A Distributed Event-Orchestrated Digital Twin Architecture for

Optimizing Energy-Intensive Industries

Nicol

o Bertozzi

1 a

, Anna Geraci

1 b

, Letizia Bergamasco

1,2 c

, Enrico Ferrera

1 d

Edoardo Pristeri

1 e

and Claudio Pastrone

1 f

Fondazione Links, Via Pier Carlo Boggio 61, 10138 Turin, Italy

Department of Control and Computer Engineering, Politecnico di Torino, Corso Duca degli Abruzzi 24, 10129 Turin, Italy

{name.surname}@linksfoundation.com

Keywords:

Distributed Microservices Architecture, Digital Twin, Event-Based Orchestration, Interoperability, Energy

Optimization, Industry, Computing Continuum.

Abstract:

This paper presents a novel distributed architecture designed to spawn digital twin solutions to improve energy

efﬁciency in energy-intensive industrial scenarios. By executing user-deﬁned workﬂows, our platform enables

the implementation of real-time monitoring, forecasting, and simulation microservices to enhance decision-

making strategies for optimizing industrial processes. Leveraging a stateless centralized orchestration mech-

anism built around an Apache Kafka-based backbone, the platform ensures scalability, fault tolerance, and

efﬁcient handling of heterogeneous data. Key features include intuitive workﬂow conﬁguration, asynchronous

communication for streamlined workﬂow execution, and API-driven scheduling for dynamic, event-based task

management. This platform will be deployed and validated in several energy-intensive industrial scenarios,

supporting the management of energy systems of different plants, to prove its effectiveness across a wide range

of energy management challenges.

1 INTRODUCTION

The rising global demand for smarter, more sustain-

able industrial practices has propelled the adoption

of Digital Twin (DT) technologies. These systems,

which create virtual replicas of physical processes

or assets, hold immense potential for optimizing op-

erations, reducing resource wastage, and addressing

the pressing challenges of climate change. Within

energy-intensive industries—responsible for nearly

27% of total energy consumption in EU countries

(Odyssee-Mure Project, 2023)—DTs offer transfor-

mative solutions by enabling real-time energy mon-

itoring, predictive analytics, and dynamic process

optimization (Zhang et al., 2021; Henriksen et al.,

2022). Implementing DT technologies is increas-

ingly necessary to enable the seamless execution of

Artiﬁcial Intelligence-driven analytics and simulation

https://orcid.org/0000-0003-2049-9876

https://orcid.org/0009-0006-2251-8347

https://orcid.org/0000-0001-6462-1699

https://orcid.org/0000-0002-4671-3861

https://orcid.org/0000-0003-4538-3044

https://orcid.org/0000-0003-0471-8434

modules, which are critical to optimizing industrial

processes and decision-making. Achieving this re-

quires the ability to integrate heterogeneous systems,

execute complex workﬂows efﬁciently, and maintain

scalability under dynamic industrial demands. More-

over, as industrial ecosystems become more complex,

DT platforms must effectively orchestrate distributed

components, ensuring efﬁcient coordination, consis-

tency, and fault tolerance.

Several challenges must be addressed for DT ar-

chitectures to reach their potential. These include

data integration and interoperability, ensuring seam-

less ﬂow across diverse systems; human-computer in-

teraction and usability, providing intuitive interfaces

for non-expert users; scalability, managing growing

workﬂows and events without performance degrada-

tion; and orchestration, enabling smart communica-

tion and coordination among modular components.

Overcoming these hurdles requires innovative archi-

tectures tailored to the complexity and dynamic de-

mands of modern industrial environments.

In response to these challenges, this paper intro-

duces a distributed event-orchestrated architecture for

digital twins, designed to streamline deployment and

enhance operational efﬁciency. This novel architec-

Bertozzi, N., Geraci, A., Bergamasco, L., Ferrera, E., Pristeri, E. and Pastrone, C.

A Distributed Event-Orchestrated Digital Twin Architecture for Optimizing Energy-Intensive Industries.

DOI: 10.5220/0013364400003944

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 10th International Conference on Internet of Things, Big Data and Security (IoTBDS 2025), pages 337-344

ISBN: 978-989-758-750-4; ISSN: 2184-4976

337

ture represents a transformative advancement in in-

dustrial digital twin solutions by seamlessly integrat-

ing Internet of Things (IoT) sensors, Artiﬁcial In-

telligence (AI) algorithms, and simulation tools. It

is designed to enable effortless customization for di-

verse use cases while addressing the key limitations

of existing platforms. The architecture’s key fea-

tures include: i) an intuitive workﬂow conﬁguration

that simpliﬁes the management of complex interac-

tions between modules; ii) a stateless microservices

design and asynchronous workﬂow execution, ensur-

ing efﬁciency in distributed environments; and iii)

API-driven scheduling capabilities, allowing work-

ﬂows to be triggered dynamically by time-based or

data-driven events. These features collectively em-

power organizations to leverage DT functionality with

greater adaptability, efﬁciency, and ease of use. This

platform has been developed within the FLEXIndus-

tries project, and will be deployed in several indus-

trial energy management settings, for use cases such

as energy demand forecasting and energy consump-

tion optimization.

This paper is organized as follows: Section 2 re-

views related digital twin approaches; Section 3 de-

scribes the presented digital twin architecture; Sec-

tion 4 details its orchestration mechanisms; Section 5

presents the use cases in which the presented platform

is deployed; and Section 6 highlights the conclusion

and future work.

2 RELATED WORK

Several studies have demonstrated the effectiveness

of DT architectures in sectors such as manufactur-

ing (Javaid et al., 2023; Alberti et al., 2024), en-

ergy (Yu et al., 2022), and smart cities (Farsi et al.,

2020), particularly in optimizing resource use and en-

hancing operational efﬁciency. One notable example

of a digital twin framework is the REPLICA system

(Rossini et al., 2020), developed as part of the RE-

CLAIM European project. REPLICA offers a scal-

able and ﬂexible architecture for realizing intelligent

digital twins, leveraging an event-based communi-

cation model and technologies like Apache Kafka.

However, the framework’s distributed orchestration

model, where each module has to manage its own

communication logic, introduces limitations. Specif-

ically, the absence of centralized orchestration can

lead to complex, interdependent workﬂows, where

mismanagement of message publishing and subscrib-

ing phases results in communication inefﬁciencies.

Furthermore, REPLICA requires manual conﬁgura-

tion and hard-coding of new workﬂows, making the

process cumbersome and time-consuming.

In contrast to REPLICA’s distributed orchestra-

tion model, in this paper we introduce a central-

ized approach, which seeks to prevent inefﬁciencies

by consolidating control in a central orchestration

component. This simpliﬁcation aims to streamline

workﬂow management, reduce conﬁguration over-

head, and improve system scalability. While cen-

tralized orchestration offers signiﬁcant advantages in

terms of control and coordination, it can introduce

risks such as bottlenecks, particularly in large-scale

distributed environments. To mitigate such risks, we

implement a stateless architecture, which contrasts

with REPLICA’s stateful approach that stores all the

intermediate variables and results locally. As high-

lighted in (Rossini et al., 2020), the stateful nature of

REPLICA can lead to delays in module availability

due to the persistence of data. Instead, our stateless

design eliminates reliance on local data storage, en-

abling more ﬂexible and efﬁcient handling of concur-

rent requests.

From the literature on orchestration, a clear need

emerges for centralized and asynchronous orchestra-

tion mechanisms. Broadly, orchestration can be cat-

egorized into two main types: service orchestration

and data orchestration. As discussed in (

Cili

c et al.,

2023; Nguyen et al., 2020), service orchestration is

commonly employed to manage workload demands.

In these studies, tools like Kubernetes (Kubernetes,

2020) are highlighted as central solutions in this do-

main. In cloud-native environments, Kubernetes’

Horizontal Pod Autoscaling (HPA) plays a crucial

role by dynamically adjusting the number of applica-

tion pods based on real-time trafﬁc patterns, ensuring

that the infrastructure scales up or down as needed

to maintain optimal performance. On the other hand,

data orchestration addresses the challenges of manag-

ing large-scale data ﬂows. For instance, solutions like

Apache Kafka (Sax, 2018) have been explored for en-

abling real-time processing of massive data streams,

ensuring smooth data ﬂow and low-latency handling

(Escrib

a-Gelonch et al., 2024; Li et al., 2023). These

works mention that Kafka offers several features mak-

ing it a good ﬁt for companies’ requirements, includ-

ing scalability, data grouping and partitioning, low la-

tency, and the ability to handle a large number of di-

verse consumers.

Asynchronism and data orchestration are straight-

forward individually, but integrating them is challeng-

ing. A decision framework (Megargel et al., 2021)

helps architects choose choreography, orchestration,

or a hybrid approach based on requirements like cou-

pling and visibility. For energy-intensive industries, a

hybrid approach—combining asynchronous commu-

IoTBDS 2025 - 10th International Conference on Internet of Things, Big Data and Security

338

nication with centralized orchestration—best handles

high data throughput and workﬂow control.

The DT extends beyond mere data orchestration,

computation, communication, and control. In addi-

tion to these properties, Digital Twins must address

security domains to ensure authorized access and re-

strict operation. In (Harper et al., 2019), the authors

emphasize secure access as a core point of Digital

Twin architectures. Clients interacting with Digital

Twins should authenticate using established security

best practices, which could be facilitated through fed-

erated identity systems. Furthermore, recent discus-

sions in (Onwubiko et al., 2023; Hemdan et al., 2023;

Salim et al., 2022) explore a blockchain-based solu-

tion for DT, which represents a promising technology

to enhance DT security in the future.

In (Jwo et al., 2022), the term “Digital Twin” is re-

named as “Data Twin”. Indeed, as we see in (He et al.,

2019; Emmert-Streib and Yli-Harja, 2022; Ma et al.,

2023), the quality of data is crucial in digital twins.

Effective data management ensures that the DT accu-

rately reﬂects the real world and supports informed

decision-making. The data sources can vary signiﬁ-

cantly based on the type of data and sensor installa-

tions. As discussed in (Sun et al., 2020; Bousdekis

and Mentzas, 2021), the interoperability layer of the

platform must be ﬂexible and robust.

Regarding task orchestration, traditional systems

like Celery (Celery, 2024), Apache Airﬂow (Apache

Software Foundation, 2024), and Luigi (Spotify,

2024) are widely adopted for managing workﬂows,

each excelling in speciﬁc areas. Celery is ideal for

high-throughput environments with its asynchronous

task execution; Luigi provides a reliable framework

for batch data processing; and Apache Airﬂow of-

fers robust tools for complex workﬂows with com-

prehensive monitoring and dependency management.

However, Celery’s stateful architecture limits scala-

bility and fault tolerance; Luigi lacks support for dis-

tributed execution and dynamic scheduling; and Air-

ﬂow’s complex conﬁguration for dynamic or event-

driven workﬂows may be challenging for new users.

Our framework builds upon the analysis of these

strengths and weaknesses of popular orchestration

platforms, to create an efﬁcient and robust data or-

chestration mechanism for our scenarios. By adopting

a stateless, microservices-based architecture, it elim-

inates the complexities associated with state manage-

ment, enabling modular and scalable components that

can be deployed independently across distributed en-

vironments. This design simpliﬁes fault tolerance

and ensures seamless scalability, even under dynamic

workloads. Intuitive APIs provide dynamic schedul-

ing capabilities, allowing workﬂows to be triggered

by events, time schedules, or data changes. Addition-

ally, our platform supports advanced workﬂow man-

agement, such as real-time monitoring and seamless

handling of complex task dependencies, all through

a user-friendly conﬁguration process. By address-

ing these trade-offs, our solution delivers unparalleled

ﬂexibility, scalability, and adaptability.

3 ARCHITECTURE

In this Section, we present an architecture developed

for enabling digital twins of energivorous industries

which enhance performance, particularly in terms of

latency and communication efﬁciency. In this con-

text, we deﬁne an Agent as an AI-driven or simu-

lation module that receives data, performs computa-

tions, and generates results. Results from individual

Agents may be also combined in a new user-deﬁned

workﬂow, to create more sophisticated outcomes. A

workﬂow deﬁnes a series of interconnected tasks de-

signed to achieve a speciﬁc outcome. When a work-

ﬂow is in execution, we call it a pipeline. There-

fore, multiple pipelines can be associated to the same

workﬂow, with or without the same input data. The

workﬂow ﬁle is a Python script that contains a set

of platform-speciﬁc functions accessible to the user

through a dedicated Python package, in order to sim-

plify the deﬁnition of each task.

A crucial aspect of DT platforms is the orches-

tration of components, which coordinates interactions

among platform elements to transform inputs into vis-

ible outputs. In a typical DT architecture, the input

might be sensor data, while the output could be a real-

time representation of a physical component’s state,

such as the energy ﬂow in an industrial plant. This

process encompasses the automatic ingestion of het-

erogeneous data, secure storage in a persistent system,

and efﬁcient distribution to all orchestrated services.

To enhance the collaboration between simulation-

based and AI-based Agents, the Multi-Agent System

(MAS) module, leveraging centralized orchestration

functionalities, handles all these communications by

publishing the corresponding control messages inside

the central bus.

One of the distinctive features of our platform is

its intuitive workﬂow conﬁguration, allowing users to

effortlessly create and modify workﬂows using the

platform’s predeﬁned services. This design offers

exceptional ﬂexibility by enabling workﬂows setup

without requiring extensive coding, thereby lower-

ing technical barriers for users. Moreover, leveraging

an asynchronous communication, our platform allows

independent workﬂows to proceed without interfer-

A Distributed Event-Orchestrated Digital Twin Architecture for Optimizing Energy-Intensive Industries

339

ence, enabling workﬂows to continue running while

AI computations are underway. This concurrent ex-

ecution is supported by a scalable, stateless design

where each component executes tasks independently,

treating every request as self-contained. This archi-

tectural approach enables seamless cloning of compo-

nents to handle high loads, ensuring that the platform

can dynamically scale to meet complex, high-demand

requirements. Finally, our solution provides an efﬁ-

cient workﬂow scheduling via APIs, supporting both

data-triggered and time-triggered events.

This framework provides four key functionalities:

Data Gathering and Management, Data Orchestra-

tion, Data Processing and Platform Integration, as il-

lustrated in the architecture diagram of Figure 1. The

Data Gathering and Management functionality aims

to manage the data ﬂows from external sources to

platform components. In particular, it is in charge

of data gathering, storage, querying and distribution.

The Data Orchestration functionality is designed to

conﬁgure, deploy, monitor, and manage the execu-

tion of all the platform pipelines. The main com-

ponent involved is the MAS, that acts as an orches-

trator, efﬁciently distributing and executing Agents

based on the data they require from the database. The

Data Processing functionality is implemented by AI

models and simulation services, such as energy cost

forecasting. It enables AI-driven decision making

by exposing the results in a platform front-end that

hosts Graphical User Interfaces (GUIs) and data vi-

sualization tools. The Platform Integration function-

ality manages the whole infrastructure by integrating

and deploying automatically the last stable versions of

all the architecture modules, while ensuring security

with customized authentication tools. In addition, this

macro block is responsible for automatically trigger-

ing the retraining of the AI services, in order to avoid

the risk of concept and data drifting.

The following paragraphs describe the role of each

module depicted in Figure 1, highlighting the partic-

ular features they add to the platform.

3.1 Data Interoperability and Storage

The Data Interoperability and Storage (DIS) compo-

nent facilitates the integration and management of

data from a multitude of sources and protocols. The

primary objective of this component is to establish a

uniﬁed interface for the collection, storage, and re-

trieval of data from heterogeneous data sources. The

compliance of this component with the Data Handler,

and therefore with the DT architecture, is guaran-

teed by the exploitation of application communication

protocols such as HTTP and MQTT, which charac-

terize the northbound of the supported interoperabil-

ity frameworks, like EdgeX (Edge Foundry, 2020),

Apache NiFi (Apache NiFI, 2006), and Eclipse Kura

(Eclipse Kura, 2014).

3.2 Communication Backbone

The Communication Backbone (CB) facilitates the

interaction between different Agents and services

within the platform. It ensures efﬁcient routing of

data and requests between Agents responsible for dif-

ferent tasks, according to the pipelines requested by

users. The success of this system hinges on an ab-

stracted asynchronous communication proxied layer,

which allows each service to receive the necessary

task information only when requested by the MAS

and, consequently, by the user. This decoupling en-

sures that Agents are not responsible for communi-

cation or orchestration, thereby maintaining the plat-

form’s ﬂexibility and adaptability to changes or im-

provements. The models used to format the messages

exchanged within the platform depend on the speciﬁc

communication. In particular, three data models are

actually adopted: one for the Agents, one for the Data

Handler and one for the Multi-Agent System.

3.3 Agents

The orchestrated Agents represent the operational

components of the platform. They can be either

simulation-based or AI-based, and they perform user-

requested computations, each handling a speciﬁc part

of the process. The MAS coordinates the informa-

tion ﬂow between these Agents to provide an aggre-

gated result to the user. In this way, the DT platform

ensures that these components cooperate seamlessly

to achieve a speciﬁc outcome. Each Agent operates

with an abstracted communication layer, receiving

only their speciﬁc tasks without full platform visibil-

ity. The physical communication between the Agents

and the Metadata and Caching Storage (MCS) allows

the formers to have access to the data they need to per-

form their activities, avoiding to exchange large mes-

sages on Kafka.

3.4 Multi-Agent System

The Multi-Agent System (MAS) serves as the plat-

form’s data orchestrator, initiating the Agents neces-

sary for user-requested pipelines. It manages asyn-

chronous operations, enabling the platform to handle

new user inputs while executing past workﬂows. Con-

nected to the orchestrated Agents and the Data Han-

dler, the orchestrator distributes computational loads

IoTBDS 2025 - 10th International Conference on Internet of Things, Big Data and Security

340

Digital Twin Plaorm

Data Interoperability

and Storage

Data Processing

Communicaon Backbone

Mul-Agent

System

Agent 1

Data Handler

Secure Plaorm Integraon and Management Tools

Plaorm

Frontend

User APIs

Gateway

/git

/api

/platform

Metadata

and

Caching

Storage

VPN Link

Core

Components

Supporng

Components

Services

Logical

Interacon

Kaa

Communicaon

Other

Communicaon

Plaorm

Entrypoints

Gateway

Agent 2 Agent 3

…

Agent n

Data Gathering and Management Data Orchestraon

Plaorm Integraon

Plaorm

Funconalies

Figure 1: Digital Twin Platform Architecture.

and ensures tasks have the necessary data to proceed.

This conﬁguration enables efﬁcient pipeline manage-

ment and task execution within the platform.

The MAS can be conﬁgured through REST APIs,

exposed by the User APIs Service. This module al-

lows the user to schedule periodic workﬂow execu-

tion, to start a new DT pipeline, to list all the pipelines

that are currently running, to get the result of the con-

cluded ones, and to deﬁne new workﬂows.

3.5 Data Handler

The Data Handler (DH) oversees all data requests

originating from the Agents, providing an asyn-

chronous interface between the DIS and other core

components. Hence, it handles the data requests in-

termediated by the MAS to the DIS. The uniﬁed data

model adopted by the DIS is then wrapped within

the DH messages, which are then transmitted over

the CB using the platform’s data models. The DH

ensures that core components can comprehend ex-

changed events by establishing a separation between

the DIS format and the data tasks format. Once the

DIS has completed a request, the DH publishes the

data to the MCS for Agent access and notiﬁes it to the

orchestrator.

3.6 Platform Frontend

The Frontend service provides access the platform

dashboard. The GUI allows the user to schedule

workﬂows, view a list of submitted tasks, and obtain

the results of completed tasks. Finally, it is possible

to retrieve the real-time data stream from the multiple

data sources connected through the DIS.

3.7 Security Platform Integration and

Management Tools

This module encompasses a suite of tools designed to

facilitate monitoring of platform activity and to guar-

antee the security of interactions between internal and

external actors. Health checks and metrics regarding

the CB are exploited to keep track of the load capacity

of the core modules, in the view of cloning some of

them in case of overload. On the other side, Secure

Sockets Layer (SSL) over Transmission Control Pro-

tocol (TCP) is exploited for protecting the APIs, while

an Identity and Access Management (IAM) frame-

work is responsible for the authentication and autho-

rization of users and microservices.

A Distributed Event-Orchestrated Digital Twin Architecture for Optimizing Energy-Intensive Industries

341

4 ORCHESTRATION

MECHANISM

The orchestration mechanism presented in this paper

is based on a completely stateless centralized manage-

ment system and employs asynchronous operations to

guarantee a complete service decoupling within the

platform. This centralized management system or-

ganizes the sequence of steps required to complete a

pipeline, thereby reducing the computational load of

each Agent. For clarity, the orchestration mechanism

follows a centralized paradigm in order to concentrate

the data management responsibility on a single com-

ponent. However, it relies on a cluster of Kafka bro-

kers and on stateless components, which ensure scal-

ability and fault tolerance. The stateless components

replication and the number of Kafka brokers is admin-

istered by the Kubernetes HPA. The asynchronous op-

erations, enabled by Apache Kafka, treat simulation

and AI requests as events. The components only re-

spond to inputs from the MAS, thereby simplifying

their operation. The statelessness of the system pre-

vents the unavailability of modules by ensuring that

Agents produce outputs based solely on inputs. The

MAS initiates new requests once the required data is

available, thereby maintaining Agent efﬁciency.

The distributed paradigm ensures that multiple

modules evaluate tasks, thus allowing for the dupli-

cation of overloaded services. The integration of the

orchestration mechanism into Apache Kafka enables

the system to handle stateless, asynchronous requests,

thereby maximizing the beneﬁts of distributed archi-

tectures. Furthermore, this feature enables the DT to

operate across the computing continuum. In this way,

the DT can be seen as a uniﬁed entity within a more

complex system comprising multiple instances of the

same platform deployed across different locations, in-

cluding far-edge, edge, fog, or cloud.

Figure 2 shows the sequence of steps leading the

platform to complete a pipeline requested by the user.

The backend APIs can be exploited as a high level

interface that allows users to run new pipelines. As

a consequence, the backend checks and, if neces-

sary, retrieves from the MCS the last version of the

workﬂow that has to be executed. At this stage, the

MAS is ﬁnally activated through an Orchestrator

Task message, sent on Kafka. This message acti-

vates the orchestrator, that has to start creating the

tasks that compose the workﬂow. The most impor-

tant ﬁelds of this message are the issuer, which is the

identiﬁcator of the user that has created the pipeline,

and the workﬂow ﬁle. In the example of Figure 2,

the MAS ﬁrstly requests some data by publishing the

Data Task messages addressed to the DH on Kafka,

as wrappers for the historical queries that Agent 1 and

Agent 2 need. The Data Task mainly contains the

query of the data that are useful for the pipeline execu-

tion. Once the DIS component completes these activ-

ities, the MAS can share the historical data provided

by the DIS as input for the orchestrated Agents. This

is achieved using the dedicated Agent Task message,

which includes a reference composed of two ﬁelds:

the issuer and the agent name. Depending on the

values of these ﬁelds, the corresponding Agent will

be triggered while others will not. Additionally, this

message contains the MCS data location. Finally,

the MAS can complete the pipeline by returning the

overall outcome to the user, using the Orchestrator

Task message. Since each of the above message types

is used for both requests and responses, a ﬁeld has

been provided within them to differentiate communi-

cation beginnings from ends.

5 APPLICATION SCENARIOS

The microservice architecture presented in this work

will be deployed in different industrial scenarios,

in the context of the FLEXIndustries project. The

project is a European Commission-funded initia-

tive aimed at transforming energy-intensive industries

through the dual focus on green energy optimization

and digital twin technology for enhanced operational

efﬁciency. The project involves companies from dif-

ferent production sectors, e.g., automotive, pharma-

ceuticals, and manufacturing. For all these diverse

scenarios, our DT platform offers a fully reconﬁg-

urable framework. The practical implementation of

each scenario is based on the deﬁnition of the interac-

tions between the orchestrated Agents and the func-

tionality they provide to the platform. In this way, it is

easy to build vertical customization by simply adding

the logic that determines the execution order and the

input data required to the set of Agents involved in

the realization of the scenario. To be compliant with

our DT notations, each of these scenarios could be as-

sociated to a different workﬂow. The following para-

graphs present examples of the covered use cases, fo-

cused on managing energy systems of the companies’

plants.

Efﬁciency Monitoring. Monitoring process pa-

rameters with the ultimate goal of optimizing the en-

ergy efﬁciency of speciﬁc equipment and reducing

overall energy consumption.

Renewable Energy Surplus Management. Fore-

casting and management of internal energy ﬂows to

support decisions about energy sales or storage.

Optimal Energy Source Selection. Selection of the

IoTBDS 2025 - 10th International Conference on Internet of Things, Big Data and Security

342

Mul-Agent

System

Data Interoperability

and Storage

Orchestrator

Task

Data Task 1

Agent Task 1

Data Task 2 Agent Task 2

User APIs

Pipeline (Workﬂow in Execu�on)

Workow

(Set of Tasks)

User

Run Pipeline

Request

Core

Components

Supporng

Components

Services

Kaa

Communicaon

Other

Communicaon

Metadata and

Caching Storage

Agent 1

Agent 2

Figure 2: Steps involved in the execution of a pipeline within the Digital Twin Platform.

optimal energy source in a speciﬁc moment of the day

for a given process.

Precise Energy Estimation. Accurate estimation

of the energy consumption for companies that pur-

chase electricity in ﬁxed quantities, to avoid penalties

for overuse or waste from underuse.

Optimized Scheduling. Generation of optimized

schedules for machine operations, in situations where

speciﬁc machines may need to be temporarily stopped

due to high energy consumption.

Energy Monitoring. Visual representation of a

company’s plant, including all components and aug-

mented information about energy consumption in

each process and machine.

A key feature of the platform is its ability to reuse

multiple Agents for various applications, enabling

users to create diverse workﬂows with ease. For

example, Agents providing energy price forecasting

and energy demand forecasting can be utilized across

multiple workﬂows serving different objectives. The

platform has been designed and developed based on

the requirements of energy-intensive industries within

the context of the project. However, these require-

ments are not overly restrictive, allowing the platform

to adapt and perform effectively across diverse indus-

try contexts and scenarios, both within and beyond the

energy sector.

6 CONCLUSION AND FUTURE

WORK

In this paper we have presented a distributed Dig-

ital Twin platform designed to seamlessly integrate

data from diverse sources, enabling real-time moni-

toring, predictive maintenance, and operational opti-

mization across the industrial ecosystem. This plat-

form’s distinctive architecture and design make it a

highly adaptable solution for modern energy man-

agement needs, equipped to serve multiple industries

with speciﬁc yet ﬂexible workﬂows.

A central feature of our DT platform is its user-

conﬁgurable workﬂow framework. The platform ab-

stracts the complexity of low-level operations, en-

abling users to derive actionable outputs by spec-

ifying high-level workﬂows deﬁnition, data inputs,

and algorithms. Beyond ﬂexibility, leveraging an

asynchronous communication and a stateless compo-

nent design, the platform offers exceptional scalabil-

ity and the ability to execute multiple workﬂows con-

currently. These features allow the platform to han-

dle high computational loads, making it an ideal solu-

tion for data-intensive, industrial environments where

computational power and responsiveness are crucial.

Finally, our platform offers smart scheduling for peri-

odic workﬂows, which can be triggered by data events

or time intervals.

The presented DT architecture will undergo em-

pirical testing across various industrial application

scenarios, in the context of the FLEXIndustries

project, ensuring the platform’s robustness, scalabil-

ity, and effectiveness in real world scenarios.

Future work will include the deﬁnition of a met-

alanguage for workﬂow conﬁguration and updates,

thus completely removing the need for programming

language expertise and further improving the plat-

form’s usability. This will enable the decoupling the

deployment of the platform and the workﬂow conﬁg-

uration, with the latter not requiring specialized ICT

background.

A Distributed Event-Orchestrated Digital Twin Architecture for Optimizing Energy-Intensive Industries

343

ACKNOWLEDGEMENTS

The research leading to these results has received

funding from the European Community’s Hori-

zon Europe Programme under grant agreement n.

101058453.

REFERENCES

Alberti, E., Alvarez-Napagao, S., Anaya, V., Barroso, M.,

Barru

e, C., Beecks, C., Bergamasco, L., Chala, S. A.,

Gimenez-Abalos, V., Graß, A., et al. (2024). Ai lifecy-

cle zero-touch orchestration within the edge-to-cloud

continuum for industry 5.0. Systems, 12(2):48.

Apache NiFI (2006). Apache niﬁ: Data processing and dis-

tributing system. Available at: https://niﬁ.apache.org

(Accessed: 4 December 2024).

Apache Software Foundation (2024). Apache airﬂow. Ac-

cessed: 2024-11-20.

Bousdekis, A. and Mentzas, G. (2021). Digital twins for

intelligent manufacturing: A survey of applications,

architecture, and challenges. Computers in Industry,

128:103440.

Celery (2024). Optimizing — celery 5.4.0 documentation.

Accessed: 2024-11-20.

Eclipse Kura (2014). Eclipse kura: Open-source iot edge

framework. Available at: https://eclipse.dev/kura/

(Accessed: 4 December 2024).

Edge Foundry (2020). Edgex: Open-source edge plat-

form. Available at: https://www.odyssee-mure.eu/

(Accessed: 4 December 2024).

Emmert-Streib, F. and Yli-Harja, O. (2022). What is a

digital twin? experimental design for a data-centric

machine learning perspective in health. International

Journal of Molecular Sciences, 23(21):13149.

Escrib

a-Gelonch, M., Liang, S., van Schalkwyk, P., Fisk, I.,

Long, N. V. D., and Hessel, V. (2024). Digital twins in

agriculture: Orchestration and applications. Journal

of Agricultural and Food Chemistry.

Farsi, M., Daneshkhah, A., Hosseinian-Far, A., Jahankhani,

H., et al. (2020). Digital twin technologies and smart

cities.

Harper, K. E., Malakuti, S., and Ganz, C. (2019). Digital

twin architecture and standards.

He, R., Chen, G., Dong, C., Sun, S., and Shen, X. (2019).

Data-driven digital twin technology for optimized

control in process systems. ISA Transactions, 95:221–

234.

Hemdan, E. E. D., El-Shafai, W., and Sayed, A. (2023). In-

tegrating digital twins with iot-based blockchain: con-

cept, architecture, challenges, and future scope. Wire-

less Personal Communications, 131(3):2193–2216.

Henriksen, H. J. et al. (2022). A new digital twin for cli-

mate change adaptation, water management, and dis-

aster risk reduction (hip digital twin). Water, 15(1):25.

Javaid, M., Haleem, A., and Suman, R. (2023). Digital twin

applications toward industry 4.0: A review. Cognitive

Robotics, 3:71–92.

Jwo, J. S., Lee, C. H., and Lin, C. S. (2022). Data twin-

driven cyber-physical factory for smart manufactur-

ing. Sensors, 22(8):2821.

Kubernetes (2020). Kubernetes. https://www.kubernetes.io.

Accessed: 23 June 2020.

Li, X., Zhang, Y., Liu, Y., Li, P., Hu, H., Wang, L., and

Zhang, C. (2023). A high-throughput big-data orches-

tration and processing system for the high energy pho-

ton source. Journal of Synchrotron Radiation, 30(6).

Ma, Y., Zhu, X., Lu, J., Yang, P., and Sun, J. (2023). Con-

struction of data-driven performance digital twin for a

real-world gas turbine anomaly detection considering

uncertainty. Sensors, 23(15):6660.

Megargel, A., Poskitt, C. M., and Shankararaman, V.

(2021). Microservices orchestration vs. choreogra-

phy: A decision framework. In 2021 IEEE 25th In-

ternational Enterprise Distributed Object Computing

Conference (EDOC), pages 134–141.

Nguyen, T. T., Yeom, Y. J., Kim, T., Park, D. H., and

Kim, S. (2020). Horizontal pod autoscaling in ku-

bernetes for elastic container orchestration. Sensors,

20(16):4621.

Odyssee-Mure Project (2023). About the odyssee-mure

project. Available at: https://www.odyssee-mure.eu/

(Accessed: 21 May 2024).

Onwubiko, A., Singh, R., Awan, S., Pervez, Z., and

Ramzan, N. (2023). Enabling trust and security in dig-

ital twin management: A blockchain-based approach

with ethereum and ipfs. Sensors, 23(14):6641.

Rossini, R. et al. (2020). Replica: A solution for next gen-

eration iot and digital twin based fault diagnosis and

predictive maintenance. SAM IoT, 2739:55–62.

Salim, M. M., Comivi, A. K., Nurbek, T., Park, H., and

Park, J. H. (2022). A blockchain-enabled secure dig-

ital twin framework for early botnet detection in iiot

environment. Sensors, 22(16):6133.

Sax, M. J. (2018). Apache kafka. In Kleppmann, M. and

Lange, C., editors, Encyclopedia of Big Data Tech-

nologies. Springer, Cham.

Spotify (2024). Luigi - design and limitations. Accessed:

2024-11-20.

Sun, S., Li, W., Zhang, Z., Feng, J., and Zhang, C. (2020).

Data twin driven intelligent manufacturing: A com-

prehensive review. Journal of Manufacturing Systems,

56:547–558.

Yu, W., Patros, P., Young, B., Klinac, E., and Walmsley,

T. G. (2022). Energy digital twin technology for

industrial energy management: Classiﬁcation, chal-

lenges and future. Renewable and Sustainable Energy

Reviews, 161:112407.

Zhang, X. et al. (2021). Digital twin for accelerating sus-

tainability in positive energy district: A review of sim-

ulation tools and applications. Frontiers in Sustain-

able Cities, 3:663269.

Cili

c, I., Krivi

c, P., Podnar

Zarko, I., and Ku

sek, M.

(2023). Performance evaluation of container orches-

tration tools in edge computing environments. Sen-

sors, 23(8):4008.

IoTBDS 2025 - 10th International Conference on Internet of Things, Big Data and Security

344