Essay 1

A Process Systems Engineering Approach to

Decision-Making in the Chemical Industry
Roderich Wallrath∗

DOI: 10.1002/cite.202400118
This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any
medium, provided the original work is properly cited.

A variety of optimization tools are used in the chemical industry, but they are often embedded in inflexible, isolated,
manual, and complex decision-making processes. This work proposes a decision-making framework that integrates models,
data, and decision-makers into data-connected decision cells, organized within a decision graph. The framework formalizes
decision processes and enables systematic and adaptive workflows across decision layers. The framework can be applied
to various use cases, including supply chain optimization and process design, to address the evolving challenges of the
chemical industry. It enables a shift from component-level optimizations toward system-wide optimization strategies that
account for the interconnected nature of complex production systems.
Keywords: Advanced optimization, Chemical industry, Decision support systems, Model-aided decision-making
Received: September 14, 2024; revised: January 27, 2025; accepted: February 18, 2025

1 Introduction decision-making process in the chemical industry ultimately

relies on communication and collaboration among different
The chemical industry is one of the most important and stakeholders.
complex sectors in the global economy. With its wide range To enable a new level of collaboration and communi-
of products and processes it affects various aspects of mod- cation in decision-making and improve the decision out-
ern life and the environment and contributes to the vision of comes, a data- and model-driven decision-making frame-
a better life and the Sustainable Development Goals [1]. The work is proposed that structures decision-making as a
chemical industry faces many challenges and opportunities, network of decision cells and data flows. The main idea
such as increasing competition, environmental regulations, is to use a decision graph (DG) [8], which is a graphi-
innovation, sustainability requirements, and digitalization. cal representation of the sequence of decisions that need
To cope with these challenges and opportunities, chemical to be made to achieve a certain goal. Nodes in the DG
companies need to make effective and efficient decisions represent decision cells, in which data and models exist
that consider multiple objectives, constraints, uncertainties, to help decision makers reach high-quality decisions. Ver-
and trade-offs. tices in the DG represent decision dependencies and data
One way to support decision-making is to adopt a process flows that link cells to other cells or external data sources.
systems engineering perspective and use computer mod- While the basic idea of decision decomposition is not new
els and process data that capture the relevant aspects of and has been studied extensively, for example, in the con-
the system under consideration, such as physical, chemical, text of mathematical programming, the rapid development
economic, or environmental phenomena. Models can help of decision-support tools and the increasing digitization of
decision-makers to analyze different scenarios, optimize chemical processes are giving new relevance to formalized
performance, design new processes, or evaluate alterna- approaches to decision-making.
tives. Model-based decision-making has successfully been The rest of the article is structured as follows: First, the
adopted in the chemical industry for various applications current role of decision-making frameworks in the chem-
[2], such as process selection and design [3, 4], pro- ical industry is reviewed. Second, the components of the
cess monitoring and control [5], planning and scheduling framework are introduced. Third, the application of the
optimization [6], or supply chain management [7].
However, the models are usually specialized for one
purpose and have a limited scope. Moreover, models are 1
Roderich Wallrath https://orcid.org/0000-0001-6580-3582
often owned by different user groups within or across ([email protected])
organizations, such as engineers, managers, operators, or 1
University of Twente, Faculty of Science and Technology, Sus-
customers. These user groups may have different perspec- tainable Process Technology, Drienerlolaan 5, 7522 NB, Enschede,
tives, objectives, knowledge, or data sources. Therefore, the The Netherlands.

Chem. Ing. Tech. , , No. 0, 1–11 © 2025 The Author(s). Chemie Ingenieur Technik published by Wiley-VCH GmbH
 15222640, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cite.202400118 by Dr.Ashok Vajravelu - National Institutes Of Health Malaysia , Wiley Online Library on [16/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2 Essay

framework to supply chain and production optimization of In [14], an integrated multimethod and multicriteria
specialty chemicals and process design for bulk chemicals decision-making framework for sustainable supplier selec-
production is discussed. Advantages when implementing tion in the chemical industry is laid out. The authors high-
the framework are discussed and a conclusion with an light that current approaches lack the inclusion of risk in
outlook for future work is provided. production and suppliers carbon footprint into the selection
process, as well as the lack of scenario-based decision-
making capabilities. The framework explains the evaluation
2 Related Work and Contribution of the environmental, economic, and social dimension with
appropriate tools. It then computes a weighting of these
The idea of a systematic approach to decision-making in dimensions using the DEMATEL method. However, the
the chemical industry has been developed in the 1950s and framework does not elaborate on the role of data flows and
1960s. In [9], the concept of decision trees is introduced is not focused on the design task of the decision system.
as a simple tool for process investment decision-making. In [15], the batch versus continuous process selection task
It structures decision-making as a sequential process of in the chemical industry is addressed by combining three
decision points and chance events, and allows to sys- different decision-support tools (QFD, AHP, and ANP) into
tematically include the realization of uncertainty as well a single framework which helps to systematically define
as to evaluate choices, risks, objectives, monetary gains, decision criteria, account for feedback and interrelationships
and information needs. In [10], the need for a rigorous among decisions, prioritize, evaluate, and rank decision
approach to production and inventory control is explained outcomes. While the framework allows the inclusion of a
in a complex multistage process application study involving variety of problem aspects and provides quantitative deci-
microbiological fermentation, chemical conversion, drying, sion support, it does not focus on the integration of complex
mechanical grinding, and blending steps. First, the author models and solving them in an iterative procedure within
derives a detailed mathematical model featuring control the decision system.
decision points and suggests a detailed scheduling pro- The current need for aided decision-making was also
cedure to optimize the economical process performance. mentioned in [16], in the context of artificial intelligence
However, process fluctuations due to changes in personnel for green manufacturing in the process industry. In [17],
resources occur which can be represented more efficiently in a graph-based approach to model chemical networks from
a simpler model. The author eventually finds that an alter- feedstocks to products is suggested and the environmental
native scheduling procedure based on simple decision rules impacts of potential processes are assessed. While the need
performs better. for data-driven approaches in early-stage decision-making
Within the same decade, the algorithmic approach [11] in the chemical industry is stated, the lack of accurate and
is developed further and linear programming [12] is intro- complete process databases as a major challenge is pointed
duced as a decision tool for production scheduling. Over out. To systematically approach decision-making, the idea of
the last three decades, the massive use of data and mod- decision graphs to problems was also explored extensively
els in decision-making has become increasingly important for mechanical manufacturing systems in [18]. The authors
as computer technologies have advanced. In today’s world mainly focus on machine selection decisions in complex
of Industry 4.0, Big Data, and artificial intelligence, the manufacturing environments.
role of data and models to improve decision-making is In [19], the idea of graph-based knowledge reuse is
receiving unprecedented attention from industrial and aca- presented to enable data-aided decision-making for new
demic researchers across sectors. Recently, holistic decision- product development in the manufacturing industry. The
making concepts have been developed that facilitate a shift authors suggest to use graphs and a navigation process to
from component-based to system-wide optimization. better access knowledge. This leverages an iterative, data-
In [13], a decision-making framework is developed that driven decision-making process that involves informed,
reorganizes current pharma decision processes involved in manually decision steps and customer feedback.
product discovery, process development, market launch, In [20], a multi-agent approach is described to support
production, and supply chain. As an example of a highly collaborative supply chain decision-making in the chemi-
conservative industry, the authors argue that new techno- cal industry. Since optimal supply chain decisions are often
logical trends and the shift to patience centricity require a impacted by plant performance, compromise decisions need
redesign of current operational strategies from traditional to be made rapidly and evaluated quantitatively. In the
hierarchical decision levels (strategic, tactical, operational) multi-agent system conflicts are resolved by distributed
to decision support layers (business, technological, physi- simulation. While the multi-agent approach leverages the
cal, temporal) that focus on patient needs across functional automatic communication of simulation models and algo-
areas and echelons of the company. While this reconfigu- rithms, it does not include manual decision steps that are
ration addresses the anticipated trends and challenges on a introduced through human collaboration.
conceptual and structural level, it does not elaborate on the A similar approach of using an agent-based intelli-
end-to-end usage of models and data within the framework. gent decision support system is introduced in [21]. The

www.cit-journal.com © 2025 The Author(s). Chemie Ingenieur Technik published by Wiley-VCH GmbH Chem. Ing. Tech. , , No. 0, 1–11
 15222640, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cite.202400118 by Dr.Ashok Vajravelu - National Institutes Of Health Malaysia , Wiley Online Library on [16/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Essay 3

framework allows combining different model types that are

commonly used in the chemical industry such as simulation,
artificial intelligence, and rule-based approaches. Agents are
defined based on different roles and tasks and can be cate-
gorized as user agents, broker agents, and task agents. The
authors show how the framework can be used to support
both process operations and design decisions. While this
flexible framework enables the integration of various tools
and models as well as user inputs, it does not elaborate
on the establishment of data flows in relation to decision
documentation.
The literature on decision-making frameworks in the
chemical industry demonstrates that many innovative
methodologies have been developed to tackle various Figure 1. Schematic decision graph (DG) consisting of nodes
(decision cells) and vertices (data flows).
challenges. However, most of these approaches primarily
focus on introducing new methods, with limited emphasis
on practical, application-based frameworks. Furthermore,
these methods do not suggest how to effectively manage 3.2 Role of Models
and exploit data, which is of increasing importance with the
digitalization of the chemical industry. This work addresses Models (Mij ) are key components of the decision cells,
the gap by proposing a comprehensive, application-oriented serving to support decision-making processes by synthesiz-
framework specifically designed for decision-making sys- ing data and generating recommendations. These models
tems in the chemical industry. It can be applied across may range from simple analytical tools to sophisticated
diverse areas such as supply chain management, process algorithms, depending on the complexity of the decision
design, strategic process selection, and process opera- task.
tions. The framework effectively integrates existing models, The model associated with each decision cell takes inputs
N
tools, data flows, and decision-making processes, offering a from upstream cells (dij−1 ), external data (dijext ), and user-
u
structured approach to optimizing decisions by combining defined parameters (dij ), producing an output decision
models, data, and human input. (dijN ). This integration of models ensures that decisions are
informed by both internal and external factors, enhancing
the robustness and effectiveness of the decision-making pro-
3 Methodology cess. Fig. 2 shows a decision cell (Nij ) featuring a model
(Mij ).
In this section, the idea behind the proposed data- and By formalizing the decision-making process within the
model-driven decision-making framework is discussed. The framework, practitioners can effectively coordinate and
construction of a decision graph (DG), the role of mod- manage decisions, data, models, and stakeholders. This
els and data, and the approach to data- and model-driven structured approach enhances decision quality, fosters col-
decision-making are explained. laboration, and enables informed decision-making across
diverse domains within the organization.

3.1 Decision Graph

The main component of the proposed framework is the

DG, a graphical representation that organizes the decision-
making process into discrete decision cells interconnected
by data flows. The DG, denoted as G = (N,d), comprises
decision cells Nij and data flows dij structured across ver-
tical decision levels j ∈ {1, . . . , m} and horizontal decision
levels i ∈ {1, . . . , n} as illustrated in Fig. 1. In each decision
cell Nij , decisions are made based on inputs from adjacent
cells and external data sources. This hierarchical structure
allows for a systematic decomposition of complex decision
problems into manageable units, facilitating modularization
and analysis. Note that dependencies between decisions can Figure 2. Decision cell Nij contains a model Mij , and receives
be represented by bidirectional data flows that enable the N
input decisions dij−1 from level j–1, external data dijext , and user
modeling of relationships to upstream cells. data diju to compute decision dijN .

Chem. Ing. Tech. , , No. 0, 1–11 © 2025 The Author(s). Chemie Ingenieur Technik published by Wiley-VCH GmbH www.cit-journal.com
 15222640, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cite.202400118 by Dr.Ashok Vajravelu - National Institutes Of Health Malaysia , Wiley Online Library on [16/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4 Essay

laboration, and enables informed decision-making across

diverse domains within the organization.

4 Application Studies

In the following, the implementation of the proposed frame-

work to real-world decision-making systems within the
chemical industry is discussed, specifically focusing on
supply chain optimization and process design. Through
these qualitative application studies, the practical utility of
Figure 3. Illustration of scenario-based decision-making in cell
the framework in addressing the complex decision-making
N21 , featuring three scenarios each with different user input.
challenges of chemical companies is illustrated. First, the
application of the framework to supply chain optimization
3.3 Role of Data for specialty chemicals is presented and secondly, the design
task of a process for bulk chemicals is discussed.
Data serves as the foundation upon which decisions are
made, encompassing a wide range of information relevant 4.1 Application Study 1: Supply Chain
to the decision-making context. These may include numer- Optimization for Specialty Chemicals
ical data, qualitative assessments, and performance metrics,
and uncertainty parameters, sourced from diverse channels 4.1.1 Background
such as databases, sensors, simulations, and market fore-
casts. Uncertainty parameters, such as demand and price The specialty chemicals industry operates within a dynamic
volatility, can be particularly impactful in the DG. environment characterized by fluctuating market demands,
Data flows (dij ) connect decision cells and external or regulatory requirements, supply chain complexities, and
user data sources, facilitating the exchange of informa- long production chains with chemical reactions, unit oper-
tion throughout the DG. By capturing dependencies and ations, large-scale apparatuses, equipment, and logistics.
relationships between decision variables, data flows enable Optimal decision-making requires balancing various factors
decision transparency and facilitate post-decision analysis, such as production capacity and shop floor operations, raw
such as sensitivity analysis and root cause analysis. material availability, distribution logistics, market trends,
and supply chain management. The primary objective is to
3.4 Data- and Model-Driven Decision-Making optimize the end-to-end production of specialty chemicals
from raw material purchasing to customer order deliv-
The main idea behind the proposed framework is a divide- ery and from the global supply chain level to the shop
and-conquer strategy, whereby the decision-making process floor level. This involves streamlining material and infor-
is decomposed into independent decision cells and data mation flows, minimizing inventory costs, and maximizing
flows. This modularization allows for flexible decision- customer satisfaction.
making, with each cell addressed individually as long as it
remains independent of others. 4.1.2 Decision Cells
Furthermore, the framework supports scenario-based
decision-making as illustrated in Fig. 3, wherein different The first step in implementing the framework involves iden-
output decisions (dijNk ) are considered as distinct scenar- tifying decision cells within the supply chain and production
ios (k). First, these scenarios, along with associated data network. Along the horizontal dimension, the decision cells
{dijuk , dijext , dij−1
N
} and models {Mijk }, are stored for docu- correspond to important horizontal stages of the supply
mentation and retro-analysis. Second, as an input to other chain process, including raw material procurement, man-
cells these scenarios can then be used to improve the ufacturing, warehousing, and distribution as illustrated in
decision quality, for example by worst-case to best-case Fig. 4.
assessments or by considering the robustness of decisions. Along the vertical dimension, the decision cells are related
Third, different user groups can work on different scenarios to the supply chain decision-making hierarchy as illustrated
independently. Based on the same input data set {dijext , dij−1
N
} in Fig. 5. At the highest level of the supply chain hierar-
groups can develop their specific scenario k and suggest a chy, global supply chain decisions are made to determine
user data set, and model {dijuk , Mijk }. production forecasts based on market and demand fore-
By formalizing the decision-making process within the casts spanning from months to years into the future. These
framework, practitioners can effectively coordinate and decisions involve strategic planning and resource alloca-
manage decisions, data, models, and stakeholders. This tion to align production capacities with anticipated market
structured approach enhances decision quality, fosters col- demand across different geographical regions. Following

www.cit-journal.com © 2025 The Author(s). Chemie Ingenieur Technik published by Wiley-VCH GmbH Chem. Ing. Tech. , , No. 0, 1–11
 15222640, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cite.202400118 by Dr.Ashok Vajravelu - National Institutes Of Health Malaysia , Wiley Online Library on [16/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Essay 5

Figure 4. Concept of decision system for process design optimization featuring a central decision cell N11 introducing sustainability
requirements to three unit operations decision cells.

global supply chain decisions, regional and local supply 4.1.3 Integration of Models
chain decisions are made to translate high-level produc-
tion forecasts into production targets for sites and plants. Models (Mij ) are developed and integrated into each deci-
These vertical decisions occur on a shorter timescale, typi- sion cell to support decision-making processes. These mod-
cally ranging from weeks to months, and involve optimizing els leverage historical and operational live data, market
production quantities, raw material purchases, sales, distri- trends, and business objectives to generate insights and rec-
bution, and production warehouse planning to meet specific ommendations for decision-making from the supply chain
regional demands and operational constraints. Once pro- to the shop floor level.
duction targets are established at the site and plant level, they At the supply chain decision level, models are often based
are propagated to the production management teams. on mixed-integer programming (MIP) optimization. These
Production management decisions entail generating models enable organizations to formulate and solve large-
detailed production schedules that consider fixed con- scale optimization problems involving strategic objectives
straints on the shop floor, such as available processing related to production planning, inventory management, and
equipment, run rates, personnel resources, and mainte- distribution logistics. MIP models can handle complex deci-
nance periods. Additionally, stochastic factors such as sion problems featuring numerous variables, constraints,
overall equipment efficiency, mean time between failure, and objectives, allowing decision-makers to identify opti-
and mean time of failure are taken into account to ensure mal solutions that maximize overall supply chain efficiency
operational excellence and minimize production disrup- while satisfying objectives and high-level production
tions. After production schedules are finalized, supporting constraints.
functions such as warehousing and logistics are typically At the site, plant, and shop floor level, models are
optimized to ensure efficient material flow and timely used to simulate detailed production processes and oper-
delivery of finished products to customers. This involves ational workflows. This involves detailed physicochemical
coordinating storage and transportation capacities as well systems modeling for chemical reactions, thermodynamic
as inventory management policies to minimize costs and first-principle models, and discrete event modeling (DES)
maximize service level. for material flow and logistics systems. Models capture the
dynamic behavior of production systems, allowing decision-
makers to assess the impact of various factors such as
equipment downtime, production bottlenecks, and resource
allocation strategies on the overall performance. Further-
more, users can incorporate rules and heuristics into models
that mimic business and operational decision logic, enabling
organizations to explore what-if scenarios, and evaluate the
robustness of proposed solutions.

4.1.4 Data Flow Establishment

Data flows (dij ) are established to facilitate the exchange

of information between decision cells and external data
sources. The framework acknowledges the existing decision-
making flow within the organization and utilizes it as a
Figure 5. Illustration of three important vertical decision levels foundation for establishing data flows. This ensures conti-
in the supply chain in specialty chemicals manufacturing. nuity and minimizes disruption to established workflows,

Chem. Ing. Tech. , , No. 0, 1–11 © 2025 The Author(s). Chemie Ingenieur Technik published by Wiley-VCH GmbH www.cit-journal.com
 15222640, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cite.202400118 by Dr.Ashok Vajravelu - National Institutes Of Health Malaysia , Wiley Online Library on [16/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
6 Essay

Table 1. Overview of the framework’s components for application study 1.

Decision level Decision cells Model types Data flows Goal

Global Procurement, sales, Rigorous optimization, e.g., linear ERP/MRP, demand forecasts, Optimize net present value,
financial controlling, programming, business decision pricing strategies, global define all material quantities
business strategy, global logic procurement and sales
production planning planning
Site, plant, Local planning, site- and Optimization, e.g., mixed-integer APS, MES, production run Plan production, allocate
logistics plant management, site programming, dynamic simulation, rates, asset availability plans primary and secondary
logistics DES and meta-heuristics, business resources
decision logic
Shop floor Process management, shift Dispatching rules, execution logic SCADA, DCS, PLC, shifts Detailed scheduling, robust
management plans execution

while fostering collaboration and knowledge sharing across data sources like IoT devices, SCADA systems, and exter-
different decision-making levels and functional areas. nal market data feeds. Additionally, middleware platforms
In supply chain operations, data flows are primarily and data integration tools can be employed to standardize
focused on transmitting information related to demand and route data efficiently across different systems, ensuring
and sales forecasting, inventory management, procurement, that information is transmitted securely and with minimal
and logistics. Data flows from sales teams to produc- latency.
tion planning teams provide insights into market demand,
which then informs production scheduling and inventory 4.1.5 Outcomes
replenishment decisions. Similarly, data flows from sup-
pliers to procurement departments facilitate the ordering The implementation of the framework to application study
of raw materials and supplies. However, these data flows 1 is summarized in Tab. 1. Based on that, four enhance-
are predominantly unidirectional, flowing from upstream ments to the decision-making process are discussed in the
to downstream processes, with limited feedback loops to following.
optimize decision-making across the supply chain.
Similarly, in site and plant management, data flows are Closing the feedback loop: By establishing bidirectional data
primarily directed towards operational monitoring, pro- flows between the production decision levels and the sup-
cess control, and performance optimization. Data flows ply chain decision level, the framework facilitates more
from sensors, SCADA systems, and control devices pro- informed decision-making and ensures alignment between
vide real-time information on process variables, equipment strategic objectives and operational realities. Since specialty
status, and production performance. This data is then chemicals plants often are producing a variety of products
utilized by operators and engineers to make operational and product grades, throughput is constrained by equip-
decisions, adjust process parameters, and ensure efficient ment changeover times, for instance due to re-tooling and
plant operation. However, these data flows typically follow cleaning. This complexity can only be considered at the site-,
a unidirectional path, with limited integration and feedback plant-, and shop floor level, requiring detailed equipment
mechanisms between different levels of plant management. knowledge and models. The framework also advocates for
At the shop floor level, data flows are focused on produc- an iterative approach to decision-making.
tion execution, quality control, and maintenance activities. Feedback data flows are established to align supply chain
Data flows from production systems, quality control devices, targets with the range of feasible production scenarios,
and maintenance logs provide insights into production rates, accounting for constraints and nuances that may impact
product quality, and equipment reliability. This informa- decision outcomes. This feedback can take place either
tion is used by shop floor supervisors and maintenance by manually providing the data sets to the owners of the
teams to optimize production schedules, ensure product adjacent decision cells or by automatically executing the
quality standards, and minimize downtime. However, simi- cells with the new data. A major advantage of formal-
lar to supply chain and plant management, these data flows izing feedback data flows is that the decision process is
often lack bidirectional communication and integration recorded. Particularly for interdependent decisions between
with higher-level decision-making processes. supply chain and production, this is often neglected because
Technically, the data flows are realized through the inte- a formal way of describing iterative decision-making is
gration of application programming interfaces (APIs). The cumbersome or completely lacking.
APIs facilitate real-time data exchange by connecting enter-
prise systems such as enterprise resource planning (ERP) Scenario-based decision-making: For specialty chemicals, the
and supply chain management (SCM) with other critical uncertainty of data sources on the supply chain level must be

www.cit-journal.com © 2025 The Author(s). Chemie Ingenieur Technik published by Wiley-VCH GmbH Chem. Ing. Tech. , , No. 0, 1–11
 15222640, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cite.202400118 by Dr.Ashok Vajravelu - National Institutes Of Health Malaysia , Wiley Online Library on [16/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Essay 7

Table 2. Concept of unique identification (ID) of decisions based historian, better production forecasts can be achieved.
data flows and model. As a result, process equipment, warehouses, and logistics
Cell services can be operated more efficiently.
1 2 3 3. Another implication can be to invest in redundancy mea-
Scenario sures to safeguard against unforeseen disruptions. In the
k
N11 1
d11 2
d11 3
d11 face of unforeseen events, such as large-scale supply chain
k
N21 1
d21 or production disruptions, the decision historian together
k 1 2 with models can be an extremely powerful tool for guid-
N31 d31 d31
ing organizations on how to react best to avoid lost
N12 {d11
2 , d 1 , d 1 , d u , d ext , M }
21 31 12 12 12 ID → d12 sales and minimize the impact on operations. Thereafter,
the decision historian can indicate where to close the
loop between decision cells to improve decision-making
considered. Demand projections often depend on the per- sustainably.
formance of other industries. Likewise, raw material lead
times, for instance from oversea suppliers, can vary strongly. Human interaction: Model recommendations generated
Commercial forecasts and pricing strategies are frequently through quantitative analysis can serve as valuable inputs
adjusted to global market conditions and can be disrupted to decision-making, providing insights and perspectives
by global events such as supply chain issues, natural disas- that may not be immediately apparent through manual
ters, geopolitical events, or technological disruptions. Given intuition alone. By incorporating Bayes’ Theorem into the
this inherent uncertainty, massive scenario-based testing decision-making process, the framework allows decision-
and evaluation can be conducted to assess the robustness of makers to integrate their beliefs with model results in a
supply chain decisions. This necessitates the exchange of a systematic manner. Bayes’ Theorem allows decision-makers
vast amount of data between decision cells, and underscores to update their prior beliefs based on new solution infor-
the importance of formalizing decision-making processes mation from models and based on the model accuracy, as
and applying concepts of version control. Herein, the frame- shown in Eq. (1). This provides decision-makers with quan-
work enables retro-analysis and decision monitoring as titative decision support that can help to validate beliefs and
mentioned in Sect. 3.4. As illustrated in Tab. 2, digital fin- reinforce the confidence in decisions.
gerprints can be attached to each decision by assigning  
unique IDs to decisions and the corresponding data sets and P M A |A ∗ P (A)
P (A|MA) = (1)
models. P (MA)

Decision historian: Based on these considerations, a database For example, supply chain decision-makers may have a
for decision-making (decision historian) can be created that prior belief, P(A), that selecting specialty chemical prod-
serves as a basis for decision performance monitoring, to uct A for the upcoming manufacturing period will yield the
detect root causes of suboptimal outcomes and to analyze highest revenues. This belief can be updated when the model
uncertainty propagation from data sources to the final deci- M also recommends product A, resulting in the posterior
sions. The idea of data historians in the chemical industry belief (conditional probability) P(A|MA ). The probability
is not new, and in addition to previous work [22, 23] this that model M recommends product A is typically bro-
paper proposes to utilize data historians to directly track and ken down as P(M A ) = P(A) ∗ P(M A |A) + P(Ā) ∗ P(M A |Ā),
improve decision-making. By analyzing historical supply accounting for the possibility that the model might recom-
chain decision data and production performance metrics, mend product A, even if it is not the revenue-maximizing
companies can identify recurring patterns, inefficiencies, option. Additionally, the model’s sensitivity, P(MA |A), must
and missed opportunities in the overall decision-making be considered, i.e., the probability that product A is rec-
processes. The decision historian also provides organiza- ommended given that it is indeed the revenue-maximizing
tions with insights into where to invest to reduce uncertainty choice. With this, the posterior belief P(A|MA ) can be
and where uncertainty is not business-critical. In specialty formed, quantifying the confidence in selecting product A
chemicals, insights from the decision historian may improve based on the model’s recommendation.
the following areas:
1. In fluctuating markets, such as agrochemicals, take-or- 4.1.6 Conclusion
pay contracts are often used to balance the risks between
buyers and sellers. The decision historian could be used The adoption of a data- and model-driven decision-making
to make material purchases more accurate, and purchas- framework enables specialty chemicals companies to opti-
ing costs can be reduced through the renegotiation of mize supply chain and production processes and address
supplier contracts. challenges such as the risk of lost sales, cost of produc-
2. The reduction of the cost of flexibility can also be tion flexibility, and supply chain disruptive events more
achieved on the production side. Based on the decision effectively. By leveraging organizational expertise across

Chem. Ing. Tech. , , No. 0, 1–11 © 2025 The Author(s). Chemie Ingenieur Technik published by Wiley-VCH GmbH www.cit-journal.com
 15222640, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cite.202400118 by Dr.Ashok Vajravelu - National Institutes Of Health Malaysia , Wiley Online Library on [16/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8 Essay

functional areas and by integrating advanced technologies, cells play a pivotal role in process design, representing a
such as DES (SimPy, JaamSim), MIP (SCIP, HiGHs), and holistic engineering view responsible for optimizing overall
deep learning (Keras, TensorFlow), into decision cells, orga- process efficiency, capital expenditures (CAPEX), opera-
nizations can enhance their decision-making capabilities tional expenditures (OPEX), overall mass balancing and
and transparency. By combining pre-existing, stand-alone stoichiometric considerations, reducing waste generation,
models and tools within chemical companies, the frame- energy consumption, CO2 emissions, and ensuring overall
work can also exploit unused synergy effects. operational flexibility.

4.2 Application Study 2: Sustainable Process 4.2.3 Integration of Models

Design for Bulk Chemicals
Models (Mij ) improve the decision-making ability in each
4.2.1 Background cell, and the model types can be selected according to the
complexity and characteristics of the unit operation.
While the bulk chemicals sector relies on established pro- First-principle models are standard tools in process
cesses and products, the growing imperative of sustainability design. They include chemical reaction models, thermody-
brings with it the need for innovation and sustainable pro- namic models, computational fluid dynamics (CFD) models
cess design. Modern processes must be designed not only and others, and are used to accurately represent process
to meet the stringent sustainability standards but also to dynamics and behavior to make informed design decisions.
accommodate operational flexibility in products, raw mate- Sustainable processes mostly operate in a transient regime
rials, and energy supply. This dual challenge necessitates the due to variability in feedstocks and energy supply. With first-
design or redesign of multiproduct plants capable of pro- principle models, the effect of dynamic heating and cooling
cessing a set of raw materials, including renewables with on the temperature profiles and the conversion rates can be
variable quality, while maintaining the ability to utilize con- explored. With the models, the operational flexibility can be
ventional feedstocks. Moreover, with the rising adoption of monitored throughout the design process and the impact of
renewable energy sources, processes must be engineered changes in design parameters on the operating window can
to efficiently utilize fluctuating energy inputs. The pro- be assessed. In addition to first-principle models, decision
cess design task becomes complex, as processes must be cells may incorporate empirical models, grey-box models,
robust within a larger operating window and have efficient and fully data-based models. For sustainable bulk chem-
ramp-up and ramp-down capabilities. The intersection of icals production they can be an advantage when dealing
sustainability goals with operational flexibility goals requires with biomass feedstocks since bioreaction kinetics are often
a comprehensive decision-making framework that helps too complicated to describe with first-principles or are not
to design for both optimal environmental and economic known.
performance. Low-fidelity models play a crucial role in connecting
first-principle models of unit operations and in supporting
4.2.2 Decision Cells model-aided process design. These low-fidelity models
serve as bridges between detailed, high-resolution mod-
Bulk chemicals are usually produced in large-scale pro- els and enable the integration and optimization of the
cess networks that contain unit operations such as storage entire process or process sections. Low-fidelity models
and pretreatment, reaction, separation, and purification. In are typically based on simple material and energy flow
addition to focusing on specific process steps and unit balances, representing the overall process or aggregated
operations, the decision cells can also be modified. For process sections on a process flow diagram level. By simu-
example, when designing a polyurethane foam recycling lating different scenarios and design options at the system
process, the decision cells can cover the collection logis- level, decision-makers can quickly find optimized process
tics and upstream and downstream sections, emphasizing configurations that balance conflicting objectives such
that the most important design decisions are made at the as cost, performance, and sustainability. Thereafter, the
intersections. On the other hand, the decision cells can be individual process components can be optimized using the
further subdivided into apparatuses. For example, in process unit-operation models. For example in [24], an MINLP
design in a biorefinery, the key energy consumption deci- superstructure model is used for the simultaneous fuel
sions could be made at the apparatus level, or apparatuses and process design of biorefineries. This model effectively
could be designed by different vendors within the project captures the interdependencies between fuel specifications
or functional areas within the organization. If model-based and the associated production processes on the system
engineering is critical to the decision-making process, the level. With the optimal design determined, specific process
cells may also be modified depending on the availability equipment can then be detailed in engineering projects and
of technical expertise and modeling tools. Central decision analyses.

www.cit-journal.com © 2025 The Author(s). Chemie Ingenieur Technik published by Wiley-VCH GmbH Chem. Ing. Tech. , , No. 0, 1–11
 15222640, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cite.202400118 by Dr.Ashok Vajravelu - National Institutes Of Health Malaysia , Wiley Online Library on [16/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Essay 9

Table 3. Overview of the framework’s components for application study 2.

Decision level Decision cells Model types Data flows Goal

Holistic Mass and energy balance, Balance equations, (mixed-integer Sustainability and operational Optimize sustainability and
sustainability performance, non-)linear programming, flexibility KPIs, environmental economic performance,
OAE, process flexibility, dynamic flowsheet simulation impact and production cost determine best design with
CAPEX, OPEX sensitivity operational flexibility
Unit operations Atom economy and First-principle models, Sustainability KPIs, robustness Optimize unit operation
conversion rates, ODE-based dynamic simulation, and safety KPIs, chemical and design under soft
sustainability performance, kinetics and thermodynamics thermodynamic properties of constraints and objectives
OEE, operating window models, CFD simulation process flows, variability of raw
materials and energy supply

4.2.4 Data Flow Establishment ments to the decision-making process are discussed in the
following.
Bidirectional data flows are essential in process design, facil-
itating an iterative approach that spans from conceptual Documenting design choices: For bulk chemicals process
design to basic engineering to detailed engineering. Data design, the framework enables rigorous documentation
flows enable the exchange of information between the deci- of design choices, enhancing the traceability of decisions
sion cells and support the refinement and optimization of throughout the process design lifecycle. By systematically
design decisions in the course of the project phases. capturing decision data and associated models and sce-
Data flows must also contain sustainability information, narios, the evolution of design decisions can be tracked,
for example using key performance indicators (KPIs), to and decision-makers can identify rationale behind design
ensure that sustainability requirements are integrated into choices at any time in the project and assess the impact of
the evaluation at every stage of design. For bulk chemicals design modifications on project outcomes. This retrospec-
production, important KPIs are energy consumed per unit tive analysis aids in improving engineering efficiency and
of product output, carbon dioxide (CO2 ), methane (CH4 ), mitigating design risks.
and nitrous oxide (N2 O) emissions, water usage, waste
generation, raw material efficiency, and renewable energy Model flexibility: One of the key advantages of integrating
integration. By including these sustainability KPIs in the models within the framework is the modeling flexibility it
data flows, decision-makers can continuously monitor and enables. Different model types can be deployed in different
evaluate the environmental impact of design choices across decision cells, each tailored to address specific character-
various unit operations, enabling mitigation of sustainability istics of the decision problem. For example, a bioreactor
risks and alignment with sustainability objectives. decision cell can feature a machine-learning model to
Data flows should also include sensitivity information on describe the reaction kinetics, while the following separation
design parameters to enable robustness and sensitivity anal- step in a distillation column can be described by a dynamic
ysis. Decision-makers can assess the impact of variations in first-principle model. Moreover, models can be easily val-
design parameters on key performance metrics across differ- idated, exchanged, or updated in individual cells without
ent unit operations. This facilitates identification of critical affecting the functionality of other cells, allowing for agile
variables early in the design process. decision-making and continuous improvement throughout
Given the dual challenge of sustainable bulk chemicals the design project.
production, it is particularly important to consider both
environmental impact and production costs in data flows. Integration of sustainability requirements: As illustrated in
These aspects are often strongly related to conversion rates Fig. 6, central engineering decision cells such as (N11 ) can
or energy consumption across unit operations. Data flows not only integrate insights from individual unit operations
should therefore allow for communication of environmen- cells, but also introduce sustainability goals and external
tal impact and production costs as well as their sensitivity ext
requirements (d11 ) into the design process. The com-
to other design parameters. This enables holistic assess- munication between individual cells is key to integrate
ment and optimization of design choices that balance both overall process requirements into the process design without
environmental sustainability and economic viability. prescribing constraints to each individual unit operation.
For example, in bulk chemical production key sustainabil-
ity goals are minimizing emission factors and optimizing
4.2.5 Outcomes atom economy on alternative synthesis routes using new
ext
feedstocks. These goals (d11 ) can be addressed on decision
The implementation of the framework to application study level 1. However, these goals can only be realized within the
2 is summarized in Tab. 3. Based on that, three enhance- unit operations (N12 ,N22 ,N32 ) on decision level 2. Optimized

Chem. Ing. Tech. , , No. 0, 1–11 © 2025 The Author(s). Chemie Ingenieur Technik published by Wiley-VCH GmbH www.cit-journal.com
 15222640, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cite.202400118 by Dr.Ashok Vajravelu - National Institutes Of Health Malaysia , Wiley Online Library on [16/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
10 Essay

Figure 6. Concept of decision sys-

tem for process design optimiza-
tion featuring a central decision
cell N11 introducing sustainabil-
ity requirements to three unit
operations decision cells N12 , N22 ,
N32 .

u
design choices (d12 , d22
u
, d32
u
) therefore must be derived in an that the data- and model-driven framework can be an
iterative process between unit operations decision cells and effective solution approach for chemical companies.
overall process design decision cells. Future work should test this framework in a real-world
implementation case study. This task has eventually to be
4.2.6 Conclusion tackled in interdisciplinary, industrial collaborations, due to
the sensitivity of the data that would be required. Further-
By defining the decision structure and decision interfaces, more, it would be interesting to investigate how automatic
the framework enables independent optimization of unit model validation can be performed within the framework,
operations in individual decision cells and thus promotes and to explore further how human intuitions and model
the adaptability of process design. Through bidirectional recommendations can be merged across decision cells.
data flows, overall process requirements are effectively prop-
agated to the cells, ensuring alignment with sustainability Symbols used
objectives and operational constraints. The framework’s
flexibility in accommodating various model types allows dij [–] data flow between decision cells
decision-makers to leverage diverse modeling approaches, Mij [–] model in decision cell i,j
enhancing the accuracy and comprehensiveness of process Nij [–] decision cell in horizontal decision level i
analyses. Moreover, the framework facilitates rigorous docu- and vertical decision level j
mentation and retro-analysis throughout the design project, P(A) [–] priori belief (probability) of hypothesis A
providing transparency and accountability in decision- P(A|MA ) [–] posteriori belief (conditional probability)
making processes. of hypothesis A, given the result of
model M on hypothesis A

5 Conclusion
Sub- and superscripts
In this study, a framework for data- and model-driven
decision-making in the chemical industry is introduced. It ext [–] external
is illustrated how complex decision systems can be struc- i [–] horizontal decision levels {1, . . . , n}
tured as decision graphs that consist of decision cells and j [–] vertical decision levels {1, . . . , m}
data flows. This systems engineering approach is discussed k [–] scenario
in two application studies of process design for bulk chem- u [–] user
icals and supply chain operations for specialty chemicals,
and potential benefits were highlighted. Given the need for Abbreviations
end-to-end optimized decision-making and an increasingly
digitalized and complex chemical industry, it is concluded API application programming interfaces

www.cit-journal.com © 2025 The Author(s). Chemie Ingenieur Technik published by Wiley-VCH GmbH Chem. Ing. Tech. , , No. 0, 1–11
 15222640, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cite.202400118 by Dr.Ashok Vajravelu - National Institutes Of Health Malaysia , Wiley Online Library on [16/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Essay 11

CAPEX capital expenditures [11] B. Giffler, G. L. Thompson, Algorithms for solving production-
DES discrete event modeling scheduling problems, Oper. Res. ,  (), 487–503.
DG decision graph [12] A. S. Lee, J. S. Aronofsky, A linear programming model for
scheduling crude oil production, J. Pet. Technol. ,  (),
ERP enterprise resource planning
51–54.
KPI key performance indicator [13] C. M. Marques, S. Moniz, J. P. de Sousa, A. P. Barbosa-Povoa,
MIP mixed-integer programming G. Reklaitis, Decision-support challenges in the chemical-
OPEX operational expenditures pharmaceutical industry: Findings and future research direc-
SCM supply chain management tions, Comput. Chem. Eng. , , 106672.
[14] C. Wu, Y. Lin, D. Barnes, An integrated decision-making
approach for sustainable supplier selection in the chemical
References industry, Expert Syst. Appl. , , 115553.
[15] F. Y. Partovi, An analytical model of process choice in the
[1] J.-F. Chen, Green chemical engineering, Engineering ,  (), chemical industry, Int. J. Prod. Econ. ,  (), 213–227.
283–284. [16] S. Mao, B. Wang, Y. Tang, F. Qian, Opportunities and chal-
[2] N. Asprion, M. Bortz, Process modeling, simulation and opti- lenges of artificial intelligence for green manufacturing in the
mization: from single solutions to a multitude of solutions process industry, Engineering ,  (), 995–1002.
to support decision making, Chem. Ing. Tech. ,  (), [17] J. Staddon, J. Smit, Z. Skoufa, D. Watson, Chemical Networks:
1727–1738. A Methodology to Rapidly Assess the Environmental Impact
[3] J. Burger, N. Asprion, S. Blagov, R. Böttcher, U. Nowak, of Chemical Processes: Applying graph theory principles to
M. Bortz, R. Welke, K. H. Küfer, H. Hasse, Multi-Objective chemical industry data enables early-stage decision making for
Optimization and Decision Support in Process Engineering – optimum decarbonisation solutions, Johnson Matthey Technol.
Implementation and Application, Chem. Ing. Tech. ,  (), Rev. ,  (), 466–478.
1065–1072. [18] R. V. Rao, Decision Making in the Manufacturing Environment:
[4] C. Bramsiepe, N. Krasberg, C. Fleischer, L. Hohmann, N. Using Graph Theory and Fuzzy Multiple Attribute Decision
Kockmann, G. Schembecker, Information technologies for Making Methods, Springer, London , Vol. , 294.
innovative process and plant design, Chem. Ing. Tech. ,  [19] C. Zhang, G. Zhou, Q. Lu, F. Chang, Graph-based knowl-
(), 966–981. edge reuse for supporting knowledge-driven decision-making
[5] Z. Ge, Z. Song, F. Gao, Review of recent research on data-based in new product development, Int. J. Prod. Res. ,  (),
process monitoring, Ind. Eng. Chem. Res. ,  (), 3543– 7187–7203.
3562. [20] R. García-Flores, X. Z. Wang, A multi-agent system for chem-
[6] I. Harjunkoski, C. T. Maravelias, P. Bongers, P. M. Castro, ical supply chain simulation and management support, OR
S. Engell, I. E. Grossmann, J. Hooker, C. Mendez, G. Sand, Spectrum , , 343–370.
J. Wassick, Scope for industrial applications of production [21] Y. Gao, Z. Shang, A. Kokossis, Agent-based intelligent system
scheduling models and solution methods, Comput. Chem. Eng. development for decision support in chemical process industry,
, , 161–193. Expert Syst. Appl. ,  (), 11099–11107.
[7] H. Paulo, M. Vieira, B. S. Gonçalves, S. Relvas, T. Pinto-Varela, [22] I. A. Udugama, C. L. Gargalo, Y. Yamashita, M. A. Taube, A.
A. P. Barbosa-Povoa, A Discrete-Event Simulation Approach to Palazoglu, B. R. Young, C. Bayer, The role of big data in indus-
the Design and Planning of Biomass Supply Chains Consider- trial (bio) chemical process operations, Ind. Eng. Chem. Res.
ing Technological Learning, Comput. Aided Chem. Eng. , ,  (), 15283–15297.
, 1285–1291. [23] J. Crompton, Data Management from the DCS to the Histo-
[8] BayesianNetworks and DecisionGraphs (Eds.: F. V. Jensen, T. D. rian, in Machine Learning and Data Science in the Oil and Gas
Nielsen), Vol. , Springer, New York . Industry (Ed.: P. Bangert), Gulf Professional Publishing, Oxford
[9] J. F. Magee, Decision Trees for Decision-Making, Harvard Busi- , 83–110.
ness Review, Brighton, MA , 35–48. [24] J. M. Restrepo-Flórez, C. T. Maravelias, Advanced fuels from
[10] R. L. Ackoff, Production and inventory control in a chemical ethanol – a superstructure optimization approach, Energy
process, J. Oper. Res. Soc. Am. ,  (), 319–338. Environ. Sci. , , 493–506.

Chem. Ing. Tech. , , No. 0, 1–11 © 2025 The Author(s). Chemie Ingenieur Technik published by Wiley-VCH GmbH www.cit-journal.com