Projects · 4 February 2026

Attack simulations in virtualised OT environments

Attack simulations in virtualised OT environments

NB: For confidentiality reasons, some parts of this research project have been intentionally omitted from this article.

Introduction & Motivations

Industrial systems – weak links in digital warfare

In 2010, the Stuxnet attack, probably carried out by the USA and Israel against Iranian nuclear facilities, raised the alarm on the risk of cyberattacks on industrial infrastructures.
By targeting the gas centrifuges at the Natanz uranium enrichment site in Iran, this intrusion raised the alarm on the risk of cyberattacks against industrial infrastructures.
In 2024, the number of attacks targeting OT infrastructures has undergone an unprecedented increase and thus, according to the Dragos report [1], 1,693 ransomware attacks against industrial installations were recorded, representing an increase of 87% in one year. Added to these attacks are the financial consequences, where a data theft represents 5.56 million dollars according to IBM [2] exposing the energy and sensitive infrastructure sectors to heavy risks of financial losses and intellectual capital.
The threats of attacks on physical information systems are today coupled with those on virtual infrastructures.
In May 2025, a time bomb gradually deployed across the world. Invisible because virtual, it gave rise to and served as a gateway for massive attacks on the data centres of companies virtualising their applications and information systems.
Its name: the conjunction of CVE-2025-22224, CVE-2025-22225, CVE-2025-22226 [3]
Its modus operandi: the joint exploitation of memory overflows in critical hypervisor processes combined with malicious code injection to elevate access privileges.

Its consequence: the possibility for an attacker to compromise an entire virtualised IT infrastructure using VMware brand products.

Its repercussions: theft of information, data compromise, theft of industrial secrets. Combined with the change in business model that the provider of this virtualisation solution, VMware, underwent following its acquisition by the American giant Broadcom, there are multiple questions regarding the sovereignty of our data in Europe and worldwide.

We no longer live in a world where computer threats are disparate and operated by isolated groups, but rather in a world where war is synonymous with hybrid warfare combining cyberattacks and geopolitics.

Faced with these threats, many companies are considering investments in protection solutions for information systems and industrial systems. One of the most adopted solutions is to simulate attacks on critical systems upstream in order to test their robustness.


This is where two simulation methods come into play :

  • The physical method, where information and industrial systems are reproduced on a reduced scale to test their weaknesses.
  • The virtual method, where information and industrial systems are reproduced virtually in simulators.

Having studied for nearly 6 months the different virtual attack simulation solutions, it seemed interesting to me to apply my knowledge to physical systems by reproducing them in a virtualised manner. This brings in the concept of Cyber Ranges, genuine virtual laboratories dedicated to cybersecurity, which we will define in greater detail throughout the report. On the financial front, the market for these virtual solutions is estimated at 520 million dollars in 2024 with an annual growth of 15% [4].

These tools were the subject of my study throughout this research period.

Academic context

This research project follows on from a collaboration I had undertaken with my engineering school ESAIP from June to August 2025 as part of research into alternatives to proprietary virtualised attack simulation solution [5].

Within the German university TH-Brandenburg and as part of my master's level Erasmus programme, I was able to access physical industrial resources within the IoT laboratory where a control-command regulation chain using PLC programmable controllers was made available to me, thus granting me a genuine playground to compare the behaviours of physical industrial systems with their virtualised digital replicas. This infrastructure enabled me to validate and invalidate the research hypotheses addressed throughout this report.

Challenges and limitations encountered

This project encountered technical challenges that consumed the majority of the 3 months of research. The difficulties notably highlighted the issues related to proprietary solutions claiming to be open to their use cases whilst being closed regarding their source codes, thus justifying the motive and subsequent motivation to imperatively seek open-source alternatives.
Added to this main challenge was that of human resources, which I lacked throughout my research. Being the only one working on a large-scale project, I had to lower my ambitions as well as my initial guiding principle of finding an entirely open-source solution, which transformed into an in-depth analysis and correction of a proprietary solution in order to better define the prototype of an open-source solution.
On 21 January 2026, a failure of one of my 20 TB backup hard drives added a further difficulty to those already present by causing the disappearance of the entire experimentation part on simulation in a virtualised environment following the Cyber Range solution developed for this purpose, representing at least 6 months of work. I therefore had to reproduce this simulation environment from scratch in one week.

Issues with the Airbus CyberRange solution

Having become a specialist in this solution despite myself, often praised for its advanced functionality features, I quickly encountered the limitations in terms of cost, flexibility and transparency that are highlighted in the official Airbus Cybersecurity documentation. [6].

Among these setbacks, we find:

  • Defective updates causing compatibility breakdowns at the inter-API routing level of the solution.
  • Disjointed nomenclatures revealing questionable design choices.
  • Incomplete documentation which, despite its official nature, concealed the essence of the operating architecture.
  • Non-existent support.
  • Extremely slow deployment of physical solution replicas, partly due to contradictory choices regarding virtual machine deployment.

Throughout the project, I rebuilt a virtual attack simulation laboratory.

Reverse engineering work

Within the framework of research based on the development of an open-source solution, it was preferable to base my research on a proven solution and to work from a proprietary solution that was "robust and efficient" according to the companies that had adopted it [7], I undertook reverse engineering work to understand and correct the platform's malfunctions. By deconstructing the proprietary solution part by part, this work made it possible to develop an architecture and its intrinsic operating processes with a view to an open-source future.

Research problem and objectives

This research project was based on the following problem:

"Can proprietary or open-source simulation solutions faithfully simulate attacks on OT environments ? »

To address this, the following research objectives were set:

  1. State of the art: drafting of an analytical report on existing proprietary or open-source Cyber Range solutions by identifying the strengths and weaknesses of each studied solution.
    (Works based on scientific articles [9], [10], [11], [12], [13], [14], [15], [16])
  2. Design of a simulation architecture: definition of an IT/OT systems virtualisation architecture enabling the simulation of multi-surface attacks on industrial systems.
    (Works based on scientific articles [13], [17])
  3. Implementation of digital twins for PLC industrial systems.
    (The experiments were based on the PLC systems from Siemens [18], [19] )
  4. Experimental validation by comparing and analysing the feasibility of attacks based on known vulnerabilities between the real physical system and its digital twin.
    (Works based on scientific articles [20], [21])
  5. Evaluation of open-source alternatives for a future vision of the research project
    (Works based on scientific articles [22], [14], [15])
  6. Recommendations on the deployment of OT cyber ranges under strong budgetary constraints.

Research questions

  1. Which cyber range architectures allow for increased compliance with a Purdue model while meeting accessibility criteria (financial and ergonomic) and reliability whilst remaining realistic in terms of simulations?
  2. What discrepancies, blocking points and latencies can be observed between a physical system and its digital twin when the simulated system undergoes a MITM attack ?
  3. What is the reliability level of proprietary PLC simulators such as Siemens' PLCSIM Advanced?
  4. Are there viable real open-source alternatives to replace proprietary solutions such as the Airbus CyberRange?

We can thus formulate the following hypotheses:

  • Cyber Ranges are genuine low-cost and high-fidelity alternatives for simulating attacks on industrial systems.
  • Digital twins enable a faithful reproduction of possible attacks on their physical model.
  • Open-source solutions can faithfully reproduce the behaviours of proprietary solutions.

State of the art

Chronology of historical attacks on industrial systems

The growing context of hybrid warfare intertwined with strong geopolitical tensions has only accentuated the criticality of industrial systems in the most sensitive sectors of States such as energy, armaments, transport and technology. Waging "war at a distance" is now commonplace and two countries such as the United States of America and Russia have demonstrated this in an "exemplary" manner in recent years. Whether in the context of destabilisation operations, responses to attacks received or total paralysis of business sectors, we observe an increasingly sophisticated level of attacks as well as increasingly precise targeting of intended targets from the 2010s to the present day.

The following chronology only reinforces the imperative to develop simulation environments to study and develop countermeasures against the scale and consequences of these threats.

History of major cyberattacks targeting industrial systems

  • Year · Attack name · Description and Impact
  • 2010 — Stuxnet [23] & [24] — A malware specifically targeting Siemens S7-300 PLCs in Iran destroyed 1,000 Iranian centrifuges by modifying the frequency of voltage converters, causing premature wear on the systems.
  • 2015 — BlackEnergy [25] — Attack on the control of the Ukrainian electrical grid resulting in a 6-hour outage affecting 230,000 households. First confirmed blackout caused by a cyberattack.
  • 2016 — CRASHOVERRIDE [26] — Modular malware specifically designed to attack electrical distributors, targeting Ukrainian substations using IEC 104, IEC 61850 and OPC DA protocols used by Siemens Siprotec systems.
  • 2017 — TRITON/TRISIS [27] — First attack targetting automated safety systems serving as safeguards in case of major problems. Schneider Electric Triconex systems were targeted to cause material damage in a petrochemical plant in Saudi Arabia.
  • 2021 — Colonial Pipeline [28] — Ransomware attack paralysing the largest American pipeline for nearly 6 days. The Americans paid 4.4 million dollars in ransom.
  • 2022 — Industroyer2 [29] & [30] — Malware variant of Industroyer developed by Russian hackers Sandworm against Ukraine targeting electrical substations using the IEC 104 protocol.
  • 2023 — Unitronics [31] — Attack on Unitronics Vision PLCs connected to the Internet affecting water treatment facilities in the United States and Israel – Defacement of interfaces and disruption of industrial processes.
  • 2024 — FrostyGoop [32] — ICS malware targeting Modbus TCP protocols, disrupting heating in 600 buildings in Ukraine.

Among the previously mentioned attacks, the CrashOverride attack of 2016 was the first to bring multi-surface OT threats to the forefront. By using specific tools targeting multiple protocols, respectively IEC 101, 104, 61850 and OPC DA, it was proof of in-depth knowledge on the part of the attackers of the control systems specific to Ukrainian electrical substation systems. It was also one of the attacks to reveal a dual phase:

  • The first involving disruption and paralysis of the electrical grid.
  • The second targeting the paralysis of the control systems of these networks by self-destructing the operating systems hosting the control-command solutions.

If we now come to the common elements shared by all these attacks, it is appropriate to establish a model representing the different levels of network segmentation facilitating the categorisation of attacks and their impacts in an approach of simulation and virtualisation thereof.

The choice was thus made to adopt the PURDUE model

The PURDUE model applied to network segmentation of industrial infrastructures

The Purdue model, named after the university that defined it, more commonly known as the Purdue Enterprise Reference Architecture (PERA), has become a standard in industrial network segmentation as supported by the work of T.J. Williams in the 1990s [33].
It aims to define a hierarchical model following 5 distinct functional levels, from functional equipment to the application software layers of an enterprise or industry.
This model was subsequently formalised and made complete by the IEC 62443 standard [34] by adding demilitarised network zones (DMZ) and specific security requirements per level. Thanks to this formalism, the categorisation of attacks by application levels was made easy when it came to defining attack surfaces and the protection level of the targeted zones.

Thus, the IEC 62443 standard is defined in four security levels:

  • Level 0: The attack surface implemented no security measures.
  • Level 1: The attack surface was protected only against accidental misuse (strict access levels, password policies and network segmentation by VLAN).
  • Level 2: The attack surface was protected against simple threats (multi-factor authentication).
  • Level 3: The attack surface was protected against sophisticated and targeted threats notably through the use of encrypted communications or by the implementation of real-time threat detection tools combined with forensic analysis plans.
  • Level 4: The attack surface was protected against persistent and high-risk threats with the help of physical and software protections implementing regular testing (blue team / red team field training exercises).

    You will find on the following page the PURDUE architecture represented graphically and summarised in table form as applied in the context of my study to OT industrial systems.
PURDUE model applied to OT/IT network segmentation
PURDUE model applied to OT/IT network segmentation

Description of the PURDUE model levels

  • Level · Designated item · Components and associated functions
  • Level 0 — Physical Process — Physical equipment: sensors, actuators, valves, motors acting or interpreting on physical or mechanical phenomena
  • Level 1 — Basic Control — Programmable logic controllers (PLCs), RTUs, safety controllers (SIS). They run control and regulation loops in real time.
  • Level 2 — Supervision — SCADA systems, HMI, engineering stations, data historians. They serve as a human-machine interface and are interfaced with the logic of archiving actions.
  • Level 3 — Operations — MES (Manufacturing Execution Systems). They manage and order production by optimizing operations.
  • DMZ — Demilitarized Zone — Firewall, antivirus, SIEM. Isolation between IT and OT.
  • Level 4-5 — Enterprise/Cloud — ERP, CRM, management systems, cloud services. Business and analytics functions.

Having defined the surfaces, attack zones and cybersecurity measures, let us complete with an overview of existing solutions:

Virtual laboratories and Cyber Ranges: Overview of existing solutions

Virtualisation approaches

According to the research by Ukwandu et al [11] augmented by NIST formalisations [35], four main approaches can be undertaken to design a Cyber Range.
Among these approaches, we find::

  • Simulation, which consists of modelling system behaviour mathematically, representing 50% of the mainly studied solutions. This approach represents the safest and least costly, but it faces the limitations of faithfully reproducing real-world infrastructures.
  • Emulation, which corresponds to the replication of operations of a target system through a mirror system, representing 38% of the mainly studied solutions. This approach offers a better level of fidelity and realism by simulating both the real infrastructure and its associated nominal behaviours such as traffic and network addressing from the real world.
  • Overlay, which consists of using an entire existing network and infrastructure on which testing and attack tools are superimposed, representing 8% of the mainly studied solutions. This solution appears to be the one with the most optimal fidelity levels, but with a significant financial trade-off, as it involves having real physical equipment such as servers, OT systems, network and storage.
  • The physical approach, which represents only 4% of the mainly studied solutions, mainly due to its extremely high costs.

Here is a graphical summary:

Summary of strengths and weaknesses of the types of Cyber Ranges
Summary of strengths and weaknesses of the types of Cyber Ranges

Overview of existing solutions

In order to complement the previously defined approaches relating to the design and operation of Cyber Ranges, here is a quick overview of existing proprietary and open-source solutions:

  • KYPO Cyber Range [14], developed by Masaryk University (Czech Republic), is an open-source Cyber Range that serves as a cybersecurity training platform. Based on OpenStack as a virtualisation solution, it offers a system of attack scenarios and incident response. KYPO is distinguished by its nature federating multiple instances such as containers, microservices and open-source software, and its ability to integrate with standardised evaluation frameworks. Its limitations lie in its steep learning curve and its substantial hardware requirements with knowledge of OpenStack clusters as a prerequisite.
  • AIT Cyber Range [15], developed by the Austrian Institute of Technology, is a Cyber Range based on Docker containers and terraform. The design can be extended to deploy the solution on a wide variety of infrastructures (both on-premises and in the cloud). Its operation is based on predefined templates for each scenario (ransomware, APT, network attacks). It is distinguished from other existing solutions by its automated scoring tool, highly sought after at the academic level. The solution is specially designed for CTF (Capture the Flag) exercises.
  • ICSSIM [36] , developed on an open-source Python base, aims to simulate industrial control systems. It supports the modelling of complete SCADA networks using various protocols (Modbus, DNP3, OPC) and attacks such as MitM, spoofing or denial of service. Its strength lies in its scalability and its very detailed documentation. However, we must contrast these advantages with its low simulation realism compared to real firmware emulators.
  • OpenPLC [22] developed on an open-source runtime implementing the IEC 61131-3 standard for PLCs. Its strength is its cross-platform availability such as Raspberry Pi, computers running operating systems such as Windows or Linux as well as microcontrollers. Added to its advantages is the support for Modbus TCP and DNP3 protocols. OpenPLC was particularly relevant in the context of my study, as it allowed me to generate and simulate PLCs at no cost.
  • Airbus CyberRange (LADE) [5]

The Airbus CyberRange solution, based on the LADE (Life and Death Engine) principle, offers advanced features for creating topologies and automated attack scenarios. The graphical interface allows for the on-the-fly definition and design of complex infrastructures combining virtual machines, Docker containers and network configurations. The "work zones" allow for partitioning resources between different users or simulation frameworks. This was the cornerstone of the entire research project conducted.

Industrial protocols, vulnerable to cybersecurity considerations.

The study framework of this project, residing primarily on industrial infrastructures, calls upon the simulation of the most widespread industrial protocols in the field. These protocols were historically designed for their robustness and performance without concern for security considerations, as specified in the report by Alanzi et al. [20]
This security debt has become, with modern hybrid warfare, a genuine attack vector around which the guiding principle of this project has been structured.

Among these protocols, we find:

  • Modbus, operating on the TCP layer.

Developed by Modicon in 1979 and adapted to work on the TCP/IP layer in 1999, it represents one of the most widespread protocols in industry due to its simplicity and openness, thus making it highly vulnerable.

Among its vulnerabilities we find:

  • The absence of authentication allowing any client to send commands without prior identification, which enables impersonation of legitimate control stations.
  • Data sharing in clear text. This absence of encryption allows the interception and modification of frames during MitM attacks.
  • The absence of frame integrity verification. This lack of signature or cryptographic checksum allows the injection of malicious commands.
  • The absence of broadcast address control. A simple transmission to address 0 allows reaching all controllers and thus facilitates the reconnaissance phases for large-scale attacks.
  • S7comm and S7comm+, both developed by Siemens, have been largely in the research spotlight, thus becoming truly documented proprietary protocols vulnerable to attacks [20].
    The vulnerabilities affecting them are this time version-dependent:
  • The S7-300/400 versions were of critical severity, as their lack of authentication granted the possibility of remotely stopping the PLC CPUs.
  • The S7-1500 versions introduced a challenge-response mechanism while remaining vulnerable to frame replay attacks if these mechanisms are poorly configured
  • OPC UA (Open Platform Communications Unified Architecture), developed by the OPC Foundation from 2004 and standardised under the IEC 62541 standard in 2011, represents a certain advancement in industrial communication security. Unlike historical protocols, it natively integrates authentication mechanisms using X.509 certificates, message encryption and signing according to three security modes (None, Sign, SignAndEncrypt), mechanisms noted by the German BSI [37]. Despite these technical praises "on paper", the practical reality is quite different, as numerous vulnerabilities have been documented by research, among which we find:
  • A frequent activation of the "None" security mode by default in commercial implementations, effectively removing all protection and exposing communications to interception and modification, as studied by Erba et al. in 2022 [38].
  • Critical flaws with 17 zero-day vulnerabilities discovered by Kaspersky ICS CERT in 2018, including heap memory corruptions allowing remote code execution with elevated privileges (Cheremushkin & Temnikov, 2018) [39]
  • Massive certificate validation defects, where nearly 80% of the analysed industrial products present security flaws allowing Man-in-the-Middle type attacks [38]
  • An extended attack surface linked to the complexity of the protocol, having led to the discovery of more than 50 vulnerabilities by Claroty Team82 between 2020 and 2023[40]

Digital twins in the service of cybersecurity

After having drawn up an overview of the protocols and possibilities for attack studies intended to be replicated in the experimental part of the project, it seemed fundamental to define a concept that was touched upon throughout the previous paragraphs, but which today constitutes a real asset for the fidelity of attack simulations on virtual systems. For a formal definition, the digital twin is a virtual replica of a physical system reflecting its state and behaviour in real time.

Originally, the digital twin was developed for predictive maintenance and process optimisations. Its application in the field of industrial security is booming, as it allows the creation of virtual copies of OT infrastructures with a view to conducting infiltration tests, attack simulations and vulnerability analyses (pentests) without impacting real production systems. This concept thus enables cost reduction in research and development stages but also facilitates the improvement of system quality and security over a very short time span.

If we were to define its usefulness in cybersecurity in one sentence, the digital twin must comply with the following rule:

  • The digital twin must faithfully reproduce the functional behaviour of the real system while replicating its network characteristics and potential vulnerabilities.

It therefore becomes interesting to study the following issue:

"To what extent does a virtualised programmable logic controller reproduce the vulnerabilities documented in the academic literature?" This is the question that will be addressed throughout the experimental section.

Experimental section

Experimental plan

Context

In order to address the issues raised in the previous paragraphs, the following experimental plan was established with the objective of evaluating the vulnerabilities of the OPC UA protocol on the digital twin of the Siemens PLC in the laboratory. The initial hybrid approach is maintained with the OPC UA protocol for I/O processes and S7Online for the HMI.

Simulated architecture

The simulated architecture retains its physical network configuration, only the IP resource addressing changes.:

  • PLC Siemens: 192.168.10.10
  • HMI WinCC Unified PC RT: 192.168.10.20
  • Hacker PC: 192.168.10.50

The libraries used for the attack codes refer to the Python libraries "opcua-asyncio & python-opcua".

Project variables

Tableau des variables du projet PLC

  • Variable · Type · Description
  • Temp_Sensor — REAL — Temperature sensor (attack target)
  • Too_HOT — BOOL — High temperature alarm
  • Too_COLD — BOOL — Low temperature alarm
  • Temp_Good — BOOL — Temperature within normal range
  • lamp_red — BOOL — Red lamp - critical alarm output
  • lamp_yellow — BOOL — Yellow lamp - warning output
  • green_process — BOOL — Normal process indicator (green)
  • red_process — BOOL — Process in alarm indicator (red)

Phase 0 — Initial OPC UA configuration

Goal: Verify that PLCSIM Advanced faithfully reproduces the nominal functional behaviour of its physical model present in the TH Brandenburg lab

Protocol:

  • Deployment of an identical thermal regulation programme on PLCSIM Advanced.
  • Connection of the WinCC HMI client to the virtual PLC.
  • Capture of S7comm traffic from PLCSIM Advanced via Wireshark
  • Recording of metrics under nominal conditions

Metrics of phase 0 of the experimental protocol

  • Metric · PLCSIM Advanced · Expected Results
  • Successful writing on the PLC memory (Yes / Non) — YES
  • Operational Regulation Program (Yes / No) — YES
  • Communication with the HMI (Yes / Non) — YES
  • Consistency of the output states according to the ST program logic — YES
  • Preliminary captures of the network activity — N/A

Phase 1 – Simulated MitM vulnerabilities on the S7Comm protocol in comparison with the scientific literature

Passive frame interception

Goal: Verify that the OPC UA traffic emulated by PLCSIM Advanced can be intercepted in clear text in accordance with the vulnerabilities documented in the scientific literature [20].

Protocol:

  • Spoofing via Ettercap between the human-machine interfaces and the simulated PLC.
  • S7comm Traffic Capture
  • Frame analysis based on the readability of values, commands, and identifiers.

Metrics on passive frame interception

  • Metric · PLCSIM Advanced · Expected results
  • ARP spoofing successful (Yes/No) — Yes
  • Plaintext readable OPC UA frames (Yes / No) — Yes
  • Extracted tag values (Yes / No) — Yes
  • Nodes of identified variables (Yes / No) — Yes

Injection of falsified values

Goal: Verify that a modification of the Temp_Sensor variable triggers the expected reaction from the virtual PLC.

Protocol:

  • Anonymous connection to the server
  • Identification of the node number corresponding to the temperature sensor (Temp_Sensor)
  • Injection of an erroneous high temperature (88°C).
  • Observation of the triggering of Too_Hot alerts and lamp_red outputs.
  • Verification that the WinCC HMI displays the injected value.
  • Injection of a very low value (-35°C) and observation of the expected behaviour.
  • Return to nominal values (24°C)

Metrics on MitM injection on the S7Comm protocol

  • Metric · PLCSIM Advanced · Expected results following [20]
  • Unauthorized Write Successful (Yes/No) — Yes
  • Alarms triggered (Yes/no) — Yes
  • Triggered outputs (Yes/no) — Yes
  • Display of Tampered Information on the HMI (Yes/Non) — Yes
  • Behavior in accordance with the physical model (Yes/no) — Yes

Expected behavioural metrics following data injections

  • Data injection · PLC Reaction · Expected HMI Results
  • 85°C (high) — Too_HOT=TRUE, lamp_red=ON — Red light on / "Too Hot" indication
  • -10°C (low) — Too_COLD=TRUE — Indication « Too Cold »
  • 25°C (normal) — Temp_Good=TRUE, green_process=ON — Indication « Good »

Evaluation of security modes

Goal: Comparison of attack resistance according to the security mode implemented

Protocol:

  • Successive configuration of each OPC UA security mode: None, Sign, SignAndEncrypt.
  • Attempt for each mode to connect without valid certificates, intercept and modify values by injection
  • Documentation of observed errors and behaviours
  • Verification of the independence between S7Online (HMI) and OPC UA communications

Expected metrics for the evaluation of security modes

  • Testing · None · Sign · SignAndEncrypt
  • Anonymous login — Accepted — Refused — Refused
  • Readable interception — In plaintext — Signed — Encrypted
  • Editing frames — Possible — Detected — Impossible
  • Injection values — Accepted — Rejected — Rejected

1Target architecture based on the Purdue model

The architecture of the simulated solution extends from levels 0 to 3 of the PURDUE model. It enables the simulation of a realistic scenario where any attacker could compromise the supervision network as well as the control network to maliciously and discreetly manipulate the targeted industrial process.

PURDUE architecture of the experiment
PURDUE architecture of the experiment

Threat model followed by the attack scenario

The threat model considered is the most prevalent one currently, involving a remote compromise enabling an attacker to carry out a MitM attack.

Attack sequence diagram
Attack sequence diagram

Following the MITRE Framework, a reference system for classifying and categorising attacks, the implemented experiment covers the following aspects:

  • Identification (T0846): Identification of equipment on the OT network via ARP scanning and traffic analysis.
  • MITM positioning (T0830): ARP spoofing to intercept traffic between the HMI and the PLC.
  • Data manipulation (T0836): Modification of sensor values transmitted and interpreted by the PLC
  • Impact on the processes (T0831): The PLC reacts to the falsified data, causing a malfunction of the nominal process.

Comparative Cost Analysis: Physical vs Virtualised Infrastructure

Attack simulations in virtualised OT environments

As discussed throughout the state of the art, the flagship argument for virtualisation is its massive reduction in infrastructure costs. Here is a summary diagram of the costs of a physical solution versus its virtual twin based on catalogue prices from distributors. [41]

To have an order of magnitude on the savings achieved by opting for a virtual attack simulation, here is a detailed budget for a complete physical infrastructure vs its 100% virtual twin.

Detailed budget for a complete physical infrastructure

  • Component · Reference · Price (€)
  • CPU S7-1516-3 PN/DP — 6ES7516-3AN02-0AB0 — 4 800
  • Module DI 32x24V DC — 6ES7521-1BL00-0AB0 — 450
  • Module DO 32x24V DC/0.5A — 6ES7522-1BL01-0AB0 — 520
  • Module AI 8xU/I/RTD/TC ST — 6ES7531-7KF00-0AB0 — 420
  • Module AO 4xU/I ST — 6ES7532-5HD00-0AB0 — 310
  • Rack S7-1500, alimentation PM 190W, connectors — N/A — 600
  • HMI MTP1500 Unified Comfort Panel 15" — 6AV2128-3QB06-0AX1 — 3 000
  • Capteurs PT100 et transmetteurs 4-20mA — Divers — 300
  • TIA Portal V20 Professional (licence) — 6ES7822-1AA08-0YA5 — 3 000
  • TOTAL PHYSICAL INFRASTRUCTURE COSTS — 13 400 €

Detailed budget for a 100% virtualised infrastructure

  • Component · Reference · Price (€)
  • TIA Portal V20 Professional (licence) — 6ES7822-1AA08-0YA5 — 3 000
  • PLCSIM Advanced V7 (licence) — 6ES7823-0BA00-1BA0 — 3 750
  • WinCC Unified Comfort (licence runtime) — 6AV2154-2GB01-6AA0 — 780
  • TOTAL VIRTUALIZED INFRASTRUCTURE COSTS — 7 530 €
  • SAVINGS — 5 870 € (-44%)

This analysis thus demonstrates that a virtualised infrastructure is synonymous with considerable cost savings. For academic and research environments, this cost reduction is significant and allows for diversifying research activities, making any system upgrade agile (a software update does not represent the same level of investment/labour as a physical upgrade). These cost savings are valid for a research and development approach. It is important to contrast these benefits with the necessity of carrying out tests on a physical infrastructure before any production deployment.

Implementation of the experiment

Complete architecture of the solution

By basing the implementation of the experiment on a virtual laboratory, I decided to use a fully virtualised architecture. Here is its architecture diagram.

Architecture diagram of the virtual laboratory
Architecture diagram of the virtual laboratory

Based on the Airbus CyberRange, I had to carry out fairly extensive reverse engineering to make functional a version that was not.

The development of this virtual laboratory solution was thus divided into two phases:

The first aimed at performing reverse engineering on an existing solution to resolve its malfunctions and understand its intrinsic operation.

The second aimed at rewriting functional source code for the continuation of experiments.

Hardware and architectural solution

On the hardware side, the Airbus CyberRange is based on two main components: an Open vSwitch virtual switch based on an Ubuntu LTS 22.04.5 operating system and a Lenovo P920 server/workstation with a 5 TB HDD storage capacity, combined with 128 GB of RAM and a computing capacity of 36 cores clocked at 3 GHz from two Intel Xeon Gold 6150 processors.

The server hosts the entire functional solution of the Airbus CyberRange, based on a VMware vSphere ecosystem emulated in VMware Workstation 16 (for testing and development purposes).

Backend of the Airbus CyberRange
Backend of the Airbus CyberRange

It is connected to the virtual switch to offer three entry points to the outside world:

  • The first corresponds to the administration network for configuring and administering the hypervisor and its virtual machines
  • The second corresponds to the user network through which the graphical simulation interface is accessed.
  • The third corresponds to a WAN access configurable according to the needs of each simulation to replicate a network extracted from the real world.

This compartmentalisation is ensured by VLANs shown in the following capture:

VLANs of the virtualisation architecture
VLANs of the virtualisation architecture

The solution is based on a virtualisation ecosystem well known to the general public, VMware vSphere, a platform that combines the hypervisor (named ESXi) along with a range of tools for virtualising networks and storage.

View of the hypervisor and Airbus CyberRange components
View of the hypervisor and Airbus CyberRange components

As we can see in Figure 9, 4 virtual machines constitute this virtual laboratory. On the main machine named Lade runs a set of microservices to orchestrate the deployment of virtual IT infrastructures and execute attack and test scenarios in a secure environment.

These four virtual machines that we find in Figure 9 correspond to the following functions:

  • LADE-Gateway : Correspond au pare-feu autorisant la connexion entre la plateforme et le monde extérieur (Web, Réseau Utilisateurs, Réseau d'administration). Il centralise, grâce à la solution OPNSense, VLANs et pare-feu.
  • LADE-CR-SRV : Correspond au serveur principal de noms résolvant tout le domaine local de la plateforme, soit « cyberrange. local ». Il centralise le serveur DNS et le serveur de temps NTP
  • LADE-ROUTER : Correspond au routeur dédié VMware vCenter qui est intégralement géré par le LADE-SERVER. Il contient un serveur de noms, un serveur NTP, un serveur de logs et un service de redirection des workzones vers le LADE-SERVER.
  • LADE-SERVER : Correspond au cœur principal de la solution, articulé autour de conteneurs Docker partitionnés en micro-services fonctionnels : interface web, interaction avec l'hyperviseur, bases de données, stockage de fichiers.

In the work zones, we find two virtual machines corresponding to the following functions:

  • WORKZONES-GATEWAY: Manages network gateways using Docker services, configures network ports using OpenVSwitch. In parallel, a dedicated network for launching scenarios and direct actions on virtual machines is created: LADE-Agent. (We will discuss this further, as this was the entry point for all my implementation and testing issues.)
  • WORKZONES-Containers: Manages Docker containers present on the work zones using OpenVSwitch for connecting containers to LADE networks.

To better understand this network architecture, refer to the diagram on p.23 where we find on the upper level the main LADE infrastructure with the LADE Server, the core of the solution, which is connected to the LADE-GATEWAY. Using NAT (Network Address Translation) rules, this gateway manages the following network flows:

  • HTTP/HTTPS traffic to the WAN
  • Authorised NAT/VLAN traffic from workzones to LADE services
  • Dedicated administration and user network connection (Admin/User Network)

The LADE Router positioned at the centre of the solution manages the interconnection between the LADE orchestrator service and its associated work zones.

Finally, the lower level represents the interconnection between the Workzone Gateways and the application containers.

Licensing framework used - budget estimation

Following a presentation during the course of this project to one of the researchers in the research department at Brandenburg University, Mr Emil Schmidt, I agreed to add this section on the legal usage frameworks of the virtualisation solution, necessary in strict usage frameworks imposed by potential partner organisations.

While the physical architecture does not present any particular problem at first glance, the software solution and more specifically the hypervisor is another matter entirely. Following the acquisition by Broadcom in 2023, the new owner switched from lifetime hypervisor licences to annual subscription licences. This is a real problem in the context of a potential project where costs would be restricted, as it would result in an ongoing non-compressible annual expense (which would still be lower than the estimated costs of a test solution using 100% physical hardware).

Now based on costs per processor core, the VMware vSphere solution comprising hypervisor and cluster orchestrator is estimated at €50 per year. It is thus estimated that for the solution used in the tests but also usable in production conditions, the cost would be 36 × 50, i.e. €1,800 in non-compressible licence fees. Combined with a one-off budget of €4,400 in the first year (including said licence fees)

Costs of the implemented solution

  • Component · Unit price · Quantity · Total
  • Lenovo ThinkStation P920 (Refurbished Chassis) — ~€1,500 — 1 — €1,500
  • Intel Xeon Gold 6150 (refurbished) — ~200 € — 2 — 400 €
  • RAM DDR4 ECC 128 Go (8 × 16 Go) — ~400 € — 1 — 400 €
  • Disques HDD 5 To (configuration RAID) — ~300 € — 1 — 300 €
  • Annual License Prices — 1800 — 1 — 1800 €
  • TOTAL 1st year — 4400 €
  • TOTAL following years — 1800 €

NB: this budget does not include the licence cost of the Airbus CyberRange solution.



Regarding the legal framework for use, 3 points should be clarified:

  • The first concerns its proprietary closed-source nature: VMware prohibits any modification of the code and any creation of derivative solutions
  • The second concerns its military-grade certification: it possesses a STIG (Security Technical Implementation Guide) from the American Defense Information Systems Agency, available at this address VMware STIG Program Overview
  • The use of the solution is authorised for the following domains:
  • Military (non-offensive: offensive guidance, nuclear, biological, chemical weapons excluded)
  • Public
  • Commercial

Regarding its certifications, the solution meets the criteria:

  • FIPS 140-2 level 1
  • ISO 27001
  • ISO 27017
  • SO, 2 Type 2
  • NIS2
  • RGPD
  • DORA
  • ANSSI pour la France

Military grades:

  • DISA STIG
  • NIST 800-53
  • FedRAMP
  • CNSSI 1253
  • Cyber Essentials+

Reference physical infrastructure

The TH Brandenburg IoT laboratory served as the reference for the virtualised solution. The simulated temperature regulation process was thus based on:

  • A Siemens S7-1500 PLC controller composed of the CPU S7-1516-3 PN/DP processing unit and its digital and analogue input/output modules
  • An HMI MTP1500 supervision interface simulated by a WinCC process in the virtual laboratory
Siemens S7-1516-3 PN/DP PLC with its I/O modules
Siemens S7-1516-3 PN/DP PLC with its I/O modules
WinCC HMI interface
WinCC HMI interface

The PLC process uses the following 13 variables, which will be used during a reconnaissance phase carried out by the malicious user:

Variables of the replicated process
Variables of the replicated process

Configuration of the virtual environment

After making a mirror copy of the TIA Portal engineer station, I replicated the solution on the Airbus CyberRange platform

Import of the TIA Portal engineer station into the Airbus CyberRange
Import of the TIA Portal engineer station into the Airbus CyberRange

Following this import, I just needed to install the VMware guest tools to facilitate the interaction of the virtual machine with the hypervisor

Virtual machine of the TIA Portal engineer station
Virtual machine of the TIA Portal engineer station

All that remained was to create a simple virtual network to simulate the attack vector network portion: thanks to the drag-and-drop deployment solution, there was nothing simpler for deploying virtual machines and networks.

The virtual experiment plan was thus structured around the following addressing and machine configuration.

Addressing and virtual configuration of the attack

  • IP address · VM · OS · Resources · Software & Role
  • 192.168.10.10 — PLC — Windows 10 Pro — 4 vCPU, 16 GB RAM — Engineering Station: TIA Portal V20, configuration and programming- PLCSIM Advanced V7: CPU emulation of type s7-1500
  • 192.168.10.50 — attacker — Windows 10 Pro — 4 vCPU, 8 GB RAM — Attack tools: Ettercap, Wireshark, MITM scripts in python
  • 192.168.10.20 — HMI — Windows 10 Pro — 2 vCPU, 4 GB RAM — WinCC Unified Comfort : interface HMI de supervision
Digital twin of the TH Brandenburg ICS lab
Digital twin of the TH Brandenburg ICS lab

We find in Figure 15 the graphical representation of the ICS laboratory on the Cyber Range. All that remained was to configure the HMI interface by installing the WinCC Unified software, then to install the PLC simulator, PLCSIM, on the PLC/engineer station machine.

Real-time configuration of the 3 machines
Real-time configuration of the 3 machines

The agility that the Airbus CyberRange solution provides is so enhanced that it was possible for me to configure all the virtual instances at the same time.

All that remained was to connect the various elements of the PLC process to its virtual network model before launching an attack simulation on the system.

The PLC-HMI link was addressed accordingly following the new network addressing plan.

PLC/HMI network addressing
PLC/HMI network addressing

All that remained was to carry out attack tests in order to obtain results in response to the hypotheses posed.

Below we find the complete virtual attack laboratory, the digital twin of its physical solution.

The digital twin under attack testing conditions
The digital twin under attack testing conditions

Please find on the following page the experimental results.

Experimental Results

Phase 0 Validation under nominal conditions

The fundamental validation of obtaining a virtual model faithful to its physical model was successfully carried out.

Writing the programme to the virtual PLC
Writing the programme to the virtual PLC

As we can observe in this capture attached above, the equivalent of the virtualised PLC, PLC Sim, was successfully flashed with the programme present on the laboratory PLC. It remained to be seen whether its operation under simulated conditions corresponded to reality.

Virtual HMI Display
Virtual HMI Display

As we can observe in the previous capture, the replica of the virtual HMI indicates proper nominal operation. The indicators are green and the virtual temperature is correctly displayed (as the PLC has no "virtual" values coming from its inputs, it displays, as in its initial programming, a minimum capped value of -40°C corresponding to state 0 of the analogue input.)

The output consistency is in accordance with the displayed result:

The -40°C state corresponds to a "low" temperature obtained by the "Temp alert – Cold" output observed on the HMI.

We can thus draw up the following table of experimental metrics.

Table summarising the experimental results of phase 0 of the experiment

  • Metric · PLCSIM Advanced · Expected
  • Successful PLC memory write — YES — YES
  • Operational regulation programme — YES — YES
  • Communication with HMI (S7ONLINE) — YES — YES
  • Output consistency (ST logic) — YES — YES
  • Preliminary network captures — 189 KB / 1375 packets — N/A

1.1. Network capture analysis (excerpts from .pcapng files)

General statistics

Table of general metrics on the content of network captures

  • Parameter · Value
  • File size — 193 532 octets (189 KB)
  • Format — pcapng v1.0
  • Captured Packets — ~1375
  • LLDP Frames — 215
  • Estimated capture duration — ~27 minutes

Detected industrial protocols

Discovered network protocols following the wireshark capture

  • Protocol · Occurrences · Comment
  • S7comm (port 102) — 0 — This absence of S7comm frames remains to be investigated further, as it was specified in the expected outcomes, being necessary for any PLC-HMI communication.
  • OPC UA (port 4840) — 0 — As the PLC was not connected to any other equipment, this behaviour is completely normal.
  • LLDP (discovery) — 215 — The discovery of Siemens Profinet network equipment is indeed active.

Network evidence – excerpts from network capture logs

Captured LLDP frames

Of the 215 LLDP frames captured, here is an exhaustive list of the identified equipment:

Table of LLDP frames captured during phase 0

  • MAC Source · Chassis ID (LLDP)
  • 02:1b:1b:6d:f0:07 — desktop-or2einm (Attaquant)
  • 02:c0:a8:0a:03:01 — plcxb1.profinet-schnittstellexb15ee1
  • 02:c0:a8:0a:03:03 — plcxb1.profinet-schnittstellexb25fa1
  • 5e:6e:59:cc:1c:9d — desktop-tpp6ej3 (Engineering)

Sources of the LLDP frames

The hexadecimal/ASCII analysis of the LLDP captures clearly reveals the identity of the equipment present on the network: on this first frame we can distinguish the identification of our S7-1500 PLC emulator

Offset  Hex                                              ASCII
─────────────────────────────────────────────────────────────────────
0000   31 06 02 00 14 08 29 53 69 65 6d 65 6e 73 2c 20   |1.....)Siemens, |
0010   53 49 4d 41 54 49 43 20 53 37 2c 20 45 74 68 65   |SIMATIC S7, Ethe|
0020   72 6e 65 74 20 50 6f 72 74 2c 20 58 31 20 50 31   |rnet Port, X1 P1|
0030   0c 71 53 69 65 6d 65 6e 73 2c 20 53 49 4d 41 54   |.qSiemens, SIMAT|
0040   49 43 20 53 37 2c 20 43 50 55 31 35 30 30 20 28   |IC S7, CPU1500 (|
0050   50 4c 43 53 49 4d 29 2c 20 36 45 53 37 20 53 49   |PLCSIM), 6ES7 SI|
0060   4d 2d 30 31 35 30 30 2d 41 50 4c 43 2c 20 48 57   |M-01500-APLC, HW|
0070   3a 20 56 65 72 73 69 6f 6e 20 30 2c 20 46 57 3a   |: Version 0, FW:|

We find on this frame the identity of the HMI station

Offset  Hex                                              ASCII
─────────────────────────────────────────────────────────────────────
0000   6f 72 74 2d 30 30 31 06 02 00 14 0a 0f 44 45 53   |ort-001......DES|
0010   4b 54 4f 50 2d 54 50 50 36 45 4a 33 0c 15 53 49   |KTOP-TPP6EJ3..SI|
0020   45 4d 45 4e 53 20 41 47 20 53 49 4d 41 54 49 43   |EMENS AG SIMATIC|
0030   20 50 43 0e 04 00 80 00 80 10 14 05 01 00 00 00   | PC.............|


On the following one, the identity of the attacking machine

Offset  Hex                                              ASCII
─────────────────────────────────────────────────────────────────────
0000   69 6e 6d 04 09 07 70 6f 72 74 2d 30 30 31 06 02   |inm...port-001..|
0010   00 14 0a 0f 44 45 53 4b 54 4f 50 2d 4f 52 32 45   |....DESKTOP-OR2E|
0020   49 4e 4d 0c 60 56 4d 77 61 72 65 2c 20 49 6e 63   |INM.`VMware, Inc|
0030   2e 20 4e 6f 6e 65 2c 56 4d 77 61 72 65 2d 34 32   |. None,VMware-42|


And finally the presentation of the Profinet network functionality of the emulated PLC as well as its firmware version

Offset  Hex                                              ASCII
─────────────────────────────────────────────────────────────────────
0000   63 78 62 31 2e 70 72 6f 66 69 6e 65 74 2d 73 63   |cxb1.profinet-sc|
0010   68 6e 69 74 74 73 74 65 6c 6c 65 78 62 31 35 65   |hnittstellexb15e|
0020   65 31 04 09 07 70 6f 72 74 2d 30 30 31 06 02 00   |e1...port-001...|
0030   14 08 29 53 69 65 6d 65 6e 73 2c 20               |..)Siemens,     |
Hex:   46 57 3a 20 56 65 72 73 69 6f 6e 20 53 33 30 2e 38 30 2e 35
ASCII: FW: Version S30.80.5, 10S C-16

Conclusion on this first experimental phase.

As expected, the behaviour of the simulated solution was perfectly compliant with the specifications. However, in-depth analysis reveals an inconsistency between the theoretical and fundamental communication between a PLC and its user interface, which in our case was not formally perceptible through network captures.
The continuation of experiments is crucial to resolve the inconsistency thus raised.

Situation report as of 21/01/2026

Following the course of the experimental phase, further experiments were planned to resolve doubts about the results from the network captures.

The hardware situation was not in my favour, as I lost 3 months of previous work in a fraction of a second due to the malfunction of my main work hard drive. The initial report being stored on it, I first had to redo all the work of setting up a virtualisation infrastructure, then redeploy the digital twin as well as write this research report that you are currently reading.

I thus plan to continue my work as part of another experimental phase potentially scheduled for early March if conditions permit.

In symbolic memory of this unexpected event that required me to work 16 hours a day from 22/01/2026 to 03/02/2026 in order to make up for the delay caused by this loss of approximately 20 TB of data, here is a tribute to the object of all my woes at the end of January 2026!

Photo of "The Void" at its peak
Photo of "The Void" at its peak

Ainsi voici ci-dessus une photo d'un des responsables des chimères intellectuelles que je dus combattre durant deux semaines entraînées par sa perte, afin de rattraper 3 mois de travaux en quelques jours.

General conclusion and answers to the research questions

This research work was a true success in terms of answering the questions raised regarding the architectural aspect of an attack virtualisation solution.

The possibilities that virtualisation brings in the context of industrial systems security are significant when compared to costly and inflexible physical solutions.

The deployed architecture thus demonstrates that it is possible to faithfully reproduce levels 0 to 3 of the established Purdue model while adhering to a controlled budget. The ability to segment networks and carry out attacks in complete isolation from the real world aligns well with the approach presented in the IEC 62443 standards recommendations [34].

The discrepancies between the physical system and its digital twin are minimal, which indicates a high degree of validity of the established model. However, the absence of S7Comm frames remains a mystery to be elucidated during the next phases of experimentation. Was it a problem related to the early signs of an error on my work disk causing data corruption and resulting in frame capture latency? Only one thing is clear: the experiments must continue.

The fidelity of the Siemens proprietary emulator PLCSIM Advanced remains to be established, for while it respects its macroscopic communication behaviour with its user interface, its internal behaviour still needs to be validated, particularly its vulnerability to OPC UA attacks as well as its expected faithful reproduction of behaviour when facing a MitM attack. The answers to these questions may in the future validate the use of the emulator for training or research purposes in the field.

Regarding viable open-source alternatives, we cannot draw conclusions as, due to lack of time and practical research elaboration on the subject per se, the question remains open. Avenues exist but remain entirely to be explored. Among those that came to mind throughout the experiments, one will note the possibility of replacing the workload imposed by the Siemens proprietary solution, which requires deploying the solution on an unnecessary operating system layer. A solution like OpenPLC appears to be the best candidate given its functional comprehensiveness and its open-source nature in terms of legal usage frameworks.

Finally, regarding the simulation platform itself, the Cyber Range, I would like to point out that despite its unparalleled features, its rapid deployment of virtual solutions, its interoperability and its usage capabilities, its semi-proprietary nature largely due to its foundation built on a VMware hypervisor strongly invites us to consider designing an entirely open source solution that would meet the major challenges of work frameworks imposed by companies and industry standards.

The design and implementation of such a project would, however, require increased human resources.

This leads us to discuss the human relationships established throughout this research work.

Being both a challenge I set for myself and a desire to work on a specific project in the field of industrial systems, my feedback is mixed.

I am very satisfied with the progress on the project but at the same time saddened that I was unable to share this project with more people. This comment is not merely an expressed sentiment; it reflects the necessity of working collaboratively on projects of this scale, thus providing greater peace of mind and security when considering the setbacks experienced on January 21, 2026. The presence of other research members would have ensured data integrity security, as it would have de facto required the various members to possess a local copy of the progress to work on the project, which in my case was compromised by the imposed centralisation of work resources.

I take nothing but positives from these final days of experimental phase 0, and this allows me to properly redefine a work framework conducive to my interests and desires while ensuring stability and security of results.

Perspectives

When it comes to the perspectives and future directions of the project, one phrase comes to mind.

"Everything will depend on the resources, motivations and desires of the many participants towards the project!"

Bibliography

  1. Dragos. 2025 OT Cybersecurity Year in Review. s.l. : Dragos Inc., 2025.
  2. IBM. Cost of a Data Breach Report 2024. 2024.
  3. VMware Vulnerabilities CVE-2025-22224, CVE-2025-22225, CVE-2025-22226: Critical ESXi Escape Chain. Kovacs, E. May 2025.
  4. Markets and Markets. Cyber Range Market Size, Share & Trends Analysis Report. s.l. : Markets and Markets Research, 2024.
  5. Baglione, Pierre-Sacha. Rapport de Stage ESAIP: Forces et faiblesses de l'Airbus CyberRange. s.l. : ESAIP, 2025.
  6. Airbus CyberSecurity. CyberRange Documentation. CyberRange Dcumentation v2.10. s.l. : Airbus Defence and Space, 2024, 2024.
  7. Toward Next-Generation Cyber Range: A Comparative Study of Training Platforms. SAMOVAR, Télécom SudParis, Institut Polytechnique de Paris & Department of Information and Communications Engineering, University of Murcia. Vol. Computer Security. ESORICS 2023 International Workshops.

  8. NICE Framework. NIST. s.l. : NIST Special Publication, 2020.
  9. Understanding Cyber Ranges: From Hype to Reality. European Cyber Security Organisation. 2020.
  10. A Review of Cyber-Ranges and Test-Beds: Current and Future Trends. Ukwandu, E., et al. 24, s.l. : Sensors, 2020, Vol. 20.
  11. Markets and Markets. Cyber Range Market Size, Share & Trends Analysis Report. s.l. : Markets and Markets Research, 2024.
  12. Cyber Ranges and Security Testbeds: Scenarios, Functions, Tools and Architecture. Yamin, M.M., et al. s.l. : Computers & Security, 2020, Vol. 88. 101636.
  13. Vykopal, J., et al. KYPO Cyber Range: Design and Use Cases. s.l. : ICISSP, 2017.
  14. AIT Cyber Range: Flexible Cyber Security Environment for Exercises, Training and Research. Landauer, M., et al. s.l. : EuroS&P Workshops, IEEE, 2021.
  15. Cyber-Range Federation and CyberSecurity Games: A Gamification Scoring Model. Diakoumakos, I., et al. s.l. : ECHO FCR, Zenodo, 2021.
  16. A Survey of Industrial Control System Testbeds. Holm, H., et al. s.l. : NordSec, Springer, 2015.
  17. AG, Siemens. SIMATIC S7-1500/ET 200MP System Manual. s.l. : Siemens Industry Online Support, 2024.
  18. —. PLCSIM Advanced V7.0 Function Manual. s.l. : Siemens Industry Online Support, 2024.
  19. SCADA Vulnerabilities and Attacks: A Review of the State-of-the-Art and Open Issues. Alanazi, M., et al. 103028, s.l. : Computers & Security, 2023, Vol. 125.
  20. MITRE. ATT&CK for Industrial Control Systems. s.l. : MITRE Corporation, 2024.
  21. OpenPLC: An Open Source Alternative to Automation. Alves, T.R., et al. s.l. : IEEE Global Humanitarian Technology Conference, 2014.
  22. W32.Stuxnet Dossier. Falliere, N., et al. s.l. : Symantec Security Response, 2011.
  23. Stuxnet Facts Report. Falco, LTC Marco De. Tallinn : CCDCOE, 2012.
  24. Analysis of the Cyber Attack on the Ukrainian Power Grid. Lee, R.M., et al. s.l. : SANS ICS, 2016.
  25. CRASHOVERRIDE: Reassessing the 2016 Ukraine Electric Power Event as a Protection-Focused Attack. Slowik, J. Montreal - Canada : Dragos Inc, VB2018, 2018.
  26. TRITON: The First SIS Cyber Attack. Networks, Nozomi. s.l. : Nozomi Networks Labs, 2018.
  27. Analysis of the Colonial Pipeline Cybersecurity Incident. Taiwo Justice Olorunlana1, Hamdiya Mohammed. s.l. : International Journal of Science, Architecture, Technology, and Environment, 2025, Vol. 2.
  28. Industroyer2: Sandworm's Cyberwarfare Targets Ukraine's Power Grid Again. Cherepanov, A., and Lipovsky, R. s.l. : ESET Research, 2022.
  29. Industroyer vs. Industroyer2: Evolution of the IEC 104 Component. Nozomi Networks - Giannis Tsaraias & Ivan Speziale. s.l. : Nozomi Networks, 2022.
  30. Team82, Claroty. From Exploits to Forensics: Unraveling the Unitronics Attack. Claroty. [En ligne] 7 Août 2024.
  31. FrostyGoop: Analysis of a Novel ICS Malware. Intelligence, Dragos Threat. s.l. : Dragos Threat Intelligence, 2024.
  32. The Purdue Enterprise Reference Architecture. Williams, T.J. s.l. : Computers in Industry, 1994, Vol. 24.
  33. Security for Industrial Automation and Control Systems. ISA/IEC. s.l. : International Society of Automation=, 2020.
  34. THE CYBER RANGE - A GUIDE. The Cyber Range Project Team and the NICE Community. s.l. : NIST, 2023.
  35. ICSSIM: A Framework for Building Industrial Control Systems Security Testbeds. Dehlaghi-Ghadim, A., et al. 103906, s.l. : Computers in Industry, 2023, Vol. 148.
  36. Informationstechnik, Bundesamt für Sicherheit in der. OPC UA Security Analysis. s.l. : Bundesamt für Sicherheit in der Informationstechnik, 2017 - 2022.
  37. Security Analysis of Vendor Implementations of the OPC UA Protocol for Industrial Control Systems. Erba et al., 2022 : Erba, A., Müller, A., & Tippenhauer. s.l. : CISPA, 2022.
  38. Temnikov, Cheremushkin &. OPC UA Security Analysis. s.l. : Kaspersky Lab ICS CERT, 2018.
  39. OPC UA Deep Dive: A Complete Guide to the OPC UA Attack Surface. Moshe, Brizinov &. s.l. : Claroty Research, 2023.
  40. Components, SPS-Service | RS. sps-service.eu | fr.rs-online.com. [En ligne]