and analysis of information. Despite the advances in process modelling, the
Biopharmaceutical industry
implementation of optimisation support has been fairly discussed on how to improve industrial competitiveness
and responsiveness to market variability while addressing different profitability criteria. In this work, we discuss
a proof-of-concept implementation prototype of a model-based decision support tool for the production and
maintenance planning optimisation in a biopharmaceutical industrial case – ICOMPASS, defining its application
framework for a successful deployment in a real production environment.

1. Introduction (2015), most industrial systems still tend to focus on simplistic models
and heuristics, often leading to sub-optimal solutions with time-con-
The dynamic structure of most production systems requires the as- suming procedures and subject to the decision maker expertise.
sessment of different business factors for the majority of strategic and With the systematisation of industrial optimisation problems, sev-
operational decisions (Varma, Reklaitis, Blau, & Pekny, 2007). In par- eral research contributions have targeted industrial problems related to
ticular, production planning activities have been striving for the im- supply-chain networks, energy demand, process design regulations,
provement of operational efficiency to deal with the multiple process social impact or environmental load, by exploring different approaches
constraints and market competition. With the digital transformation of to explore the solutions’ space rapidly, reduce uncertainty and make
current facilities, companies have adopted the use of process automa- better, faster and safer decisions (Barbosa-Póvoa, 2012; Castro,
tion technologies that enable an enhanced control of tasks parameters Grossmann, Veldhuizen, & Esplin, 2014; Grossmann, 2012; Maravelias,
and access to on-line operational data. These capabilities unveiled an 2012; Méndez, Cerdá, Grossmann, Harjunkoski, & Fahl, 2006). One of
opportunity to potentiate its use to support decision-making, providing the main research fields relies in industrial planning and scheduling
the replacement of inefficient manual spreadsheets with advanced optimisation, where major achievements include exact (LP, MILP and
analytical tools. As reviewed by Harjunkoski et al. (2014), model-based MINLP), non-exact (meta/heuristics, evolutionary algorithms and arti-
methodologies have been successfully applied to industrial problems to ficial intelligence methods) and simulation-optimisation approaches.
address the optimisation of multiple operational criteria. But despite Some methodologies tackle the uncertainty associated with operational
the major research advances in modelling approaches towards the in- parameters by introducing stochastic optimisation, allowing decisions
dustrial complexity, the ability to implement optimisation methodolo- to occur at points in time driven from previous outcomes. However,
gies in decision-making faces significant challenges with insufficient most of these cases tend to focus on the modelling point of view and
discussion among research works. An effective decision support tool disregards the implementation challenges as a decision support appli-
needs to consider, for example, the process network interconnectivity, cation. For example, several issues arise on how to structure the in-
the operational uncertainty and the computational performance, al- dustrial planning and scheduling decisions into a functional operational
lowing the decision-maker to evaluate the quality of solutions ac- system. The reference to protocols as ISA-95 has provided as interna-
cording to strategic options. However, as noted by Figueira et al. tional standard for the interactions of the different information layers in


Corresponding author at: CEG–IST, Instituto Superior Técnico, Universidade de Lisboa, Av. Rovisco Pais, 1, 1049-001 Lisboa, Portugal.
E-mail addresses: migueljvieira@tecnico.ulisboa.pt (M. Vieira), tania.pinto.varela@tecnico.ulisboa.pt (T. Pinto-Varela),
apovoa@tecnico.ulisboa.pt (A.P. Barbosa-Póvoa).

https://doi.org/10.1016/j.cie.2019.01.045
Received 15 September 2017; Received in revised form 11 May 2018; Accepted 22 January 2019
Available online 25 January 2019
0360-8352/ © 2019 Elsevier Ltd. All rights reserved.
M. Vieira et al. Computers & Industrial Engineering 129 (2019) 354–367

4 - Establishing the basic plant schedule – production,


material use,delivery and shipping. Determining inventory
levels
Time Frame: Months, weeks, days

3 -Workflow/ recipe control to produce the desired end-


products. Maintaining record and optimising the
production process
Time Frame: Days, shifts, hours, minutes, seconds

2 - Monitoring, supervision control and automated control


of production process
Time Frame: Hours, minutes, seconds, subseconds
1 - Sensing the production process,manipulating the
production process
0 - The actual production process

Fig. 1. Hierarchical automation pyramid (ANSI/ISA-95).

automated interfaces between enterprise and control systems (ANSI/ decision support system for the planning and scheduling activities of a
ISA-95.00.01-2000, 2000). As shown in Fig. 1, the referential hier- pulp and paper mill company to generate plans with different levels of
archical automation pyramid considers the bottom levels composed by customization; Eberle et al. (2016) suggested a model solution for the
the monitoring control systems/sensors, followed by the manufacturing optimisation scheduling of a liquid-drug pharmaceutical process by
execution system (MES) which deals with more advanced production established planning horizons for real commercial productions with a
controls and models, while the top the pyramid drives the long-term potential reduction of makespan when compared to standard expert
strategic and tactical business decisions of the supply chain manage- schedules; and Dehghanimohammadabadi, Keyser, and Cheraghi
ment based on enterprise resource planning (ERP). And, although the (2017) developed an effective optimisation based simulation frame-
integration proposed by the standardised automation pyramid may not work for manufacturing processes allowing to reduce the problem
work on in all environments, it can provide a general structure for complexity resulting in better performance measures of scheduling
successful implementations. strategies. Some of these support tools presented practical interfaces
Some successful implementations in process industries can provide using the capabilities of common spreadsheet/database software (e.g.
relevant inputs for the development of a framework solution approach: MS Excel) and optimisation solvers with new integration approaches of
the studies by Janak, Floudas, Kallrath, and Vormbrock (2006) and ERP systems. For example, to extract production orders and launch the
Shaik, Floudas, Kallrath, and Pitz (2009) discussed a model approach to optimisation model on a remote server, and then allowing the scheduler
the scheduling problems involving recipes for hundreds of products and to make final adjustments in the solution. Other tools can be found in
pieces of equipment in a large-scale chemical process, either involving commercial software like ABB CPM P&P Suite for a pulp and paper
batch and continuous processes, to be used in a daily basis for an effi- plant, that offers a collaborative solution able to manage most planning
cient medium-term utilisation of the main equipment units; Wassick tasks (order planning, trimming, quality, delivery planning, asset
(2009) and Wassick and Ferrio (2011) presented an integrated chemical management, etc.) for the improvement of the overall efficiency of the
case analysing different optimisation problems related to production plant.
planning, energy scheduling, plant design and waste treatment sche- In this work, we assess the implementation of a proof-of-concept
duling, proposing a solution tool that resulted in a significant reduction model-based decision support systems in the operational management
on scheduling related shutdowns, reduction of third party drumming of biopharmaceutical industrial processes, to our knowledge, not yet
providers and improved customer level; Choy, Leung, Chow, Poon, and addressed in literature. Following a systematic approach towards the
Kwok (2011) proposed a simulation-optimisation support on make-to- optimisation of industrial planning problems, the discussion is focused
order scheduling problems in the mould industry, improving opera- on the core decision-making characteristics of industrial processes,
tional functions by reducing the time in performing tasks and number of considering the guidelines for an analytical model approach to the
late projects; Pehrsson, Ng, and Stockton (2013) developed a decision planning problem optimisation, and addresses the application chal-
support methodology combining production engineering data, financial lenges of real production environments with a prototype interface of a
data and optimisation technology applied to an automotive production decision support tool ICOMPASS. The structure of this paper starts with
system with improved investment solutions; Vieira, Pinto-Varela and the discussion of the applicability of planning/scheduling decision
Barbosa-Póvoa (2015) presented an model-based optimisation case in support solutions and its industrial deployment challenges in Section 2.
the paint industry for the production scheduling improvement, pro- Based on that, in Section 3 we discuss the decision support framework
viding the decision maker with the workload capacity for a given de- applied to the biopharmaceutical planning problem, exploring the in-
mand plan, which proved relevant to study the requirements of tem- dustrial requirements by extending a model of the literature and de-
porary workforce according to the production demand variability fining the decision-making process. This solution approach is illustrated
throughout the year; Figueira et al. (2015) developed a model-based with the development of an application tool applied to an industrial

355
M. Vieira et al. Computers & Industrial Engineering 129 (2019) 354–367

case in Section 4; finally, in Section 5 is summarized the advantages of


the decision support system and the directions for further improve-
ments.

2. Decision-support implementation

The development of highly automated operational control has en-


abled the digital transformation of industrial processes with the in-
tegration of advanced decision support systems (Biel & Glock, 2016;
Kallestrup, Lynge, Akkerman, & Oddsdottir, 2014; Ngai, Peng,
Alexander, & Moon, 2014). In concept, these systems involve a human-
computer interaction with the knowledge/experience held by the pro-
cess stakeholders, structured in a way that a model-based methodology
provides the information to managers decide the best solution for a
complex problem. Shim et al. (2002), Power and Sharda (2007) and
Stephanopoulos and Reklaitis (2011) have reviewed the evolution of
functionalities and applications for decision support systems using
flexible infrastructures for the data acquisition/processing, decision
analysis, mathematical optimisation and simulation technologies, Fig. 2. IT application architecture.
identifying behavioural/technical issues with impact in its performance
and adoption. But, as also noted as Harjunkoski et al. (2014), one of
main challenges for its application in industrial optimisation relies on application concept should include (Fig. 2): (i) an interface environ-
the deployability issues. These authors discussed the importance, in one ment for the problem formulation/representation and analysis of so-
hand, of the role of the production scheduler towards the conflicting lution results, (ii) a database spreadsheet for the input parameters
organisation goals and the reliability of information available, and on processing and platform for generated data, and (iii) a processing
the other hand, the planning impact induced by day-to-day disruptions modelling system for the optimisation model and solver. The mathe-
from the supply-chain, manufacturing and logistics functions (e.g. rush matical models are commonly expressed in an equation form (typically
orders, late deliveries, equipment breakdown, economics variability, MILP models, or MINLP when considering explicit nonlinearities)
etc.), demanding an ongoing monitor, negotiations and adjustments of which allows to be easily implemented in commercial modelling sys-
the schedule. With this wide variability in process metrics, it is re- tems like AMPL, GAMS or ILOG, that runs optimisation proprietary
cognised that if the information flow could be computerized, then each solvers such as CPLEX, XPRESS or BARON. Furthermore, these model-
decision could rely less on the scheduler’s expertise. But acknowledging ling systems offer the interface with spreadsheets, database or graphics
that a fully automated decision process is unreasonable for the majority packages in common MS Office environments, facilitating the deploy-
of industrial problems, an accurate evaluation of the impact of the ment of a user interface to be managed by non-specialists. Finally, the
improved planning/scheduling results is crucial to establish the align- selected applications must have a compatible application programming
ment and commitment of all stakeholders. Therefore, with a proof-of- interface (API) with the protocols for the developed software integra-
concept study it is be possible to approach each industrial problem tion and connectivity of components. As well, recognising the techno-
specifications and, based in the previous reviewed examples, follow logical advances in computer network security, cloud-based server
some key guidelines for a successful industrial application development systems can reduce the implementation investment. For example,
and implementation: considering an in-house database platform that communicates with the
modelling environment (via an API) to define/validate the optimisation
• The modelling approach should be flexible to accommodate dif- problem, then submitting the data commands/conditions to centralized
ferent decisional problems and generic enough to integrate several hosted servers which generates the solution with the solver (con-
information sources to improve data quality and control, exploiting sidering a predefined solution termination/convergence criteria), and
particular problem specifications for computational efficiency; connects the output results in the form of raw data to the database for
• The scheduler requirements should be the main focus of the solution further analysis and graphical display (subject to pre- and pos-proces-
provided, allowing an effective user interface control of inputs sing data architecture).
modifications, data processing, schedule manipulations, and gra- With a reasonable number of commercial software for industrial
phical visualisation (e.g. Gantt chart) with KPI trends analysis or planning/scheduling support suited for the process industry (e.g.
reporting, which could be easily recognisable as an enhancement to aspenONE PM&E, SuperPro Designer, BioSolve Process, etc.), some
their current job; decision support tools can also be found as extension packages in
• Amid the scheduler assumption of the constraints contingencies of common business planning software, such as SAP or Oracle systems.
the problem, the solution must be robust, scalable, and stable (with However, certain issues can be identified upon the quality and accuracy
at least a coherent feasible solution reported), allowing the possi- of the solution provided, since most systems rely on heuristic ap-
bility to efficiently generate new adjusted solutions reactive to proaches which often lead to sub-optimality, as well as to the suitability
process disturbances; extent of its functionalities to the problem case. The system integration
• The solution should be easily integrated and maintainable with in- with the technology of the process enhances the flow of information
dustrial software platforms (e.g. ERP-software and plant control across the hierarchical functions, but the analysis, control, revision and
systems), representing a good return-on-investment on the hard- translation of that data for an effective decision support requires par-
ware/software structure deployment costs and license fees. ticular attention performed by a decision manager. With regard to the
complexity of the industrial problem, its on-site implementation could
Concerning the IT infrastructure requirements and the commercial require a timeline of several months, concerning the requirements
considerations for a decision support tool, the architecture for an analysis of process specifications, the system design/development with

356
M. Vieira et al. Computers & Industrial Engineering 129 (2019) 354–367

validation/testing, and its installation with the necessary training and which commonly dominates the processing cycle time (e.g. a purifica-
operation. tion stage, following the active biological ingredient fermentation
harvest, can take 2–3 days while quality control 3–4 weeks). With such
2.1. Bio/pharmaceutical planning optimisation challenges multiple constraints arising from the complexity of biopharmaceutical
manufacturing, the optimisation approach to these problems should
The pharmaceutical industry has been the scope for several en- assess the overall structure of a novel decision support tool to provide
terprise optimisation studies due to the strong competition among valuable, flexible and efficient solutions for an industrial production
manufacturers and its process network complexity (Shah, 2004). The scale.
research field on applied optimisation methodologies spreads from the
management of drug development portfolio pipelines (e.g. SimBio- 3. Solution framework
pharma prototype decision support tool proposed by Farid, Washbrook,
and Titchener-Hooker (2007)) to the cost-effective design of specific Considering the application case to the biopharmaceutical industry,
steps of the manufacturing process (e.g. processing optimisation of the development framework for a decision support system is now dis-
bioreactor and chromatography sizing strategies by Liu, Farid, and cussed for the planning optimisation of the production process. Based in
Papageorgiou (2016)). In particular, the collaboration between aca- the methodology proposed by Moniz, Barbosa-Povoa, de Sousa, and
demia and industry has promoted the development of multiple mod- Duarte (2014), this framework can be designed under three compo-
elling approaches for the optimisation of planning/scheduling solutions nents, depicted in Fig. 3: the first is related to the definition of the
in real pharmaceutical plants (Kopanos, Méndez, & Puigjaner, 2010; problem structure, providing the components of the decision-making
Stefansson, Sigmarsdottir, Jensson, & Shah, 2011). As examples, Moniz, and the processing of the operational data for the optimisation model;
Barbosa-Povoa, and de Sousa (2014) proposed a unified scheduling secondly, the optimisation component defines the modelling approach
methodology based in a discrete-time optimisation model for an in- to generate the solutions set for the given problem; and finally, the
tegrated decision support solution in a chemical-based pharmaceutical analysis of solutions via an user-interface provides the interactive
industry, while Costa (2015) presented a metaheuristic optimisation procedure to support the decision taken. This framework structure
engine embedded in a production scheduler for medium-term forecast discusses a generic approach to industrial planning/scheduling pro-
analysis applied to lyophilised drug manufacturing processes. blems in order to allow a flexible adaptation to different problem cases.
Despite the dominance of chemical-pharmaceutical cases, the pro- For the defined scope, the detail of the different components will allow
duction of highly effective bio-based drugs (e.g. cell or gene therapies, the definition of an integrated electronic decision support tools suitable
therapeutic proteins hormones, monoclonal antibodies, cytokines and for the problem specificities of biopharmaceutical production, named as
tissue growth factors) have defined new modelling challenges, either to ICOMPASS.
address the strict process regulatory policies on products storage shelf-
life and biological variability, or with the operational equipment 3.1. Problem structure
compliance to product quality. In fact, each production step requires a
legal validation by the regulatory authority, subject to the control and The problem structure addresses not only the definition of the in-
supervision of government agencies such as EMA (European Medicines dustrial process network but foremost the definition of the essential
Agency) and FDA (US Food and Drug Administration). These autho- objectives to the decision-making. In particular, it should be able to
rities establish documented evidence to assure a consistent production integrate the different types of knowledge and evolve according to the
according to the pre-determined specifications and quality attributes, decision process complexity. Understandably, the increasing problem
imposing a set of regulations to comply with Good Manufacturing complexity leads to the discussion on how to solve large scale industrial
Practices (GMP). Regarding planning/scheduling optimisation appli- cases, to which the computational restrictions for the mathematical
cations, it is acknowledged that only a short number of literature re- optimisation are a known topic (Méndez et al., 2006). In order to ad-
search is dedicated to modelling formulations to address biopharma- dress the flexibility and scalability of problem representations, several
ceutical production constraints, from the optimal capacity planning to advantages were identified with the development of model re-
campaign-lots scheduling with maintenance integration (Eberle et al., presentation based in general networks. As reviewed by Harjunkoski
2016; Kabra, Shaik, & Rathore, 2013; Lakhdar, Zhou, Savery, Titchener- et al. (2014), numerous examples of State-Task Network (STN) (Kondili,
Hooker, & Papageorgiou, 2005; Liu, Yahia, & Papageorgiou, 2014; Pantelides, & Sargent, 1993) and Resource-Task Network (RTN)
Shaik, Dhakre, Rathore, & Patil, 2014; Siganporia, Ghosh, Daszkowski, (Pantelides, 1994) representations applied to planning/scheduling
Papageorgiou, & Farid, 2014; Vieira, Pinto-Varela, Moniz, Barbosa- problems can be found in literature, exploring different modelling scale
Póvoa, & Papageorgiou, 2016; Vieira, Pinto-Varela, & Barbosa-Póvoa, configurations with integrated operational decisions (e.g. plant/process
2017b). However, the work by Leachman, Johnston, Li, and Shen design). The general network structure of a production process enables
(2014) is one of the few that developed a database and optimisation the required trade-off between the required detail of the optimal solu-
application engine for the supply chain planning of a real biopharma- tions and the computational intractability of the problem. And the in-
ceutical company, generating the batch processing schedule for each dustrial collaboration becomes essential to improve the development of
production facility and finished (single) product availability, according such optimisation approaches, supporting the structure of how the data
to demand regulations of the different markets. As noted by the author, is generated and related decisions are made at the production en-
the structure of most biopharmaceutical facilities tends to be small in vironment.
comparison to typical chemical-pharmaceutical plants, since many For our approach, the problem consists in determine the optimal
plants are only dedicated to one product to control cross-contamination production planning of an arbitrary process network of a biopharma-
and storage regulations. However, it has been verified a steady growth ceutical process, addressing the campaign plan of batch and/or con-
in this sector following the clinical success revenues with the devel- tinuous process steps, multiple intermediate deliveries, sequence de-
opment of new biologic drugs and biosimilars (Otto, Santagostino, & pendent changeovers operations, product storage restricted to shelf-life
Schrader, 2014). Significant investments are also required in manu- limitations, the track-control of the campaign lots due to regulatory
facturing technologies (batch or continuous operation) to enhance the policies, and the maintenance requirements of equipment subject to
process control of each stage and comply to rigorous quality testing, performance decay. The manufacturing process is commonly refereed

357
M. Vieira et al. Computers & Industrial Engineering 129 (2019) 354–367

Fig. 3. Structure of the planning framework methodology based in Moniz et al. (2014b).

as two-stages: biological raw-materials undergo upstream cell-fermen- decision targets, with some objective functions modelled to allow dif-
tation to produce the intermediate active biological ingredient, fol- ferent solution analysis, being the most common the production profit
lowed by downstream purification and stabilisation processing to ob- maximisation of the process operation.
tain the final product. The final product is later labelled/packaged and
shipped to retail distributors/consumers. Since each bioprocess stage 3.2. Optimisation methodology
involves multiple sequential steps with long processing times, the main
optimisation decisions are usually focused on the campaign size, allo- The development of a mathematical optimisation approach implies
cation and sequencing for each aggregated upstream/downstream the identification of the characteristics of the industrial problem in
suite, as proposed by Lakhdar et al. (2005). This aggregated approach order to formulate the constraints of the process. However, the com-
considers problems with 1–2 years horizons, which are typically de- binatorial nature of these planning/scheduling problems quickly esca-
fined as medium-term planning. However, the problem formulation lades with the binary decisions of task-lot-unit assignment and se-
should enable the possibility to consider any problem configuration, for quencing, storage or sequence-dependent changeovers tasks. For
example, the short-term scheduling of a more task detailed sequence to problems considering a significant number of processing units for long
match the planning supplies. The definition of a standard representa- time horizons tend to be difficult to solve by exact methods, being
tion is of key importance for this problem adaptability, as the RTN that necessary to resort to decomposition techniques or heuristic approaches
addresses all process representations in terms of tasks and resources: to obtain a satisfactory solution in a reasonable time. This enables the
each task is defined to consume and/or produce resources, either ma- wide research on multiple models and solutions required to fit the re-
terials, utilities or equipment, constrained through a given set of pro- quirements of the decision-making process and its suitability to the
cess parameters. And, while considering a given demand plan and all scope of the industrial problems (Grossmann, 2005).
the process data, the set of equipment resources available is allocated to With the definition of the problem structure, the model develop-
perform the corresponding process tasks, subject to an objective func- ment could test different mathematical approaches, constraints or ob-
tion. The decision-maker should then be able to draw different sce- jective functions to ensure solutions feasibility. As noted by Moniz,
narios and analyse key performance indicators (KPI) of the problem Barbosa-Povoa, de Sousa, and Duarte (2014), in the periodic planning
solution, as well as to provide the ability to generate new solutions of real manufacturing environments, the optimisation system is highly
when process data or demand is revised. Thus, for each problem case, relevant for rescheduling purposes, since, at every time point, is ex-
the goal is to determine the optimal solution for the task-unit assign- pected that some processing units are under operation with allocated
ment and sequencing, sequence dependent changeover/set-up, the production orders. So, the disturbance of new orders arrival could
temporary storage allocation, produced campaign lots, sales/late de- justify the revision of current plans or evaluate operational alternatives.
liveries and maintenance tasks schedule of process units, given: The model responsiveness should provide the decision-maker a com-
parison and analysis of solutions, as well as relevant KPI computed at
(i) the product recipes in terms of their respective RTN framework; the post-processing phase. Moreover, multi-objective approach method
(ii) the time horizon, product demands and due dates; (e.g. augmented ε-constraint method by Mavrotas, 2009) can also be
(iii) the characteristics of the processing/storage units allocated and implemented to analyse Pareto sets according to strategic process in-
sequence-dependent changeover/setup times; dicators, providing a useful insight on how the production reacts to
(iv) processing times and the task-unit suitability; changes in each criterion (Korošec, Bole, & Papa, 2013; Vieira, Pinto-
(v) the shelf-life storage of intermediaries/products; Varela, & Barbosa-Póvoa, 2017a).
(vi) the manufacturing, changeover and storage costs for all materials, With a preliminary study of a model approaches suitable to the
the value of the products, and late delivery penalties; industrial problem, we have researched in the literature a MILP for-
(vii) the duration and cost of maintenance operation, with information mulation to address the main bioprocess constrains. Noteworthy, the
of the decaying yield with the number of batches produced of the lead work by Lakhdar et al. (2005) firstly proposed a discrete- time
corresponding process unit. MILP model for the optimal production and cost effective sequence of
manufacturing tasks for a medium term horizon, underlining the ad-
Finally, the objective function can be defined according to the vantages of planning optimisation for industrial practices. More

358
M. Vieira et al. Computers & Industrial Engineering 129 (2019) 354–367

recently, Kabra et al. (2013) developed a multiple-grid continuous-time 3.2.1. Resources balance constraints
model for multi-product campaign scheduling of biopharmaceutical Regarding the resources constraints given by Eqs. (1)–(7), the
processes based on State Task Network (STN) framework and Liu et al. availability in each time event t considers the amount available in the
(2014) later proposed the simultaneous production and maintenance previous event adjusted by the amounts discretely or continuously
planning under a performance decay by extending Lakhdar et al. (2005) consumed/produced by all tasks starting or ending on that event. By
discrete time model. Both these models were compared in the recent considering that in each step could exist equipment technology options
work by Vieira et al. (2017b), which have discussed the advantages of a with either batch and continuous operation modes, it is necessary to
continuous-time single-grid formulation in an integrated production assure the correct task sequencing of the corresponding batch and
and maintenance planning of processes under performance decay, with continuous tasks in consecutive time intervals given by the optimisation
improved planning solutions applied to industrial-based examples. The solution. Following the same reasoning discussed by Vieira et al.
proposed model now extends the formulation by Vieira et al. (2017b) to (2016), along with the set of continuous tasks which consumes an in-
address process steps with either batch and continuous operation mode, termediate material r produced by both operation modes, IrBCc
, it is now
in order to generalise the multiple technological process equipment necessary to define a subset of continuous tasks that produced the same
possibilities. Therefore, additional terms are introduced in the con- intermediate r in the previous process step, IrHyb
c
, to assure that both
straints to consider either batch Ib and continuous Ic tasks, since the continuous tasks (production/consumption) can occur at the same time
model proposed by Vieira et al. (2017b) assumed that all equipment/ interval [t , t + 1].

Rr , l, t = Rr0, l t =1 + Rr , l, t 1t>1 + Rrp, l, t + Rrc, l, t Wr , l, t + r , l, t r M, l Lr , t T (1)

Rrp, l, t = vp
i Ib i, r i, l, t 1
+ c
i Ic (Ima IrHyb ) i, r i, l, t 1 + i Ima i, r FPi, l, t + c
i IrHyb i, r i, l, t

r M, l Lr , t T (2)

tasks in each process step had a continuous operation mode. This im-
Rrc, l, t = vc + +
plies that the formulation now considers the diverse balance and se- i Ib i, r i, l, t c
i Ic IrBC i , r i, l , t 1 c
i IrBC i, r i, l, t

quencing of these tasks, since batch tasks interact with resources dis- r M B, l Lr , t T (3)
cretely at its start and finish, while continuous tasks flow continuously
at a rate that remains constant throughout its duration. By process
definition, in a batch operation mode the materials are entirely con-
l LBr
Rrc, l, t = l LBr ( vc
i Ib i, r i, l, t
+ c
i Ic IrBC i, r i , l , t 1 + c
i IrBC i , r i, l , t )
r B, t T
sumed at the start of the respective production task, and the amount
(4)

Rr , t = Rr0 + Rr , t 1 + (µi,pr Ni, l, t 1 + µic, r Ni, l, t ) r E, t T


(t = 1) t>1 i Ib Ic l Lr (5)

produced is made available only at its finish. However, in a continuous


0 Rr , l, t Rrmax r M, t H
operation, a production flow rate is verified along the duration of the
l Lr (6)
task, which allows that sequential continuous tasks can occur simulta-
neously. A resume of the mathematical formulation is here presented,
the detailed explanations can be found in the seminal paper. Con- Rr , l, t = 0 t= T
sidering the indices, sets, parameters and variables notation detailed in r M RM l Lr (7)
Appendix A, it should be noted that the main decision variables respects

Tt + 1 Tt i Ib Ic
µirp ( l Li
( i Ni, l, t + i i, l, t |i Ib + i, l, t i
max
|i Ic ) + i Ib Ic i, i )
Ci, i , t + 1 + r MAr , t

r E Est , t T, t |T | (8)

Tt + 1 Tt H 1 ( i Izw Imr l Li )
µi,pr Ni, l, t + i Izw Imr
µi,pr ( l Li
( i Ni, l, t + i i, l, t |i Izw + i, l , t i
min
|i Imr ) + i Izw Imr i, i )
Ci, i , t + 1 + r MAr , t

r E Est , t T, t |T | (9)

the RTN representation given by Ni, l, t (binary) and i, l, t (continuous/ 3.2.2. Timing constraints
integer), respectively related to the occurrence of task i (batch or con- The timing constraints given by Eqs. (8)–(15) considers the duration
tinuous) of lot l in the slot [t , t + 1] of a common time-grid to the of the interval composed by a single processing task (batch or con-
equipment resources. tinuous) per unit plus, the corresponding sequence-dependent change-
over/setup time (feasible connections given by duration parameter i, i )

359
M. Vieira et al. Computers & Industrial Engineering 129 (2019) 354–367

or maintenance operations required, and can be bounded when tasks l Lr


( r , l, t ) t T
d Dr
(Yt , d Qr , d ) t T
l Lr
( r , l, t )
are subject to zero-wait policies. The duration of each processing task is t t t <t

related to parameters i, i , for batch tasks, and the processing rates r P, t T, t > 1 (19)
min
, imax for continuous tasks. And considering a set of time events t to (20)
i r , l , t =0 r M P , l Lr , t T
be arbitrary placed in a given horizon H, [T 0, T H ], the allocation of each
multiple due date d is optimised with variable Yt , d .
3.2.5. Operational constraints
Yt , d = 1 d D
The boundaries of each task extent are limited by Eqs. (21) and (22),
t T
t 1 (10) by the capacity of the equipment Vimax
,l in the case of batch tasks, or to a
maximum campaign length Campimax ,l in the case of continuous tasks.
Tt hd Yt , d t T, t 1 Lower bounds can be useful to impose a minimum batch/campaign size
d D (11)
Vimin
, l Ni, l, t i, l, t Vimax
, l Ni, l, t i Ib, l Li , t T, t T (21)
Tt hd Yt , d + H 1 Yt , d t T, t 1 max
Campimin Ni, l, t Campimax max
Ni, l, t i Ic , l Li , t T, t T
d D d D (12) ,l i i , l, t ,l i

(22)
Tt + 1 Tt 1 t T, t T (13)

T1 = T 0 (14) 3.2.6. Sequence-dependent changeover/setup constraints


Eqs. (23)–(26) constraint the sequence of tasks for changeover
TT TH (15)
procedures, considering also the allocation to each task the required
setup time for every new production. The model ability to allocate this
3.2.3. Lot constraints changeover time to the boundary intervals of a time event needs to
Regarding the lots of material resources that are able to be blended/ avoid the overlap of sequential continuous tasks, therefore, this feature
splitted, Eqs. (16) and (17) controls the lot assignment constraints, is only applied to batch process steps and to the first step of a con-
where for a given set of lots, the sequential allocation of lot number tinuous process sequence.

l Li
µi,pr Ni, l, t 1 + l Li
µi p, r Ni , l, t i Ib Ic l Li
t T µi p, r Ni , l, t 1 + µi,pr µip, r Ci, i , t |i Ib Ic
first + µi,pr µip, r Ci, i , t + 1
t t <t t >t
r EEst , i Ib Ic , i Ib Ic , i i, t T , t > 1, t T, t t, t |T| (23)

index l is able to provide the tracking control of each productions


throughout the planning horizon for regulatory protocols.

vi,pr Ni, l, t vi,pr Ni, l 1, t


vip, r Ni , l 1, t
+ i, r Ni, l, t i, r Ni, l 1, t i , r Ni , l 1, t 0
t T i Ib t T t T i Ic t T
t t 1 i i t t t t 1 i i t t
i Ib i Ic

r M RM , i Ib Ic , l Lr , l > 1, t T, t T (16)

p
t T vi, r Ni, l, t + t T i, r Ni, l, t 1
i Ib
t |T|
i Ic
t |T| l Li
µi,pr Ni, l, t i Ib Ic l Li
t T µi p, r Ni , l, t µi,pr Ci, i, t + 1 + µi,pr Ci, i, t |t > 1
t <t
(17)
t>1
r M RM , l Lr
r Esu, i Ib Ic , t T, t |T| (24)

3.2.4. Multiple deliveries constraints i Ib Ic µi,pr µ i p, r Ci, i , t 1


i Ib Ic
Eqs. (18)–(20) considers the total demand profile with corre- i i

sponding due dates per each final product Qr , d , while the amount of r E Est , t T , t > 1, t |T| (25)
production lots delivered r , l, t is balanced with the unmet demand ur , l, t
in each time event t .
µi,pr C 1 r Esu
u u i, i, t
(Yt , d Qr , d) ( r , l, t ) i Ib Ic t T
d Dr l Lr r , l, t l Lr l Lr r , l, t 1
t >1 (26)
r P, t T, t > 1 (18)

360
M. Vieira et al. Computers & Industrial Engineering 129 (2019) 354–367

3.2.7. Storage and Shelf-life constraints NMr , t 1 (j 1) + NMrmax ·(1 + MAr , t 1 ABr , j, t ) r Ema, j, t > 1
Eqs. (27)–(35) accounts for the storage and shelf-life constraints, (40)
assessing the general case of a storage task activation in the interval
[t , t + 1] when an excess material of an intermediary/product is verified NMr , t j NMrmax ·(1 ABr , j, t ) r Ema, j, t > 1 (41)
at event t . The amount of material wasted Wr , l, t is determined when
shelf-life is exceeded Si, l, t , t and the corresponding storage tasks are QMr , j, t 2 r Ema, t > 1
j (42)
verified in the interval [t , t ], t > t . The variable x r , l, t is also modelled to
account of a storage time cost penalty in the objective function (Section QMr , j, t ABr , j, t ABr , j 1, t r Ema, j > 1, t > 1 (43)
4.2.9).
QMr , j, t ABr , j 1, t ABr , j, t r Ema, j > 1, t > 1 (44)
Vimin
, l Ni, l, t Rr , l, t Vimax
, l Ni, l, t i IstB IstC , l Li , t T, t T
r Irst
ABr , j, t + 1 = µi,pr i, l, t 1 r Ema, t > 1
(27) j i Ir l Lr (45)
(Tt Tt ) HSi, l, t , t i Ist , l Li , t T, t T , t > t, t T FPrtot
,t = ydr , j · ABr , j, t r Ema, t > 1
i, l

(28) j (46)
(Tt Tt ) H (Si, l, t, t 1) I Ist , l Li , t T, t T , t > t, t T FPrtot µi,pr FPi, l, t
,t = r Ema, t > 1
i, l

(29) i Ir l Lr (47)

Wr , l, t Rr , l, t 1 Vstmax 1 + (t t) Si, l, t , t + Ni, l, t r M RM , l Lr , t T , t < |T|, t > t


i Irst t t <t (30)

Wr , l, t Rr , l, t 1 r M RM , l Lr , t T, t > 1 (31) FPi, l, t i, l , t 1 i Ima , t > 1 (48)

Wr , l, t =0 r M RM , l Lr
t T (32) NMr , t = NMr0 r Ema, t = 1 (49)

x r , l, t (Tt + 1 Tt ) (
H 1 i Irst
Ni, l, t )
r M RM , l Lr , t T , t < |T| (33) 3.2.9. Objective functions
In compliance with the decision-maker needs, the solution objective
xr , l, t (Tt + 1 Tt ) evaluates the profit maximisation of final products sales rs by pena-
r M RM , l Lr , t T , t < |T| (34) lising the costs of manufacturing cimf , storage allocations crst , inter-
mediate changeovers/setup cich
, i , extended shelf-life wasted amounts cr ,
d

x r , l, t H Ni, l, t r M RM , l Lr , t T , t < |T| late backlogs c r and cost of maintenance operations cr , given by
u ma

i Irst (35) equation (50).

max profit = r( r,l,t ) c mf


i i,l,t c str r R r,l,t + x r,l,t 60 Nii,l,t c ch
i,i Ci,i ,t
r P l Lr t T i I l Li t T r Mst l L r t T i I i I i I t T
t >1 t |T| t |T| i i t> 1

c dr Wr,l,t c ru u
r,l,t cma
r MA r,t
r M RM l L r t T r P l Lr t T r Ema t T
t> 1 t> 1 t >1 (50)

3.2.8. Maintenance constraints 3.3. Decision-making approach


The maintenance operations are constraint by Eqs. (36)–(49), where
the binary variable MAr , t indicates whether the maintenance is per- The complexity of the overall decision process in an industrial plant
formed in equipment resource subject to maintenance r Ema at time can be perceived with the enterprise functions network of the Purdue
event t , accounting the number of productions with NMr , t and con- reference model from ISA-95 (Fig. 4). Whereas planning and scheduling
sidering that each plan starts with a defined maintenance condition. decisions can be a function of the planning department, they generally
involve the collaboration and analysis of other areas such as sales,
NMMr , t NMrmax ·MAr , t 1 r Ema, t > 1 (36) procurement or R&D. The challenge for a decision support system relies
in the definition of the core decision processes able to be optimised,
NMMr , t NMr , t 1 r Ema, t > 1 (37) which can vary according to the company’s strategies and the scheduler
expertise.
NMMr , t NMr , t 1 NMrmax ·(1 MAr , t 1) r Ema, t > 1 (38) The assessment of the solutions provided should not disregard that
some simplifications are required for the formulation of the problem.
NMr , t = NMr , t 1 NMMr , t + ABr , j, t r Ema, t > 1 For example, in some cases a linear approximation/aggregation of
j (39) constraints is needed to keep the problem computationally tractable,

361
M. Vieira et al. Computers & Industrial Engineering 129 (2019) 354–367

Fig. 4. Purdue enterprise-control model with organisational structure of functions (ANSI/ISA-95, 2000).

else it could lead to infeasible solutions. Thus, the quality of the solu- approaches, the proposed ICOMPASS tool relies on a centrally hosted
tions proposed and the computational time required is a critical factor secure system based in cloud servers with API interfaces, attending the
for the decision-making. Considering the proposed model approach, the issues related to maintenance and evolution of software, promoting
continuous-time formulation requires an iterative search for the updated integrations with real-time plant management platforms and
number of time events to find the optimal solution for a given problem allowing a universal access from web-enabled devices. Acknowledging
case. To implement an automatic procedure in the decision support that the complexity of the IT programming system represents a field
tool, it was defined, based in previous model studies, that the optimal that far extends the scope of this work, the detailed inferences of the
schedule proposed should be found in a solution pool of up to n + 4 software architecture for the platform application will be discussed in
time events, being n the number of due dates, and that each model run future works.
should not exceed 3600 s of computational time. By defining a time Finally, the successful implementation of the decision support
limit for the solutions output for a given problem, it is possible that the system will require a progressive adjustment to the production en-
optimality gap could not be closed due to computational limitations, vironment. The assessment of the different stakeholders will benefit
being the solution quality subject to the decision-maker criteria. Despite from a preliminary test-case development to show the potentialities of
this procedure is find suitable to the proposed problem structure, the the optimisation system. This deployment should point the collabora-
algorithm should evolve according to the diverse requirements by in- tive process and test the robustness of the model performance applied
tegrating new types of model constraints, alternative mathematical to different problem scales, adapting the capabilities of the decision
approaches or empirical planning rules to convey additional decisional support tool according to the industrial requirements. As follows, we
objectives (e.g. matheuristics and simulation-optimisation approaches). report the implementation of the ICOMPASS tool applied to a case study
R1-2An effective decision support system lastly requires an inter- in the biopharmaceutical industry.
active flow of information among the different stakeholders of the
process decisions. The development of a suitable software-based tool is 4. Case-study application
essential to promote an integration with existing plant management and
regulatory systems, allowing the access to the process data and the The architecture of the ICOMPASS prototype tool follows the pre-
analysis of solutions (Saha, Aqlan, Lam, & Boldrin, 2016). The im- viously discussed implementation framework of a computerised system
plementation of the software application is typically assigned to IT for the decision support on planning optimisation of a biopharmaceu-
consultants, to incorporate all the identified specificities of the decision- tical production. The structure of the collaborative solution approach is
making in a collaborative process. A user–friendly visual interface able to assess the major decisions in a correct way, understanding its
should be developed with the relevant planning information displayed model limitations when the complexity cannot be overcome, and match
in Gantt charts, and well as advanced options to access solution details the decision maker expectations and needs of its users. The platform
to support the process optimisation analysis. In this work, we propose a provides easy-to-use visual interfaces and access via mobile devices,
concept of a decision support platform provided as a “Software as a allowing the decision-maker to assess the process data and analyse
Service” for the analysis, management and optimisation of an industrial different scenarios to take optimal planning decisions. The prototype
production planning operation. Following current technology was developed in collaboration with the industrial partners, balancing

362
M. Vieira et al. Computers & Industrial Engineering 129 (2019) 354–367

Fig. 5. ICOMPASS prototype architecture with the information flows and interaction.

their decision-making protocols, functional requirements and the flow solutions are generated from the optimisation algorithm, composed by
of communication among different information layers, with the struc- Eqs. (1)–(50) and implemented in GAMS modelling software. Using MS
ture displayed in Fig. 5. Excel interface capabilities to provide the input/output source data, the
The user interaction with the ICOMPASS tool is sourced at the solution is generated using the high performance commercial solver
Planning department level to set the planning order, edit the database CPLEX supported by GAMS. A Model manager is advised to control the
with the operational parameters (or preferable connection with MES, performance of generated solutions, assess the reliability of data input/
ERP and related systems), and analyse the generated optimal planning output processing, and review the mathematical formulation to address
solutions through interactive charts and KPI reports. The data is ac- additional production/process constraints to the structure. As follows,
cessed through web-interfaced menus linked to MS Office tools and the we define a planning problem example to discuss the application of the

363
M. Vieira et al. Computers & Industrial Engineering 129 (2019) 354–367

Fig. 6. Plan with optimal solution (464.7 real monetary units).

concept tool to a biopharmaceutical industrial case-study. sales/late deliveries and maintenance tasks schedule.
The ICOMPASS interface provides the interactive visualisation of
4.1. ICOMPASS application example the raw planning solution chart with a link to the inventory profile. For
this example, the reported optimal planning solution displayed in Fig. 6
A case-study was designed from the collected information of in- (with the user interface display on the left) was obtained with nine
dustrial cases (Vieira et al., 2016), considering a biopharmaceutical event points, considering the CPU time limit imposed of 3600 s. The
company with the following process characteristics: scheduler can visualise the production plan intervals tracking the size/
length of campaign lots [L#] of the four products (P_Bio# refers to final
• The company operates in 24/7 shifts production with a portfolio product while I_Bio# to the intermediate product in upstream proces-
sing), with thirteen intermediate sequence-dependent changeovers
composed by four biopharmaceutical products (Bio1-4) and has
available a set of certified equipment (with batch and continuous ([CO] mark), eight storage activations ([S] mark) and eight inter-
operation) which are aggregated in upstream and downstream suites mediate maintenance operations ([M] mark). In brackets are identified
for planning purposes: the company has recently invested in up- the amounts processed in each task, being the numbers in downstream
stream capacity by adding to a batch fermentation reactor, UPX1, units subject to yield decay. It must be noted that while UPX1 operate in
two new continuous perfusion reactors, UPX2 and UPX3, while as a batch mode, which implies that the amount of an intermediate pro-
downstream suites there are two continuous chromatography sets, duction is only available at the end of the task, the ones processed in
DWX1 and DWX2; UPX2 and UPX3 operate in a continuous mode. For example, lot [L2(6)]
• The network assessment of the process has characterised all avail- of P_Bio2 is scheduled to occur in the [120, 180 days] interval of DWX1,
consuming the lots of intermediary I_Bio2 [L3(4)] (batch) and [L4(2)]
able production suites in terms of products suitability, operational
limits, process economics and regulatory norms compliance (lots, (continuous). Also by definition, only if the amount processed in batch
storage, changeover, maintenance, etc.), defining the blend of lots is not consumed immediately, a storage activation is required, e.g.
allowed only to Bio1 and Bio2 production for regulatory purposes; I_Bio2[L5].
• The processing times of upstream/downstream stages, which re- Further, to allow a reactive approach to production disturbances, an
interface option in ICOMPASS platform allows a rescheduling strategy
presents an aggregated production cycle for an average integer
amount, are assumed constant for each suite specifications and re- of a planning solution already under process. For example, when the
vised accordingly by the scheduler; demand of a certain product order is requested to be increased, the
• The planning approach considers unlimited storage units, assuming decision-maker can evaluate if the utilisation rate of its production
equipment in the ongoing plan have the capacity to accommodate that
that the cost of product storage is minor compared to the revenue
loss of unfulfilled deliveries; production increase. Since all equipment share the same planning grid,
• Regarding the downstream units, the yield decay is assumed con- it is possible to continue the initial plant until a certain time event, from
which the demand profile will be changed. Considering the planning
stant and the planning of maintenance operations is accounted for
the chromatography protein-A resins, while the uncertainty in up- solution of Fig. 6, the case now considers that on the 250th day of the
stream production titres is considered negligible (each new plan is ongoing plan the order of Bio3 for the 540th day is requested to be
considered to start with a regenerated resin, MAr , t = 1 t = 1). increased in seven batches. It was approved by the production sche-
duler to follow the initial plan until 300th day event of the ongoing
Considering the plant production network mapped, the ICOMPASS plan, from which a new planning solution is calculated taking into
platform allows the scheduler to define a planning problem by allo- account stored materials, downstream yield status and updated order
cating to the set of equipment suites a given production demand hor- backlogs, fixing those values for the initial time event. In this solution
izon with multiple deliveries, an obtain a solution according to the the model formulation disregards the setup time of the first replanned
selected objective function, in this case, the profit maximisation. The tasks, considering the timing constraints A.1 and A.2 instead of Eqs. (1)
plan outlined 56 orders of the four products for six delivery dates, ad- and (2). In Fig. 7 is shown the reported optimal solution displayed in
ditional information regarding problem data is available from the au- the platform for the production plan from the 300th day event onwards,
thors upon request. The optimal planning solution is generated in a host obtained with five time events. Additional KPI reports allows the
server by GAMS/CPLEX (GAMS 24.5.6 WIN VS8 ×86) on an Intel Xeon scheduler to analyse that the new plan provides a demand satisfaction
X5680 64-bit at 3.33 GHz with 24 GB of RAM and the output results are of 98.3% (1.05 left), a similar result to the initial plan, 98.6% (0.75
stored in the company’s database to generate the Gantt chart solution left), but with improved results on sales profit and capacity utilisation,
for the task-unit assignment and sequencing, sequence dependent whereas these minor backlogs can be satisfied before the beginning of
changeover/set-up, the storage allocation, campaign-lots production, the next production plan.

364
M. Vieira et al. Computers & Industrial Engineering 129 (2019) 354–367

Fig. 7. Plan with optimal solution for review plan (314.6 real monetary units).

5. Conclusions robust user-control planning tools. The ICOMPASS platform represents


a prototype concept to explore these operational problems and, in close
In this work, we discussed the importance of decision support collaboration with industrial players, to potentiate the implementation
methodologies into industrial production optimisation, surveying in- of planning/scheduling methodologies for the optimisation of industrial
dustrial successful applications to structure a framework solution. The processes. One of the major issues identified relies on how the in-
development of a novel solution application for the biopharmaceutical formation is accessed and translated to the decision-maker in order to
industry with a model-based decision support tool ICOMPASS, was be perceived the competitive advantages with the corresponding op-
validated in an industrial proof-of-concept study for the production and erational and economic gains. As future work, ongoing studies with
maintenance planning optimisation of a biopharmaceutical process industrial partners will enable the development of additional features
under performance decay. The generic approach proposed for the de- concerning the interactivity with the user-interface (e.g. software
cision support tool is highlighted with the application to a bio- structure and updates, connection with enterprise systems, input/
pharmaceutical production optimisation, matching the decision-maker output data review, solution sensitivity assessment) and evaluation of
requirements with a flexible modular structure that suits these in- alternative mathematical optimisation methodologies to deal with
dustrial problems with planning/replanning solutions. And despite the larger planning problems with more detailed activities and sub-pro-
implementation challenges of these tools in real production environ- cesses variability (e.g. matheuristics and simulation-optimisation ap-
ments, due to inherent complexity of multiple operational criteria, proaches).
successful industrial applications proved that the investment on deci-
sion support solutions potentiates several advantages to the decision-
making process. Acknowledgments
The deployment of integrated, user-friendly and efficient decision
support platforms still requires an extensive research work, either from The authors would like to acknowledge the financial support of
the modelling point of view to tackle different mathematical ap- Fundação para a Ciência e Tecnologia under the grant SFRH/BD/
proaches (for example, the uncertainties constraints in biopharmaceu- 51594/2011 and from FEDER through the POCI program under the
tical manufacturing), either in the continuous development of more Project POCI-01-0145-FEDER-016418.

Appendix A

• Notation
Indices
d delivery dates
i, i task
l, l lot
r resource (process unit, intermediate or final product)
j number of batches for performance decay tracking
t, t , t event point
Sets
B resources in which lots are allowed to be blended
D delivery dates
Dr delivery dates associated with resource r
E processing equipment
Est subset of equipment related to storage
Ema subset of equipment subject to maintenance
Eutil subset of equipment to verify utilisation rate
Esu subset of equipment subject to initial setup
I process tasks
Ir tasks that require resource r
Ib batch tasks
Ic continuous tasks
Icup continuous tasks (upstream step)

365
M. Vieira et al. Computers & Industrial Engineering 129 (2019) 354–367

Ist storage tasks


Izw tasks subject to zero-wait policies
Imr tasks that must exceed a certain minimum rate
c
IrBC subset of continuous tasks that consumes the intermediate material r produced by a batch task (or in a hybrid operation step)
c
IrHyb subset of continuous tasks that produces the intermediate material r in a hybrid operation step
IstB , IstC subset of storage tasks related to intermediates and final products produced by batch and continuous tasks
Irst storage tasks associated with material resource r
Ima Processing tasks subject to performance decay in equipment subject maintenance
L lots
Lr lots associated with resource r
Li lots associated with taski
LBr lots of resourcer B that are able to be blended
M material resources
Mst subset of material resources able to be stored
P subset of final products
RM subset of raw material resources
T event points
Parameters
i constant term in the processing time of task i
i proportional term to the extent variable in the processing time of task i
µic, r allocation coefficient for the binary extent of resource r (equipment unit) in task i relative to the start of the task
µi,pr release coefficient for the binary extent of resource r (equipment unit) in task i relative to the end of the task
c
i, r consumption coefficient for the continuous extent of r resource (intermediary or final product) in task i relative to the start of task
p
i, r
production coefficient for the continuous extent of resource r (intermediary or final product) in task i relative to the end of task
r, i, r coefficient for the rate of consumption of resource r (production/consumption of task i)
cimf cost of manufacturing processing task i
cru backlog penalties cost of material resource r
crd waste disposal cost of material resource r
crst storage cost of material resource r
cich sequence dependent task {i, i′} changeover cost
,i
crma maintenance cost at equipment unit r
r sales price of resource r
H time horizon
hd absolute time of demand point d
i, i time relative to sequence changeover/setup {i, i′}
i, l lifetime of the processed products of the task i lot l
Campimin max minimum and maximum campaign length of the processing task i lot l
, l , Campi, l
min
, imax minimum and maximum allowable rate of task i
i
Qr , d demand amount of each resource at delivery date d
Rr0 , Rr0, l resource availability in the beginning of the planning horizon
Rrmax maximum resource availability of resource r
Vimin max
, l , Vi, l
minimum and maximum capacity of resource r (processing units) for task i of lot l
ydr ,j yield of resin for the batch labelled j in unit r
NMrmax maximum number of batches production allowed after maintenance in unit r
Variables
Ci, i , t binary variable that assigns a sequence dependent changeover procedure of tasks i ,i before time event t (i, i corresponds to initial setup)
Ni, l, t binary variable that assigns the task i of lot l to start at event point t
i, l, t total amount of material processed by task i and lot l starting at event point t
Rr , l, t resource availability r of lot l and time point t
Rrc,l, t , Rrp, l, t amount consumed/produced of resource r of lot l and time point t
Si,l, t, t binary variable that accounts when the product lifetime stored through task i lot l was extended within the interval [t, t′]
xr , l, t accounts for the absolute storage time if storage task i lot l is active in the event interval [t, t′]
Wr , l,t waste disposal amount when resource shelf-life is exceeded
Yt ,d binary variable that assigns a specific event point t corresponds to a demand points d
r , l, t amount expedited at a corresponding due date
u
r , l, t unfulfilled demand of product r of lot l
MAr , t binary variable to assigns if unit r is maintained at event t
NMr ,t number of batches of final products produced in unit r since last maintenance by event t
NMMr ,t auxiliary variable ≡ MAr , t × NMr , t
ABr , j, t auxiliary binary variable to assign if the batch labelled j is produced in unit r at even t
QMr , j,t Auxiliary binary variable ≡ |ABr,j,t − ABr,j-1,t|, 0 if two consecutive batches labelled j and j-1 in unit r are produced at event t, otherwise 1
FPi, t amount produced by task i starting at time event t considering yield decay
FPrtot
,t
total amount of final product r starting at time event t

• Auxiliary equations
Tt + 1 Tt i Ib Ic
µirp l Li i Ni, l, t + i i, l, t |i Ib + i, l, t
max + i Ib Ic , i i i, i
Ci, i , t + 1 + r MAr , t
i i Ic

r E\Est , t T, t |T | (A.1)

366
M. Vieira et al. Computers & Industrial Engineering 129 (2019) 354–367

Tt + 1 Tt H 1 ( i Izw Imr l Li )
µi,pr Ni, l, t + i Izw Imr
µi,pr l Li i Ni, l, t + i i, l, t |i Izw + i, l, t

i
min
i Imr
+ i Izw Imr , i i i, i
Ci, i , t + 1 + r MAr , t

r E Est , t T, t |T| (A.2)

References chemical production scheduling. AIChE Journal, 58(6), 1812–1828.


Mavrotas, G. (2009). Effective implementation of the ε-constraint method in multi-ob-
jective mathematical programming problems. Applied Mathematics and Computation,
ANSI, ISA-95.00.01-2000 (2000). Enterprise-control system integration – Part 1: Models and 213(2), 455–465.
terminology. NC, 27709: ISA: Research Triangle. Méndez, C. A., Cerdá, J., Grossmann, I. E., Harjunkoski, I., & Fahl, M. (2006). State-of-the-
ANSI, ISA-95.00.03-2005 (2005). Enterprise-control system integration – Part 3: Activity art review of optimization methods for short-term scheduling of batch processes.
models of manufacturing operations management. NC, 27709: ISA: Research Triangle. Computers & Chemical Engineering, 30(6–7), 913–946.
Barbosa-Póvoa, A. P. (2012). Progresses and challenges in process industry supply chains Moniz, S., Barbosa-Povoa, A. P., & de Sousa, J. P. (2014a). Simultaneous regular and non-
optimization. Current Opinion in Chemical Engineering, 1(4), 446–452. regular production scheduling of multipurpose batch plants: A real chemical–-
Biel, K., & Glock, C. H. (2016). Systematic literature review of decision support models for pharmaceutical case study. Computers & Chemical Engineering, 67, 83–102.
energy-efficient production planning. Computers & Industrial Engineering, 101, Moniz, S., Barbosa-Povoa, A. P., de Sousa, J. P., & Duarte, P. (2014b). Solution metho-
243–259. dology for scheduling problems in batch plants. Industrial & Engineering Chemistry
Castro, P. M., Grossmann, I. E., Veldhuizen, P., & Esplin, D. (2014). Optimal maintenance Research, 53(49), 19265–19281.
scheduling of a gas engine power plant using generalized disjunctive programming. Ngai, E. W. T., Peng, S., Alexander, P., & Moon, K. K. L. (2014). Decision support and
AIChE Journal, 60(6), 2083–2097. intelligent systems in the textile and apparel supply chain: An academic review of
Choy, K. L., Leung, Y. K., Chow, H. K. H., Poon, T. C., & Kwok, S. K. (2011). A hybrid research articles. Expert Systems with Applications, 41(1), 81–91.
scheduling decision support model for minimizing job tardiness in a make-to-order Otto, R., Santagostino, A., & Schrader, U. (2014). Rapid growth in biopharma: Challenges
based mould manufacturing environment. Expert Systems with Applications, 38(3), and opportunities. McKinsey Pharmaceutical and Medical Products Insights.
1931–1941. Pantelides, C. C. (1994). Unified frameworks for optimal process planning and scheduling.
Costa, A. (2015). Hybrid genetic optimization for solving the batch-scheduling problem in Proceedings on the second conference on foundations of computer aided operations. New
a pharmaceutical industry. Computers & Industrial Engineering, 79, 130–147. York: Cache Publications.
Dehghanimohammadabadi, M., Keyser, T. K., & Cheraghi, S. H. (2017). A novel Iterative Pehrsson, L., Ng, A. H., & Stockton, D. (2013). Industrial cost modelling and multi-ob-
Optimization-based Simulation (IOS) framework: An effective tool to optimize sys- jective optimisation for decision support in production systems development.
tem’s performance. Computers & Industrial Engineering, 111, 1–17. Computers & Industrial Engineering, 66(4), 1036–1048.
Eberle, L., Capón-García, E., Sugiyama, H., Graser, A., Schmidt, R., & Hungerbühler, K. Power, D. J., & Sharda, R. (2007). Model-driven decision support systems: Concepts and
(2016). Rigorous approach to scheduling of sterile drug product manufacturing. research directions. Decision Support Systems, 43(3), 1044–1061.
Computers & Chemical Engineering, 94, 221–234. Saha, C., Aqlan, F., Lam, S. S., & Boldrin, W. (2016). A decision support system for real-
Farid, S. S., Washbrook, J., & Titchener-Hooker, N. J. (2007). Modelling biopharmaceu- time order management in a heterogeneous production environment. Expert Systems
tical manufacture: Design and implementation of SimBiopharma. Computers & with Applications, 60, 16–26.
Chemical Engineering, 31(9), 1141–1158. Shah, N. (2004). Pharmaceutical supply chains: key issues and strategies for optimisation.
Figueira, G., Amorim, P., Guimarães, L., Amorim-Lopes, M., Neves-Moreira, F., & Almada- Computers & Chemical Engineering, 28(6–7), 929–941.
Lobo, B. (2015). A decision support system for the operational production planning Shaik, M. A., Dhakre, A., Rathore, A. S., & Patil, N. (2014). Capacity optimization and
and scheduling of an integrated pulp and paper mill. Computers & Chemical scheduling of a multiproduct manufacturing facility for biotech products.
Engineering, 77, 85–104. Biotechnology Progress, 30(5), 1221–1230.
Grossmann, I. (2005). Enterprise-wide optimization: A new frontier in process systems Shaik, M. A., Floudas, C. A., Kallrath, J., & Pitz, H. J. (2009). Production scheduling of a
engineering. AIChE Journal, 51(7), 1846–1857. large-scale industrial continuous plant: Short-term and medium-term scheduling.
Grossmann, I. E. (2012). Advances in mathematical programming models for enterprise- Computers & Chemical Engineering, 33(3), 670–686.
wide optimization. Computers & Chemical Engineering, 47, 2–18. Shim, J. P., Warkentin, M., Courtney, J. F., Power, D. J., Sharda, R., & Carlsson, C. (2002).
Harjunkoski, I., Maravelias, C., Bongers, P., Castro, P., Engell, S., Grossmann, I. E., ... Past, present, and future of decision support technology. Decision Support Systems,
Wassick, J. (2014). Scope for industrial applications of production scheduling models 33(2), 111–126.
and solution methods. Computers & Chemical Engineering, 62, 161–193. Siganporia, C. C., Ghosh, S., Daszkowski, T., Papageorgiou, L. G., & Farid, S. S. (2014).
Janak, S. L., Floudas, C. A., Kallrath, J., & Vormbrock, N. (2006). Production scheduling Capacity planning for batch and perfusion bioprocesses across multiple biopharma-
of a large-scale industrial batch plant. I. Short-term and medium-term scheduling. ceutical facilities. Biotechnology Progress, 30(3), 594–606.
Industrial & Engineering Chemistry Research, 45(25), 8234–8252. Stefansson, H., Sigmarsdottir, S., Jensson, P., & Shah, N. (2011). Discrete and continuous
Kabra, S., Shaik, M. A., & Rathore, A. S. (2013). Multi-period scheduling of a multi-stage time representations and mathematical models for large production scheduling pro-
multi-product bio-pharmaceutical process. Computers & Chemical Engineering. 57, blems: A case study from the pharmaceutical industry. European Journal of
95–103. Operational Research, 215(2), 383–392.
Kallestrup, K. B., Lynge, L. H., Akkerman, R., & Oddsdottir, T. A. (2014). Decision support Stephanopoulos, G., & Reklaitis, G. V. (2011). Process systems engineering: From Solvay
in hierarchical planning systems: The case of procurement planning in oil refining to modern bio- and nanotechnology. A history of development, successes and pro-
industries. Decision Support Systems, 68, 49–63. spects for the future. Chemical Engineering Science, 66(19), 4272–4306.
Kondili, E., Pantelides, C. C., & Sargent, R. W. H. (1993). A general algorithm for short- Varma, V. A., Reklaitis, G. V., Blau, G. E., & Pekny, J. F. (2007). Enterprise-wide modeling
term scheduling of batch-operations. 1. Milp formulation. Computers & Chemical & optimization—An overview of emerging research challenges and opportunities.
Engineering, 17(2), 211–227. Computers & Chemical Engineering, 31(5), 692–711.
Kopanos, G. M., Méndez, C. A., & Puigjaner, L. (2010). MIP-based decomposition stra- Vieira, M., Pinto-Varela, T., & Barbosa-Póvoa, A. P. (2015). Periodic versus non-periodic
tegies for large- scale scheduling problems in multiproduct multistage batch plants: A multipurpose batch plant scheduling: a paint industry case study. In J. P. Almeida, J.
benchmark scheduling problem of the pharmaceutical industry. European Journal of F. Oliveira, & A. A. Pinto (Vol. Eds.), Operational research, CIM series in mathematical
Operational Research, 207(2), 644–655. sciences: Vol. 4Springer.
Korošec, P., Bole, U., & Papa, G. (2013). A multi-objective approach to the application of Vieira, M., Pinto-Varela, T., & Barbosa-Póvoa, A. P. (2017a). A multi-objective approach on
real-world production scheduling. Expert Systems with Applications, 40(15), the optimal production and maintenance planning of biopharmaceutical processes under
5839–5853. performance decay using a continuous-time formulation. FOCAPO/CPC2017 conference,
Lakhdar, K., Zhou, Y., Savery, J., Titchener-Hooker, N. J., & Papageorgiou, L. G. (2005). Tucson AZ, USA.
Medium term planning of biopharmaceutical manufacture using mathematical pro- Vieira, M., Pinto-Varela, T., & Barbosa-Póvoa, A. P. (2017b). Production and maintenance
gramming. Biotechnology Progress, 21(5), 1478–1489. planning optimisation in biopharmaceutical processes under performance decay
Leachman, R. C., Johnston, L., Li, S., & Shen, Z.-J. (2014). An automated planning engine using a continuous-time formulation: A multi-objective approach. Computers &
for biopharmaceutical production. European Journal of Operational Research, 238, Chemical Engineering, 107, 111–139.
327–338. Vieira, M., Pinto-Varela, T., Moniz, S., Barbosa-Póvoa, A. P., & Papageorgiou, L. G.
Liu, S., Farid, S., & Papageorgiou, L. (2016). Integrated optimization of upstream and (2016). Optimal planning and campaign scheduling of biopharmaceutical processes
downstream processing in biopharmaceutical manufacturing under uncertainty: A using a continuous-time formulation. Computers & Chemical Enginnering, 91, 422–444.
chance constrained programming approach. Industrial & Engineering Chemistry Wassick, J. M. (2009). Enterprise-wide optimization in an integrated chemical complex.
Research, 55(16), 4599–4612. Computers & Chemical Engineering, 33(12), 1950–1963.
Liu, S., Yahia, A., & Papageorgiou, L. G. (2014). Optimal production and maintenance Wassick, J. M., & Ferrio, J. (2011). Extending the resource task network for industrial
planning of biopharmaceutical manufacturing under performance decay. Industrial & applications. Computers & Chemical Engineering, 35(10), 2124–2140.
Engineering Chemistry Research, 53, 17075–17091.
Maravelias, C. T. (2012). General framework and modeling approach classification for

367

You might also like