Digital-Twin-Enabled Process Monitoring for a Robotic Additive Manufacturing Cell Using Wire-Based Laser Metal Deposition

Abstract

Digital Twins (DTs) are transforming manufacturing by bridging the physical and digital worlds, enabling real-time insights, predictive analytics, and enhanced decision making. In Industry 4.0, DTs facilitate automation and data integration, while Industry 5.0 emphasizes human-centric, resilient, and sustainable production. However, implementing DTs in robotic metal additive manufacturing (AM) remains challenging because of the complexity of the wire-based laser metal deposition (LMD) process, the need for real-time monitoring, and the demand for advanced defect detection to ensure high-quality prints. This work proposes a structured DT architecture for a robotic wire-based LMD cell, following a standard framework. Three DT implementations were developed. First, a real-time 3D simulation in RoboDK, integrated with a 2D Node-RED dashboard, enabled motion validation and live process monitoring via MQTT (message queuing telemetry transport) telemetry, minimizing toolpath errors and collisions. Second, an Industrial IoT-based system using KUKA iiQoT (Industrial Internet of Things Quality of Things) facilitated predictive maintenance by analyzing motor loads, joint temperatures, and energy consumption, allowing early anomaly detection and reducing unplanned downtime. Third, the Meltio dashboard provided real-time insights into the laser temperature, wire tension, and deposition accuracy, ensuring adaptive control based on live telemetry. Additionally, a prescriptive analytics layer leveraging historical data in FireStore was integrated to optimize the process performance, enabling data-driven decision making.

1. Introduction

Industry 4.0 represents a paradigm shift in manufacturing, driven by the integration of advanced production systems and next-generation information technologies (new ITs) such as cyber–physical systems (CPSs), the Internet of Things (IoT), cloud computing (CC), artificial intelligence (AI), and big-data analytics (BDA) [1]. It envisions the evolution of traditional manufacturing to a new generation of smart factories, characterized by enhanced intelligence, flexibility, reconfigurability, and sustainability [2]. These new manufacturing systems operate at a cyber–physical level of automation, where digital and physical components interact dynamically to optimize production processes, resource efficiency, and adaptability to changing demands [3]. Now, a decade after its inception, the transition toward Industry 5.0 is emerging, shifting the focus from purely technology-driven advancements to a value-centered approach that prioritizes human-centric, sustainable, and resilient manufacturing systems [4].

At the core of this cyber–physical integration lies the digital twin (DT) concept, a virtual replica of physical systems that mirrors real-world operations, enabling the advanced simulation, predictive analysis, and optimization of manufacturing processes in real time [5]. Ciano et al. [6] emphasize that DTs are a prerequisite for cyber–physical production systems, positioning it as a core element of Industry 4.0. Today, DTs are transforming manufacturing, optimizing product lifecycle management, and enhancing predictive maintenance strategies [7]. Leveraging AI-based algorithms, DT enhances decision making by enabling real-time anomaly detection, self-adaptive process control, and predictive maintenance in manufacturing environments. This intelligent integration not only optimizes manufacturing efficiency but also fosters innovation, allowing industries to develop smarter products, streamline operations, and create new value-driven business models [5,8].

On the other hand, additive manufacturing (AM) is a pivotal technology within Industry 4.0, enabling the production of complex geometries and customized components while minimizing material waste [9]. As defined by the ISO 52900 standard [10], AM is the “process of joining materials to create parts from 3D model data, typically layer by layer, as opposed to subtractive or formative manufacturing methods”. This standard further classifies AM techniques into seven primary categories including binder jetting (BJT), directed energy deposition (DED), material extrusion (MEX), material jetting (MJT), powder bed fusion (PBF), sheet lamination (SHL), and vat photopolymerization (VPP).

Figure 1a provides a comprehensive overview of AM technologies, with particular emphasis on the DED category. Within this classification, laser metal deposition (LMD) processes are specifically highlighted in blue boxes in the figure. Among these, wire-based LMD emerges as one of the most advanced and widely utilized metal 3D printing techniques in the industry [11,12]. Its superior material efficiency, high deposition rates, and capability to fabricate complex, large-scale components make it a preferred choice for critical applications in the aerospace, automotive, and heavy-manufacturing sectors.

Figure 1b illustrates a wire-based LMD system, highlighting its key components and operational principles. The system comprises a high-power laser source, which delivers energy through an optical fiber to the deposition head. The deposition head features a nozzle that precisely directs the laser beam onto the substrate, generating a localized melt pool. Simultaneously, a metallic wire is fed into the melt pool, where it is fused to the substrate, enabling controlled material buildup. To ensure process stability and high-quality deposition, the system includes an inert gas supply, which shields the melt pool from oxidation, and a coolant circuit that circulates water to regulate the temperature of the laser optics and nozzle. This integrated approach allows for the precise, layer-by-layer fabrication of metallic structures, making wire-based LMD an essential technology for AM in high-performance industries.

By integrating LMD with industrial robotics, it is possible to produce diverse components within a single manufacturing cell, significantly reducing setup times between production cycles and enhancing the overall operational efficiency [13]. This integration provides distinct advantages for various industrial applications, optimizing production capabilities and meeting specific technical requirements [14]. However, ensuring consistent quality in LMD processes presents significant challenges because of variations in the laser power, deposition rates, material behaviors, and thermal dynamics. Real-time monitoring and predictive analysis are essential to detect potential defects, optimize process parameters, and enhance the overall system reliability. DTs provide a powerful solution by creating virtual representations of physical systems that can simulate, predict, and respond to operational changes in real time. By integrating DTs into LMD–wire processes, manufacturers can achieve greater precision, reduce material waste, and improve the overall production efficiency, reinforcing the principles of Industry 4.0 and Industry 5.0.

Despite the growing body of research on DTs and their roles in AM metal processes, a critical gap persists in the seamless integration of DTs with ISO-23247-based frameworks [15,16,17,18,19] for real-time process monitoring, predictive maintenance, and defect detection in robotic-wire-based LMD. Existing approaches often lack a structured methodology to unify these capabilities in a standardized architecture, limiting their scalability and industrial applicability. This study directly addresses this research void by proposing a comprehensive DT framework that not only consolidates disparate monitoring programs but also ensures robust synchronization between digital and physical components. By bridging this gap, the proposed approach enables advanced simulation, predictive maintenance, and real-time process optimization, ultimately enhancing the reliability and efficiency of robotic LMD systems.

In this context, the current work details the computational architecture and real-time implementation of a DT for an AM cell utilizing wire-based LMD with a KUKA KR70 R2100 robot and a Meltio engine robot’s integration LMD head. The proposed DT adheres to the ISO 23247 standard, which defines a structured approach for DTs in manufacturing, emphasizing automation and system integration. The framework incorporates CAD/CAPP/CAM (computer-aided design, computer-aided process planning, and computer-aided manufacturing) for design, planning, and execution, along with RoboDK for operational monitoring and Node-RED with FireStore Cloud for data processing and storage. Furthermore, this work presents DT implementations: KUKA iiQoT (Industrial Internet of Things Quality of Things) and the Meltio dashboard, both utilized to enhance process monitoring and defect detection and provide valuable insights into the LMD–wire process. Their integration not only improves operational efficiency and quality control but also strengthens system resilience, a key aspect of Industry 5.0, by enabling adaptive responses to changing demands and technological advancements. In addition, these implementations incorporate the 3D digital model with the physical setup of the KUKA robot and Meltio tool. This real-time synchronization between digital and physical systems optimizes operational efficiency and enhances the precision of the LMD process.

The remainder of this paper is structured as follows: Section 2 presents a literature review on the DT concept, exploring its applications in metal AM for process monitoring and defect detection. Section 3 examines reference models from the literature for DT development in manufacturing, with a particular focus on the ISO 23247 framework as a structured approach for digital representations of industrial processes. Section 4 provides an overview of the proposed DT architecture for the robotized wire-based LMD cell, detailing its design based on ISO 23247 and describing the three DT implementations developed in this work. Section 5 presents a prescriptive analysis based on historical process data stored in FireStore, demonstrating how data-driven insights enhance decision making and system optimization. Section 6 highlights the key contributions of this research, emphasizing the advancements in real-time simulations, predictive maintenance, and interactive monitoring. Finally, conclusions and future research directions are discussed in Section 7.

2. Literature Background

This section explores the research background of this study, reviewing the concept of DTs, reference models for DTs in manufacturing, and their applications in AM.

2.1. The Concept of DTs

The concept of DTs stands as one of the most significant technological advances of recent decades. Its origins can be traced to 2003, when Dr. Michael Grieves first introduced the concept during his product lifecycle management (PLM) course at the University of Michigan [20]. His proposal laid the foundation for a groundbreaking DT model, emphasizing the integration of physical products (real space), virtual counterparts (virtual space), and the seamless exchange of interconnected data between them. Since then, advances in the IoT, big-data analytics, cloud computing, and artificial intelligence have evolved the original concept until it reached a compelling level of maturity.

In 2012, NASA introduced a comprehensive definition of the concept of DTs: “…is an integrated multi-physics, multi-scale, probabilistic simulation of an as-built vehicle or system that uses the best available physics models, sensor updates, fleet history, etc. to mirror the life of the corresponding flying twin” [21]. In the manufacturing context, Tao et al. [22] view DTs as the bridge between the physical and digital realms for smart manufacturing. Meanwhile, the ISO 23247 standard [15] defines a manufacturing DT as a fit-for-purpose digital representation (digital entity) of an observable manufacturing element (physical entity) that are synchronized with each other. Subsequently, a substantial number of definitions of DTs have been formulated depending on their application domain or their relationship to similar terms, such as digital thread, product avatar, surrogate models, and real-time simulations, as reported in [23].

Tao et al. [24] have thoroughly examined the relationship and distinctions between CPS and DTs, analyzing key aspects, such as their origins and evolution, physical-to-digital mapping, hierarchical structures, and functional implementations. Their in-depth comparison positions CPS as a scientific paradigm, closely associated with Industry 4.0, while DTs are characterized as an engineering-driven concept. Despite their distinct identities, both frameworks converge on the fundamental principle of cyber–physical integration, leveraging shared technologies to achieve this synergy. Together, they drive the advancement of smart manufacturing, fostering a dynamic interplay between science and engineering to enable groundbreaking innovations.

According to Kritzinger et al. [25], DTs can be classified in three categories based on the levels of data integration and autonomy in information exchange: the digital model, which represents a physical system but lacks automated data exchange, relying solely on manual updates; the digital shadow, where real-time data flow automatically from the physical to the digital counterpart, but updates in the opposite direction require operator intervention; and the digital twin, which enables full bidirectional data integration, allowing real-time interaction and optimizations between both states. Using this classification scheme, Chen et al. [26] introduced a complementary classification based on the autonomy of data exchange. At the lowest level, a digital model requires an operator to mediate all the interactions. In a digital shadow, while data flow automatically from the physical to the virtual system, updates in the reverse direction still depend on human intervention. Finally, an autonomous DT achieves fully automated, bidirectional communication, where the physical and digital counterparts continuously interact without operator involvement, though human oversight remains possible for advanced decision making. These classifications provide a structured understanding of DT evolution, from static digital representations to fully autonomous, self-adaptive systems.

Figure 2 illustrates the three levels of data integration in a metal AM process, such as wire-based LMD, focusing on process monitoring and defect detection. In the digital model, monitoring and defect detection rely entirely on offline analysis. The process is simulated using preset parameters, such as the wire feed rate and laser power, without real-time feedback. Defects, like porosity or irregular layer formation, are identified post processing through manual inspection or the separate analysis of fabricated parts, requiring iterative adjustments by the operator.

The digital shadow introduces a real-time, one-way data flow from the physical process to its digital counterpart. Sensors continuously track key parameters, such as the melt pool temperature, head load (force exerted on the deposition head), and laser stability. Anomalies, such as overheating or layer inconsistencies, are detected automatically, allowing for early-stage analysis and predictive defect identification. However, corrections still depend on operator intervention, as the digital model cannot directly modify the physical process.

In the DT, full bidirectional data exchange enables real-time process adaptation. Advanced image-based analysis, powered by pretrained machine-learning models, continuously monitors the deposition quality, detecting macro-defects, such as “balling” and “strubbing”, in real time. If anomalies are identified or thermal imaging detects excessive heat accumulation, the DT autonomously adjusts the laser power, wire feed rate, or printing speed to correct inconsistencies and maintain the optimal build quality. This AI-driven approach minimizes material waste, enhances part reliability, and reduces the need for human intervention in defect mitigation.

2.2. Research on DT Applications in AM

An exploration of the literature using the Scopus database, with a search incorporating key terms, such as “additive manufacturing”, “3D printing”, and “rapid prototyping”, combined with “digital twin”, revealed a substantial body of 626 documents. This reflects the increasing significance of DTs in transforming manufacturing within the framework of Industry 4.0. As shown in Figure 3, the annual publication trend has demonstrated exponential growth since 2016, with research output reaching 215 papers by 2024, more than five times the number in 2020. Notably, as of February 2025, an additional 30 publications have already been recorded. This surge underscores the global recognition of DTs as a crucial technology for real-time process monitoring, predictive maintenance, and seamless integration in AM manufacturing processes. The recent spike aligns with the widespread adoption of data-driven and digital-first approaches in advanced manufacturing, further cementing DTs as a foundational pillar for smarter AM systems.

The development DTs in AM has become a focal point of contemporary research. Professor Debroy’s team has pioneered the development of this subject by conducting a series of research studies aimed at reducing or replacing the time-consuming and expensive trial-and-error experiments commonly used to predict the process parameters that affect the structure and mechanical properties of printed metal parts [27,28]. On the basis of these studies, Mukherjee and DebRoy [29] applied and tested the building blocks of the DT proposed in previous work to efficiently estimate cooling rates, solidification parameters, secondary-dendrite-arm spacing, velocity distributions, and micro-hardness in powder-bed fusion and direct-energy deposition processes. They argue that the components of a DT for AM should include mechanistic and statistical models, machine learning, big-data analytics, and sensing and control [29,30].

Several studies [31,32,33] have conducted comprehensive reviews on the roles of DTs in AM, emphasizing their increasing relevance in enhancing process efficiency and quality [31] to highlight ability of DTs to minimize AM’s reliance on trial-and-error through real-time monitoring and control while stressing the need for further advancements in AI and data integration. Another study [32] emphasizes the capacity of DTs to stabilize AM processes by leveraging the IoT and machine learning for defect mitigation and performance optimization. Bem Amor et al. [33] emphasize that DT development in AM is an iterative process requiring close collaboration among domain experts, data scientists, engineers, and manufacturing teams to ensure an accurate and evolving digital representation of the process. These studies reinforce the transformative potential of DTs in driving AM forward within Industry 4.0.

This work dedicates particular attention to reviewing research on the application of DTs in metal AM, with emphases on process monitoring and defect detection. Given the critical roles of real-time data acquisition, analysis, and feedback in ensuring process stability and part quality, the following sections explore key contributions in these areas.

DTs for Process Monitoring and Defect Detection in Metal AM

DTs play crucial roles in process monitoring and defect detection in metal AM, ensuring build quality, stability, and repeatability. By continuously tracking parameters, such as the melt pool behavior, deposition height, thermal variations, and material flow, through real-time sensor data, imaging, and computational models, DTs enable adaptive control strategies to correct deviations before they lead to defects. Defect detection, particularly for macro-defects, like lack of fusion, balling, and warping, is essential for maintaining part integrity and process efficiency. By leveraging machine learning, computer vision, and in situ sensing, DT-driven systems provide real-time insights that enhance predictive modeling and closed-loop control, minimizing material waste and production delays. This section explores key advancements in DT-enabled monitoring, focusing on data integration and early fault identification techniques.

Several studies have demonstrated the application of DTs in metal AM processes, such as WAAM to enhance process monitoring and predictive control. Reisch et al. [34] developed a DT framework for WAAM that enables real-time data analysis and proactive, context-aware process adjustments. Their approach integrates system component reviews, a robot-based setup, and experimental validation of welding torch orientation effects, highlighting the potential of DTs to improve precision and efficiency in WAAM-based hybrid manufacturing. Similarly, Mu et al. [35] introduced a DT framework incorporating an adaptive online simulation model for real-time distortion prediction in WAAM. To overcome the computational limitations of traditional numerical simulations, they employed a hybrid AI approach, combining a vector-quantized variational autoencoder and a generative adversarial network (VQVAE-GAN) for spatial feature extraction alongside a recurrent neural network (RNN) for temporal fusion. Their model, pretrained with finite element method (FEM) data, significantly outperforms conventional FEMs and artificial neural networks (ANNs) in predicting distortion fields using laser-scanned point clouds.

For PBF processes, DTs have been explored to optimize process parameters and mitigate defects. Malik, Mahmood, and Liou [36] developed a DT architecture by integrating finite element modeling with machine-learning techniques—specifically, recurrent neural networks and reinforcement learning—to predict and prevent lack-of-fusion defects. Their system adjusts process parameters in real time by leveraging sensor data and physics-based simulations, reducing reliance on costly trial-and-error methods and improving the process efficiency. In a related study, Klingaa et al. [37] analyzed the influences of build chamber conditions, including the gas flow speed, pressure, and oxygen content, on part quality in PBF. Their response surface models enabled in-line assessments for real-time feedback control, optimizing process variables and enhancing decision making under uncertainty in metal AM.

Real-time monitoring and predictive control strategies have also been developed for laser-based PBF. Yavari et al. [38] employed a DT integrating in situ melt pool temperature measurements with a graph-based thermal simulation model. By continuously updating predictions layer by layer, their system enables the early detection of process anomalies, such as parameter drifts, machine malfunctions, and cyber intrusions. Validated on stainless-steel (316L) impellers, their approach significantly improves process stability and defect prevention. In a broader framework, Phua, Davies, and Delaney [39] proposed a hierarchical DT structure for metal AM, emphasizing its roles in part qualification, certification, and process optimization. Their model identifies four levels of DT complexity, integrating surrogate modeling, in situ sensing, hardware control, and intelligent policies, to move away from costly physical testing. Expanding on this, Phua et al. [40] introduced a smart recoating framework combining DT technology with Bayesian optimization to dynamically control powder spreading in PBF. Unlike conventional fixed-action recoating, their method adapts recoater and platform movements in real time, optimizing the layer quality using discrete-element-method simulations and setting the foundation for an intelligent DT capable of autonomous control and process optimization.

In DED processes, multisensor fusion and multiscale modeling have enhanced DT-based defect detection and monitoring capabilities. Chen et al. [41] introduced a multisensor-fusion-based DT framework for defect detection in robotic laser-based DED additive manufacturing. By integrating data from a coaxial melt pool’s vision camera, an acoustic microphone, and an infrared thermal camera, the study developed a spatiotemporal fusion method to synchronize multisensor features with real-time robotic motions. Supervised machine-learning models trained on labeled defect data achieved a high accuracy rate (96%) and an ROC-AUC of 99%, enabling the generation of a virtual quality map that closely matched optical microscope observations. This approach eliminates the need for destructive testing and facilitates automated defect removal via robotic machining. On the other hand, Hartmann et al. [42] developed a multiscale DT for laser-based DED processes, integrating global and local models to balance accuracy and computational efficiency. The global model simulates the overall heat distribution, while the local model focuses on high-resolution laser–powder interactions and rapid cooling rates. Validated through in situ monitoring on an industrial laser-based DED machine, the DT achieved high accuracy rates in predicting clad dimensions and temperatures, with errors below 5% and 7%, respectively.

Beyond conventional AM monitoring, DTs have also been leveraged for advanced defect detection using non-destructive testing (NDT) techniques. Zhang et al. [43] proposed a DT-based method for laser-based ultrasonic metal defect detection, addressing data scarcity through finite element modeling and generative adversarial networks (GANs) to enhance the data quality. They applied a continuous wavelet transform to convert ultrasonic signals to time–frequency images, analyzed using a ResNet50-RA model optimized for real-time synchronization with physical detection equipment. Experimental validation demonstrated high detection accuracy rates and strong generalization across different defect scenarios.

Finally, a cloud-based approach has been explored to enhance data integration across different lifecycle stages of AM systems. Liu et al. [44] proposed a DT-enabled collaborative data management system in which a cloud DT communicates with distributed edge DTs. Their use case for layer defect analysis, implemented within the Manuela Project platform using a deep-learning model, demonstrated significant potential in reducing development times, lowering costs, and improving product quality and production efficiency.

These studies collectively illustrate the transformative impacts of DTs on AM process monitoring and defect detection, demonstrating their potential to enhance predictive modeling, adaptive control, and in situ optimization across various AM technologies.

Table 1. Summary of DT-based monitoring and defect detection in metal AM.

Reference	AM Process	Techniques Used	Purpose of the Study
[34]	WAAM	Real-time data analysis and Context-aware process adjustments	DT framework for WAAM, enabling proactive monitoring and improved precision.
[35]	WAAM	VQVAE-GAN, RNN, FEM, and Laser-scanned point clouds	Real-time distortion prediction through hybrid AI modeling.
[36]	PBF	FEM, RNN, Reinforcement learning, and Sensor data	Predicting and preventing lack-of-fusion defects with DT-based parameter control.
[37]	PBF	Response surface models and In-line assessments	Optimization of build chamber conditions for improved process control.
[38]	Laser-based PBF	In situ thermal monitoring and Graph-based simulation	Early detection of process anomalies and defect prevention in AM.
[39]	Metal AM	Hierarchical DT structure and Surrogate modeling	DT framework for part qualification, certification, and process optimization.
[40]	PBF	Bayesian optimization and DEM simulations	Smart recoating framework for adaptive powder-spreading control.
[41]	Laser-based DED	Multisensor fusion, Machine learning, and Acoustics	DT-based defect detection with high-accuracy virtual quality mapping.
[42]	Laser-based DED	Global/local DT models and In situ monitoring	Multiscale DTs for balancing computational efficiency and accuracy.
[43]	Laser Ultrasonics	FEM, GANs, ResNet50-RA, and Wavelet transform	DT-based NDT for high-accuracy metal defect detection.
[44]	Metal AM	Cloud DT, Edge DTs, and Deep learning	Collaborative DTs for lifecycle-wide AM data integration and defect analysis.

provides a structured summary of the reviewed DT-based approaches in metal AM, outlining the specific techniques, technologies, and objectives of each study. This comparative overview offers valuable insights into the diverse methodologies employed for real-time monitoring, defect detection, and process optimization, underscoring the critical roles of DTs in advancing manufacturing precision, efficiency, and reliability.

3. Reference Models of DTs for Manufacturing

Because of the pressing need to describe how the interaction between physical assets and their digital counterparts should proceed within the complex dynamic environment of Industry 4.0, several efforts have been made to consolidate a reference model and an implementation architecture of DTs in industrial production systems. The purpose of a DT reference model is to guide manufacturers and practitioners in arranging and preparing the technologies, procedures, standards, and best practices that can be used to implement DT technology in manufacturing applications.

3.1. Three-Dimensional Models

The three-part DT model, as introduced by Grieves [20], is depicted in Figure 4a. This model comprises a physical product existing in real space and its synchronized virtual representation, interconnected through a bidirectional communication channel. Similarly, Lu et al. [5] have proposed a three-component DT model tailored for the realization of smart manufacturing (refer to Figure 4b): The first component is an information model mapping physical assets and creating abstractions; the second involves a data-processing module generating representations of physical assets based on extracted information and knowledge; and, finally, a two-way communication mechanism facilitating interaction between physical and digital spaces.

3.2. Five-Dimensional Model

Tao et al. [45] introduced a comprehensive DT model grounded in five dimensions: a physical part, a virtual part, services, data, and connections. Their emphasis on the equal importance of each dimension underscores a holistic approach for crafting impactful DTs. The inception of the physical part catalyzes the creation of its virtual counterpart, unlocking a realm of possibilities for simulation, decision making, and control services. This interconnected ecosystem thrives on the wealth of knowledge derived from data transmitted through the intricate connection mechanism. As the digital and physical seamlessly converge in this model, it not only reflects innovation but also paves the way for a future where DTs redefine the landscape of SM (smart manufacturing) with unparalleled effectiveness.

3.3. Digital Twin Framework for Machining

STEP Tools, Inc., Troy, NY, USA [46] developed an interoperable DT framework for machining, showcased in demonstration meetings in 2016 with active participation from Boeing, OMAC, and the ISO TC184/SC4 Committee. This groundbreaking framework (Figure 5) facilitates closed-loop machining by integrating real-time inspection data, adhering to digital thread standards, such as the STEP-NC (Standard for the Exchange of Product model data–Numerical Control), MTConnect (manufacturing technology connect), and QIF (quality information framework). By leveraging the STEP-neutral file format, STEP AP242 (ISO 10303-242) and STEP-NC AP238 (ISO 10303-242) are employed to encapsulate data related to design, process planning, and inspection. MTConnect contributes by making machine status data accessible over the internet through the REST API, catering to consumption by web clients. Furthermore, QIF serves as the communication channel for relaying inspection results back to the design domain.

3.4. Other Related Proposals

Several notable contributions have been made to the field of manufacturing DT models. A notable study by Alam and El Saddik [47] introduces an architecture reference model of DTs for cloud-based CPSs (cyber–physical systems), named C2PS. This model incorporates a smart interaction controller grounded in a fusion of fuzzy logic rules and Bayesian networks. On the other hand, Aheleroff et al. [48] proposed an architecture reference model for DTs as a service (DTaaS) within the context of Industry 4.0. They have successfully implemented this reference model in an industrial case, demonstrating its potential benefits in scheduled maintenance, real-time monitoring, remote control, predictive functionalities, and mass individualization. Further insights into DT models are explored in related proposals, as documented in references [49,50,51,52]. These diverse approaches contribute to the evolving landscape of DT frameworks in manufacturing, each offering unique perspectives and applications.

3.5. ISO 23247—DT Reference Framework for Manufacturing

The ISO 23247 standard [15], introduced in 2021, represents a significant milestone in the standardization efforts of the ISO Committee TC 184/SC 4. The primary aim of this standard is to provide a comprehensive DT architecture framework tailored for industrial manufacturing applications within the context of Industry 4.0. It offers invaluable guidance in the construction of DTs for manufacturing, emphasizing interoperability among systems and the seamless integration of data from diverse sources. The standard comprises four integral parts, as outlined below. These parts offer guidance on analyzing modeling requirements, setting scope and objectives, using common terminology, specifying a generic reference architecture to instantiate DTs, and supporting information modeling and synchronization between the DTs and physical system [53].

• Part 1—Overview and general principles [15]: Provide an overview of the general principles of DTs, as well as definitions, requirements, and development guidance;
• Part 2—Reference architecture [16]: Provides an architecture reference model for a manufacturing DT framework;
• Part 3—Digital representation of manufacturing elements [17]: Helps to identify the physical elements that need to be mapped to the digital model;
• Part 4—Information exchange [18]: Establishes the requirements for the proper synchronization and exchange of data throughout the DT framework;
• Part 5 (Currently in a draft version under development)—Digital threads for digital twins [19]: Describes how digital threads facilitate the creation, connectivity, management, and maintenance of manufacturing DTs throughout the product lifecycle by outlining principles, demonstrating methodologies, and providing case studies.

Figure 6 illustrates the DT framework derived from ISO 23247, Part 2 [16], delineated into four distinct domains.

The first layer pertains to the observable manufacturing element (OME) domain, encompassing devices, sensors, machines, materials, products, processes, and facilities requiring monitoring and control.

The second layer encapsulates the data collection and device control entities (DCDCEs), entrusted with overseeing sensor data and managing actuated devices within the OME domain. Additionally, this domain facilitates synchronization between OME entities and digital twin entities.

The DT domain (third layer) comprises diverse system entities providing access to manufacturing services, like simulation, provisioning, management, monitoring, analysis, and optimization. This domain serves as the nucleus of the DT framework for manufacturing, emphasizing the interaction and integration capabilities of DT entities for system interoperability.

Lastly, the fourth layer corresponds to the DT user domain, housing entities seeking to leverage DT services. These entities could range from individuals and companies to other systems.

Furthermore, a cross-system entity may span across domains, delivering essential functionalities, such as information exchange, data assurance, and security support. This strategic placement ensures seamless collaboration and interoperability across the entire DT framework.

3.6. Research on the ISO 23247 Framework

The ISO 23247 framework is starting to gain attention within the research community, marking a shift from its initially underexplored status. In particular, NIST has used this framework as a template for implementing DTs in its SMS Testbed Project, unveiling three distinct use cases: machine status and condition monitoring, production scheduling and routing, and virtual commissioning [54].

Several studies have further explored the potential of ISO 23247, assessing its adaptability across different sectors. The study by Shao, Frechette, and Srinivasan [55] aimed to both inform the manufacturing community and assess the applicability of the ISO 23247 standard in emerging sectors, like biomanufacturing and AM. Using a bottom-up approach, they examined key concepts, provided interpretations, and highlighted how the framework enables further development and industrial adoption. Their work clarifies the scope and potential of the standard, guiding its implementation in novel industrial manufacturing applications.

ISO 23247 has also served as a foundation for DT implementations in AM. Kim, Shao, and Jo [56] pioneered the application of the framework to wire + arc additive manufacturing (WAAM), tackling integration and interoperability challenges while incorporating machine-learning-based anomaly detection for real-time decision making. Expanding on this foundation, Kang et al. [57] introduced an edge-computing-based DT framework to enhance data-processing capabilities. By leveraging ISO 23247 as a reference architecture and incorporating a data fusion model, their approach improves real-time interactions between physical and digital systems. A comparative evaluation between the proposed framework and conventional DTs for WAAM demonstrated significant reductions in latency, optimized data transmission, and enhanced decision-making efficiency. Together, these works establish a progressive evolution in DT architectures, advancing WAAM through enhanced real-time processing and predictive analytics.

Cabral, Rodriguez, and Alvares [58] presented a DT implementation based on the ISO 23247 framework, focusing on real-time monitoring and 3D simulation in a CNC Haas MiniMill. By leveraging MQTT and MTConnect protocols, the system efficiently transmits machining process and machine status data, enabling both edge monitoring via Node-Red and cloud-based 3D simulation using React.js, Three.js, and IBM Watson. This integrated approach demonstrates the feasibility of applying DT solutions across various machines and manufacturing environments. In extending this approach to metal AM, Cabral, Alvares, and Caribé [59] developed a DT architecture for a robotic cell consisting of a robotic arm, a positioning worktable, and a welding machine. Their system allowed real-time process monitoring, cloud-based data storage, and both online and post-deposition analysis for detecting quality-affecting process instabilities.

Other studies have leveraged ISO 23247 to enhance DT capabilities in flexible and modular production environments. Wallner et al. [60] explored the implementation of DTs in flexible manufacturing cells, leveraging ISO standards, including ISO 21597, ITU-T Y.3090, and ISO 23247. The goal was to enhance the efficiency of the DT application setup and maintenance by identifying key features, parameters, and assets. The focus encompassed manufacturing elements, such as machine tools, robots, and peripheral devices, alongside information systems, like CAM, PLM (product lifecycle management), and the MES (manufacturing execution system). To streamline lifecycle changes, the paper proposes linking to a lifecycle meta-layer, simplifying the design, deployment, and updates of DT applications and, ultimately, reducing maintenance efforts. Similarly, Le et al. [61] presented a DT architecture for modular production systems (MPSs) using the ISO 23247 framework. Their approach integrates web services, databases, and production elements to enable real-time interaction between physical and digital environments. Validated in real manufacturing setups, this system enhances monitoring, optimization, and control across industrial processes.

The adoption of ISO 23247 has also been extended to specialized manufacturing applications. Melo et al. [62] developed a DT architecture compliant with ISO 23247 and aligned with RAMI 4.0 to optimize dimensional measurements in automotive assembly lines. Their implementation enhances real-time defect detection at body shop inspection stations, preventing defects at early stages and avoiding their propagation to downstream processes, thereby ultimately improving manufacturing precision and efficiency. In another study, Caiza and Sanz [63] developed an ISO 23247-compliant architecture for immersive DTs in flexible manufacturing, integrating augmented reality and gesture tracking. Validated at an MPS 500 sorting station, the system enhanced real-time interactions, monitoring, and simulations. Their methodology covered functional models, manufacturing attributes, and communication protocols. The results showed improved process efficiency and faster error resolution, reducing production delays.

In the domain of robotic automation, Minh et al. [64] developed a real-time machine-health-monitoring DT framework for robotic machine-tending applications based on ISO 23247. Their system integrates OPC UA for communication and ROS for data exchange with a physical setup that includes a UR10e cobot, an On-Robot RG2 gripper, and vision sensors. The DT architecture consists of three key entities: a data collection and control unit (DCDCE), a core entity (CE) for edge computing and AI-driven decision making, and a user entity (UE) for cloud–human interactions. Their evaluations, based on metrics like the RMSE (root-mean-square error) and R² score, confirmed the accuracy of the DTs in replicating cobot movements, enhancing process repeatability, reproducibility, and interoperability in machine-tending applications.

Other studies have explored the roles of ISO 23247 in predictive maintenance and industrial sustainability. Sobowale et al. [65] employed DT technology within the ISO 23247 framework to monitor aging-crane degradation, focusing on the load capacity and structural stability. Using Ansys Twin Builder, their DT model enhanced the predictive maintenance and risk assessment, improving safety and extending cranes’ lifespans. Meanwhile, Cederbladh, Ferko, and Lundin [66] examined advancements in battery systems driven by sustainability and electrification. They highlighted emerging paradigms, like reconfigurable and software-defined batteries, which enhance runtime adaptability and efficiency, optimizing performance across industries such as automotive, railway, and aerospace. Additionally, Shtofenmakher and Shao [67] adapted the ISO 23247 standard from manufacturing to the aerospace sector. The focus on collision avoidance in low-Earth orbit (LEO) demonstrated the feasibility of applying existing DT standards to aerospace challenges, providing a standardized framework for this crucial process. The study extends its scope to the space-based detection of sub-ten-centimeter-class orbital debris, highlighting the framework’s versatility for diverse aerospace applications.

Building upon the foundation of ISO 23247, Rodriguez, Alvares, and Riaño [68] proposed a cyber–physical architecture for STEP-NC in manufacturing, integrating the digital thread and DT technologies into a robust digital ecosystem through a standard-driven approach. At its core, this architecture relies on STEP and STEP-NC, supported by key open standards, such as QIF, MTConnect, OPC-UA, and MQTT. By incorporating contextual information, the system enhances DT capabilities, enabling real-time virtual monitoring, process optimization, intelligent knowledge generation, and data-driven decision making. Although initially designed for advanced manufacturing, its modular and scalable nature makes it applicable across various industrial processes, reinforcing the vision of ISO 23247 for interoperable and adaptive DT solutions.

Ongoing efforts to refine the ISO 23247 framework face challenges in achieving consistent implementation, highlighting the need for standardized practices to enhance DT as a valuable tool for optimizing production processes and operations. In this context,

Table 2. Comparative analysis of DT implementations based on ISO 23247.

Reference	AM Process	Technologies Used	Main Contribution
[55]	Biomanufacturing, AM	ISO 23247 and Bottom-up modeling	Clarifies the applicability of ISO 23247 in new industries.
[56]	WAAM (Wire + Arc AM)	Machine learning and Anomaly detection	DTs for real-time decision-making in WAAM.
[57]	WAAM (Wire + Arc AM)	Edge computing and Data fusion	Reduces latency and optimizes data flow in DTs.
[58]	CNC machining	MQTT, MTConnect, and React.js	DTs for real-time monitoring and 3D simulations.
[59]	Metal AM robotic cell	Cloud-based monitoring	Online and post-process quality analyses.
[60]	Flexible manufacturing	ISO 21597 and ITU-T Y.3090	Lifecycle meta-layer for DT maintenance efficiency.
[62]	Automotive assembly	RAMI 4.0 and Defect detection	DTs to prevent defect propagation during production.
[63]	Flexible manufacturing	Augmented reality and Gesture tracking	Enhances interactions and real-time monitoring.
[64]	Machine tending (robotics)	OPC UA, ROS, UR10e cobot	AI-based monitoring for process repeatability.
[61]	Modular production	Web services and Databases	The real-time optimization of industrial processes.
[65]	Aging cranes	Ansys Twin Builder	Predictive maintenance and risk assessment.
[66]	Battery systems	Reconfigurable, software-defined batteries	Enhance runtime adaptability and efficiency.
[67]	Aerospace	ISO 23247 adaptation	DTs for space debris tracking and avoidance.
[68]	AM processes	ISO 10303, STEP, STEP-NC, QIF, MTConnect, MQTT, and OPC-UA	STEP-NC cyber–physical architecture
This work	Wire-based LMD	ISO 23247, Kuka iiQoT v8.7 (Virtual Box v.7 (Ubuntu 24.04 LTS, OPC UA, and MQTT)), MQTT Mosquitto v2.0.18, Node-Red v4.0, RoboDK v5.7.0, Meltio Dashboard v2, FireStore Cloud, Faster R-CNN, YOLOv5s, FCN, TensorFlow v2.10, PyTorch v2.5.0+cu118, cuDNN v8.1, CUDA v11.2 and the Python 3.9 library	Real-time process monitoring, 3D simulation, and defect detection

summarizes the main characteristics of the most relevant works previously mentioned, including our proposed approach, providing a structured perspective on the evolution of ISO 23247-based DT implementations.

4. Overview of the ISO-23247-Based Framework for the Robotic AM Cell

The research design of this study is structured around three main components: (i) the development of a digital twin architecture based on ISO 23247, (ii) the integration of multiple DT implementations for real-time monitoring and predictive maintenance, and (iii) the experimental validation of the proposed system in a robotic wire-based LMD cell.

Three DTs compliant with ISO 23247 were developed. The first DT is a real-time 3D simulation DT using RoboDK, developed by integrating CAD/CAM tools for process planning and collision detection. The second is an Industrial IoT-driven DT based on KUKA iiQoT, implemented to capture and analyze real-time telemetry data, thereby facilitating predictive maintenance through machine-learning algorithms. The third DT is the Meltio dashboard, employed for the remote monitoring and visualization of process parameters. Communication between the physical cell and the DTs is maintained via MQTT and OPC-UA protocols, while data storage and historical analysis are managed through FireStore Cloud.

Figure 7 depicts an overview of the proposed ISO-23247-based DT framework for the robotic AM cell using wire-based LMD. This architecture is composed of interconnected domains that facilitate data acquisition, simulation, optimization, and decision making, ultimately contributing to enhancing the efficiency and intelligence of the system. This framework integrates the five application domains—DThE (digital thread entity), OME (observable manufacturing element), DCDCE (data collection and device control entity), DTE (digital twin entity), and DTUE (digital twin user entity), synchronized with each other, following the guidelines set forth in ISO 23247-5. The synchronization among these layers ensures seamless operation, enabling 2D monitoring via a dashboard and real-time 3D simulation of the LMD additive manufacturing cell. Additionally, the architecture incorporates KUKA iiQoT for predictive maintenance and Meltio dashboard for the visualization of printing data.

The OME domain contains the physical elements of the wire-based LMD robotic cell. The primary component of this setup is a KUKA KR70 2100 industrial robot, equipped with a Meltio laser deposition head. These components are complemented by their respective controllers, a wire-feeder system, a chiller for temperature regulation, and argon gas cylinders for shielding.

A fundamental aspect of any AM process is the digital thread, which defines the sequence of operations from the initial design to the final fabrication. This aspect is captured within the digital thread entity (DThE) domain, where multiple software tools orchestrate the workflow. The design phase is carried out using Rhino or another CAD tool, while the slicing and toolpath generation processes are executed with Grashopper, Meltio Space, and KUKASim. The final toolpath is converted to KUKA robot language (KRL) codes, ensuring seamless execution in the robotic system. This digital workflow establishes continuity between design, manufacturing, and execution, forming a fundamental component of the DT system.

For a DT to function effectively, it must rely on robust data collection and exchange mechanisms. The DCDCE domain plays crucial roles in acquiring, processing, and transmitting information from the physical system to the digital domain. Various communication protocols, including MQTT, OPC UA, and CAN Bus/USB, are employed to gather real-time process data. The architecture also incorporates dedicated servers, which run both virtual machines and custom-built applications, to process and route collected data, ensuring high-fidelity synchronization between the physical and digital environments.

Once the data are collected, they are transmitted to the digital twin entity (DTE) domain, where the core DT applications reside. Here, real-time sensor data are integrated with historical records, enabling advanced simulation, monitoring, and diagnostics. Simulation tools, such as RoboDK, facilitate real-time 3D visualization, while KUKA iiQoT supports predictive maintenance. Additionally, tools like Meltio dashboard and Node-RED provide process-variable monitoring. Computer vision techniques are also leveraged for defect identification, ensuring that manufacturing anomalies are detected early in the process. The integration of these elements creates an intelligent DT ecosystem capable of predicting potential failures and dynamically adjusting process parameters.

Finally, the digital twin user entity (DTUE) domain acts as the interface through which users can access and interact with the DT system. The applications in this domain expose DT services via REST APIs (representational state transfer application programming interfaces) based on the HTTP (hypertext transfer protocol) or WebSockets, enabling users to visualize real-time data, control manufacturing parameters, and leverage AI-driven analytics for decision making. Operators can remotely monitor process variables, analyze system performance, and implement closed-loop control strategies to optimize production quality. Additionally, FireStore Cloud is utilized to provide scalable data storage and synchronization, ensuring persistent access to process logs and DT insights.

Figure 8 illustrates the functional modeling based on IDEF0 (integration definition for function modeling) for the three digital twins implemented in the robotic additive manufacturing cell, demonstrating the deployed functionalities [69]. It is important to note that the KUKA DT (iiQoT dashboard) and Meltio DT (Meltio dashboard) are proprietary solutions from KUKA and Meltio, respectively, and require licensing for their use. Therefore, their implementation details will not be further elaborated. The detailed focus will be on the development of the solutions referred to as DT RoboDK, Node-RED, MQTT, and Firebase for the additive manufacturing cell.

Although Figure 7 and Figure 8 depict individual DT modules grouped under the ISO 23247 framework, the overall DT architecture ensures full unidirectional synchronization between the physical system and its digital counterpart. This synchronization is essential for real-time process monitoring and offline analysis for future corrective actions. Offline corrective actions are based on stored process data in databases, as well as recorded process images.

However, in the proposed architecture and its current implementation, the real-time optimization of the LMD–wire process is not feasible. This limitation arises from the absence of key sensory data, such as images of the molten pool formed between the wire and the deposited material or the substrate, which could be captured using a dedicated melt-pool-monitoring camera. By applying AI algorithms to analyze these images, it would be possible to classify material transfer mechanisms and dynamically adjust the process by varying the laser power, wire-feed rate, and/or robot speed. At present, no data flow exists to enable real-time process optimization from the digital to the physical domain because of the lack of sensors capable of effectively monitoring the process and the material transfer mechanisms from the wire to the substrate/layers.

Process optimizations, defect elimination, and other feedback actions are performed offline, after the part is printed, by analyzing stored sensor data and recorded images from cameras used for process monitoring.

The IDEF0 diagram, Figure 8, illustrates the integration of three interconnected digital twins (DTs) within a real-time additive manufacturing cell using the laser metal deposition (LMD) process. These digital twins, specifically, the real-time additive manufacturing cell LMD (digital twin—RoboDK, Node-Red, MQTT, and Firebase), the KUKA iiQoT, and the Meltio dashboard, collectively enable advanced monitoring, control, and data analysis of the manufacturing workflow. Each digital twin plays a distinct yet complementary role in this architecture, contributing to the seamless flow of data and ensuring the precise coordination of the system.

The digital twin—RoboDK, Node-RED, MQTT, and Firebase (Activity A1) is the central element of the system. It processes critical inputs, such as joint position, joint speed, and motor temperature data from the robotic system, as well as additional information from the KUKA controller and the Meltio engine. This digital twin provides outputs such as real-time 3D simulations through platforms like RoboDK and RoboDK Web, cloud-based data storage in FireStore Cloud, and a 2D data visualization interface via Node-RED. Adhering to the ISO 23247 digital twin framework, it employs MQTT and OPC-UA protocols to ensure reliable and interoperable communication with other components.

The KUKA iiQoT digital twin (Activity A2) focuses on integrating data from KUKA robotic systems into an Industrial Internet of Things (IIoT) platform. It communicates unidirectionally with the real-time additive manufacturing cell digital twin, exchanging data through a shared interface to enable real-time monitoring and predictive maintenance. It provides visualizations via a dedicated 2D data panel in the KUKA iiQoT system and leverages cloud connectivity through Microsoft Azure and the KUKA Connectivity Box, ensuring scalable and secure data integration.

The Meltio dashboard digital twin (Activity A3) facilitates the monitoring and control of the Meltio additive manufacturing system. It receives operational parameters from the Meltio engine system and Meltio server, transmitting these data to a dedicated Meltio dashboard for real-time process supervision. This digital twin is also integrated with a Linux-based virtual machine and the Meltio cloud, allowing advanced data-processing and storage capabilities. It communicates with the other digital twins via a dedicated interface, aligning its functionalities with those of the overall manufacturing system.

This integrated framework showcases the synergistic potential of combining multiple digital twins to achieve real-time monitoring, optimization, and control in additive manufacturing. By enabling precise coordination between robotic systems, cloud platforms, and operator interfaces, the system enhances the process reliability, minimizes errors, and supports advanced analytics, representing a significant advancement in smart manufacturing technologies.

This multi-domain DT architecture, based on ISO 23247, establishes a seamless digital representation of the wire-based LMD process, ensuring real-time adaptability, intelligent automation, and enhanced process efficiency. By integrating real-world data with predictive modeling and digital analytics, the proposed DT framework facilitates a fully interconnected and intelligent manufacturing ecosystem. The following sections provide a detailed breakdown of each domain within the proposed DT architecture, highlighting interactions and roles in the integration of physical and digital processes in the robotic AM cell.

4.1. Setup of the Robotic LMD AM Cell

The setup of the robotic LMD AM cell, depicted in Figure 9, is centered around a KUKA KR70 R2100 robotic arm, which provides six degrees of freedom and high positional accuracy, allowing for complex deposition paths and adaptive manufacturing strategies. Mounted on the robotic arm is the Meltio engine robot’s integration LMD head, which serves as the primary processing tool for wire-based LMD. This head is equipped with a 1.2 kW laser system, composed of six fiber-coupled diode lasers, each delivering 200 W of power at a wavelength of 976 nm, ensuring uniform energy distribution across the melt pool.

The robot and the Meltio head are seamlessly integrated, with precise synchronization between motion control and laser power modulation. The robot’s controller, a KUKA KR C5, communicates with the Meltio engine control unit, allowing for real-time adjustments to processing parameters. The LMD process relies on a dual wire-feeding system: an external welding wire feeder ensures a continuous supply of 316L stainless-steel wire, while the Meltio head incorporates an internal traction mechanism for the fine control of the wire deposition.

To maintain stable processing conditions, the system includes a water-cooled laser chiller, which regulates the temperature of the laser source, preventing overheating and ensuring consistent performance. Additionally, an argon-shielding-gas system, fed from high-pressure cylinders, protects the molten material from oxidation, creating the optimal environment for layer-by-layer metal deposition. The gas flow is carefully regulated to maintain uniform coverage over the melt pool.

The substrate for material deposition is securely mounted on a heavy-duty, precision-leveled steel worktable, designed to withstand high thermal and mechanical stresses during processing. This table also serves as a base for real-time monitoring systems, including a thermal camera for temperature tracking and a high-speed video camera for process observation and defect detection. These sensors provide crucial feedback for process control and quality assessment.

Ensuring operator safety is a paramount consideration in the design of the AM cell. The entire setup is enclosed within a protective safety cabin, featuring an interlocked security system. The interlock mechanism immediately stops the process and prevents laser activation if the enclosure door is opened or if any safety violations are detected. This system ensures compliance with industrial safety standards, mitigating risks associated with high-power laser operations.

4.2. Description of the DThE and DCDCE Domains

This implementation leverages advanced CAD/CAPP/CAM software, real-time data acquisition, and predictive analytics to enhance the efficiency, precision, and adaptability of manufacturing processes. As a fundamental component of the DT framework, the DThE encompasses CAD, CAPP, and CAM systems essential for the design and preparation of robotic AM. This layer utilizes Rhino3D v.7 for geometric modeling, Grasshopper v7 for parametric design, and Meltio Space v26.04.24 for G-code generation and slicing. The process begins with the preparation of the part program of a test part called “Gravia”, where deposition parameters, such as the layer height, material type, infill percentage, and part geometry, are defined. Additionally, a 3D simulation of the robotic cell is integrated to visualize the printing process, ensuring collision avoidance and optimal path planning, as depicted in Figure 10. Once the slicing process is complete, the system generates KRL code, which dictates the robotic motion and extrusion commands. This code is transmitted to the OME domain, where it controls the robotic system’s execution in real time. The interaction between the DThE and OME is facilitated through TCP/IP sockets, OPC DA communication, and digital I/O signals, ensuring synchronized operations.

The DCDCE plays crucial roles in acquiring and processing data from the robotic system. This layer utilizes the KUKA connectivity box type A or a dedicated Linux-based virtual machine to collect real-time parameters, such as joint positions, speeds, and temperatures. Data from the Meltio deposition process, including the wire feed rate, laser power, and sensor feedback, are also acquired through CAN Bus and USB protocols. These data are transmitted to the DTE via MQTT, ensuring a low-latency, high-efficiency communication channel. The integration of the OPC UA further enhances interoperability, allowing seamless data flow between heterogeneous systems. Additionally, data from the process are stored on an Intel i7 industrial computer, housed within the Meltio controller, ensuring reliable logging for historical analysis and predictive maintenance applications.

Figure 11 illustrates the data acquisition flow of the developed system (first digital twin: RoboDK, Node-RED, MQTT, and Firebase), where MQTT is used as the communication protocol. The data from the robot controller are published to a broker running on a local server, with different types of data assigned to specific topics. Various applications are subscribed to these topics, including the real-time simulation environment in RoboDK and the process monitoring dashboards built with Node-RED, ensuring seamless data distribution and system synchronization.

Table 3. Robot data, from the robotic LMD cell, collected from the KUKA controller via MQTT.

Variable	Collection Method	Description	MQTT Topic
Status	The KUKA KRC5 system sends a “TRUE” message upon initialization and “FALSE” upon termination to indicate the connection status.	Indicates whether the adapter is online with the KUKA KRC5 controller.	“Status”
Joint Position	The KUKA system retrieves the “ActualPosition” variable for each joint.	Represents the angular position of the six joints (from 1 to 6) in degrees.	“Joints_Position”
Joint Speed	The KUKA system retrieves the “ActualSpeed” variable for each joint.	Indicates the angular speed of each joint in degrees per second.	“Joints_Speed”
Joint Motor Temperature	Accesses the “MotorTemperature” variable for each joint’s motor.	Displays the motor temperature for each joint in degrees Kelvin.	“Motor_Temperature”
Timestamp	Created in JavaScript for data transmission references.	Provides an instantaneous timestamp for the current data cycle.	“Timestamp”

presents the key variables acquired from the KUKA KRC5 system via MQTT and TCP/IP sockets, outlining their collection methods, descriptions, and corresponding MQTT topics. The collected data include the robot’s operational status, joint positions, speeds, motor temperatures, and a timestamp to ensure accurate references during data transmission.

4.3. Implementation of DTs in the Robotic LMD Cell

The DT architectures for the robotic LMD cell have been developed through three distinct implementations, each focusing on a critical aspect of the process. These DTs enable advanced simulations, predictive maintenance, and real-time monitoring, creating a fully integrated and intelligent manufacturing system.

The first implementation is a real-time 3D simulation and process validation system using RoboDK, which is presented in Figure 12. This DT provides a highly detailed and kinematically accurate representation of the robotic system, enabling the preemptive validation of manufacturing trajectories before execution. Through direct integration with CAD/CAM workflows, the toolpaths for the wire-based LMD process are automatically generated and optimized to ensure deposition accuracy. Additionally, RoboDK performs collision detection and workspace analysis, allowing engineers to identify and mitigate potential hazards before physically executing the process. The simulation continuously updates based on live data from the physical system, ensuring that trajectory modifications and process adjustments are dynamically reflected. This unidirectional communication between the virtual and real environments improves operational efficiency and reduces errors, making it a fundamental component in ensuring high-precision additive manufacturing.

Associated with the framework of the first implementation, there is also a real-time monitoring and control system developed using Node-RED. In Node-RED, an MQTT input node was configured to subscribe to the relevant topics corresponding to the robot’s operational parameters, such as position, speed, and process temperature, as illustrated in Figure 13.

These incoming messages were then processed using function nodes to parse and format the data appropriately. A database node was incorporated to store historical records in Firebase, ensuring traceability and enabling later analysis. For real-time visualization, a dashboard UI was developed in Node-RED, using gauge, chart, and text nodes to display live telemetry data from the system. Each dashboard element was linked to an MQTT node that dynamically updated as new data arrived.

The second DT implementation focuses on Industrial IoT (IIoT)-driven monitoring and predictive maintenance through KUKA iiQoT [70,71], as presented in Figure 14. This DT enables comprehensive real-time diagnostics and long-term performance analytics of the robotic system. By continuously streaming telemetry data from the robot controller, including motor loads, joint temperatures, and energy consumption, the system can proactively detect anomalies and inefficiencies. Machine-learning algorithms analyze historical performance trends to predict potential failures before they occur, enabling a predictive maintenance strategy that minimizes unplanned downtime and extends the equipment’s lifespan. Additionally, when the system identifies abnormal behavior, automated alerts and recommended corrective actions are sent to operators, allowing for proactive intervention. The cloud-based nature of KUKA iiQoT ensures that data are not only accessible remotely but also used for long-term trend analysis, contributing to continuous process optimization and operational resilience.

The third DT implementation is associated with the Meltio dashboard [72], which enables the remote monitoring of the Meltio controller’s display/terminal via TCP/IP sockets, as shown in Figure 15. It provides real-time video of the printed part, temperature data of the diode lasers, load cell force data, and wire-feeding tension data, among other parameters, all presented graphically. Further details on its usage and features can be found in the videos available at [70,71].

Together, these three DT implementations form a comprehensive digital ecosystem that enhances the efficiency, adaptability, and intelligence of the robotic LMD cell. By integrating real-time simulation, predictive maintenance, and interactive monitoring, this architecture not only aligns with Industry 4.0 principles but also establishes a robust foundation for the future of data-driven, intelligent manufacturing systems.

It is important to note that the system operates under soft real-time constraints, which are sufficient for the needs of robotic additive manufacturing. Although the adopted protocols (MQTT and OPC-UA) and hardware are not designed for hard real-time applications, they provide reliable and timely communication to ensure effective process-monitoring and control.

5. Prescriptive Analysis of Data Stored in Firebase and Meltio Engine

Figure 16 presents two superimposed graphs depicting the time-series monitoring of the robot’s joint positions and motor temperatures. During the printing of the “Gravia” part, a higher heating degree of the base motor (38 °C) is observed, while the motor of joint 3 exhibits a comparatively lower temperature (32 °C). This variation in the temperature suggests differential thermal loads across the robotic joints, which could impact the longevity and efficiency of the motors over prolonged operational periods.

Figure 17 illustrates two superimposed graphs showing the time-series monitoring of the robot’s joint positions and motor speeds. The deposition strategies rely on linear movement commands (KRL LIN) for material deposition, while approach and retraction maneuvers are executed using point-to-point commands (KRL PTP). The speed variations across different joints are attributed to the serial kinematics of the robot, requiring the synchronized acceleration of all the motors. Notably, joint 4 exhibits the highest speeds and the most significant variations in positioning, which could be indicative of its critical role in maintaining the deposition precision.

Figure 18 correlates the robot joint speed parameters with motor temperatures, highlighting that the highest temperatures are recorded in the base motor (38 °C), despite its relatively low kinematic speeds. Conversely, the highest speeds are observed in joint 4, which maintains an average temperature of 34 °C. These findings reinforce the need for targeted thermal management strategies, especially for the base motor, which endures the highest thermal strain during the printing process.

These analyses provide valuable insights into the relationship between robot joint kinematics and motor heating, underscoring the importance of predictive maintenance strategies. Because the base motor experiences the highest working temperature and endures significant strain during the 3D printing of complex parts, such as the “Gravia” part, preventive monitoring and cooling strategies should be considered to enhance the system reliability.

This analysis extends to data generated by the Meltio dashboard, which integrates a Meltio engine robot’s integration head equipped with a load cell to monitor stress variations during printing. This load cell detects excessive stress when the wire fails to fuse properly to a layer or bead of the printed part, leading to wire entanglement. Additionally, it functions as a safety mechanism to prevent damage in cases where the laser fails or does not activate, causing the wire to continue feeding without fusing. When the load surpasses approximately 60 kgf, the system triggers an emergency stop (trip), interlocking both the KUKA robot and the Meltio head to prevent further damage.

Figure 19 presents a graph illustrating the stress experienced by the Meltio head over time, as measured by the load cell. Ideally, the stress should remain minimal, appearing as a nearly linear trend with minor oscillations. The peaks in the graph indicate wire fusion and deposition issues, which can lead to defects in the printed part.

These defects can be categorized based on their root cause as follows [73]:

• Balling Defects: Occur when the material melts before making contact with the build plate, forming a bright, spherical bead in mid-air because of the incorrect laser alignment;
• Dripping Defects: Arise when the wire momentarily loses contact with the build plate, leading to incomplete fusion. Minor cases may be tolerable, but pronounced dripping can disrupt subsequent layers;
• Necking Defects: Manifest as a thinning melt pool before detaching from the wire. This indicates insufficient material, leading to weak inter-layer adhesion and wavy surface defects;
• Overbuilding Defects: The opposite of necking, where excessive material deposition results in a gradual increase in the layer height beyond the specified parameters, disrupting the uniformity;
• Stubbing Defects: Occur when insufficient energy reaches the wire, preventing proper melting and leading to excessive force on the Meltio head. This can result in severe process failure and potential machine damage (Figure 20).

Balling defects are primarily caused by incorrect laser alignments, where the focal point is set too high, leading to the premature melting of the wire. Dripping, similarly, results from an energy imbalance, where excessive energy is directed to the wire rather than the substrate. Adjustments to the tool center point (TCP) and deposition feed speed can mitigate these issues.

Necking defects stem from underbuilding, where the layer height is shorter than intended, causing cumulative discrepancies across successive layers. This increases the distance between the nozzle and the part, shifting the energy balance away from the substrate. Increasing the deposition feed speed and refining the laser alignment can help to resolve this problem.

Overbuilding defects occur when excessive material is fed into the melt pool, necessitating a slight decrease in the deposition feed speed. Stubbing, on the other hand, is often linked to misaligned lasers, emphasizing the need for precise calibration.

By analyzing the recorded images [74,75] of the printed part and correlating them with load cell readings, the peaks in the graph (Figure 19) coincide with instances of unfused wire adhering to the printed part, confirming the occurrence of stubbing (Figure 20). Addressing this issue requires meticulous laser alignment to ensure an even energy distribution between the substrate and the wire.

Furthermore, reviewing timestamped videos of the printing process [70,74,75] provides additional validation for the observed defects. Corrective measures, including the recalibration of the collimator’s positioning within the Meltio engine robot’s integration optical components, are crucial for mitigating these issues and ensuring consistent, defect-free deposition.

6. Contributions of This Work

This study introduces three distinct DT implementations, iiQoT Kuka, Meltio dashboard, and RoboDK/Node-RED/MQTT/Firebase—each addressing specific aspects of the AM process. This ISO-23247-based approach enhances real-time monitoring, process control, and optimization as follows:

• Standardized Framework: The DT architecture aligns with ISO 23247, ensuring structured integration between digital and physical systems. This adherence to standards facilitates scalability and interoperability in industrial environments;
• Interdisciplinary Approach: By integrating concepts from mechanical engineering, robotics, and computer science, this work establishes a solid foundation for future research and technological advancements in AM;
• Validation in a Real-World Environment: The proposed architecture has been implemented and tested in an operational setting, demonstrating its feasibility and potential for broader adoption in advanced manufacturing;
• Enhanced Decision Making with Real-Time Data: The DT system enables dynamic simulations, real-time monitoring, and predictive analytics, improving the process adaptability and reducing production inefficiencies;
• Predictive Maintenance Capabilities: The integration of KUKA iiQoT with the Meltio dashboard facilitates continuous monitoring and data-informed maintenance strategies, contributing to failure reduction and improved equipment lifespans;
• Advanced Technology Integration: This research leverages CAD/CAPP/CAM, RoboDK, Node-RED, and FireStore Cloud for seamless process monitoring, data management, and quality control, ensuring a fully integrated manufacturing ecosystem;
• Contributions to Industry 4.0 and 5.0: This work demonstrates how digitalization, automation, and cyber–physical integration enhance efficiency, flexibility, and customization, aligning with the principles of Industry 4.0 while also considering the human-centric, sustainable, and resilient focus of Industry 5.0 for the next generation of manufacturing systems.

Figure 21 and Figure 22 illustrate the key advancements in defect detection and quality assurance. Figure 21 shows the use of YOLO (you only look once) and faster region-based convolutional neural network (faster R-CNN) models for predicting and classifying macro-defects, such as balling and strubbing, in LMD–wire-printed parts. The dataset construction is currently in progress [76]. Figure 22 presents the preliminary implementation of a fully convolutional neural (FCN) model for microstructural defect segmentation in microscopy images, supporting enhanced quality assessment in the wire-based LMD process.

7. Conclusions

This research has successfully developed a robust and modular DT architecture tailored for robotic LMD AM, fully aligned with the ISO 23247 standard. By integrating real-time data acquisition, advanced analytics, and cloud connectivity, the proposed DT framework lays the foundation for a highly adaptive and intelligent cyber–physical system capable of enhancing manufacturing efficiency, predictive maintenance, and process optimization.

The architecture is composed of three interconnected DT implementations: iiQoT Kuka, Meltio dashboard, and RoboDK/Node-RED/MQTT/Firebase. Although the first two are proprietary solutions, the third represents a novel, open-source contribution [77] that leverages MQTT for high-speed, reliable communication. This implementation not only adheres to the DT domains outlined in ISO 23247 but also demonstrates the potential of open-source tools in digital manufacturing ecosystems.

The integration of a structured DT architecture based on ISO 23247 has proven to be effective in enhancing soft real-time process monitoring, predictive maintenance, and adaptive control in robotic wire-based LMD. The experimental results support the assertion that combining multiple DT implementations—each optimized for specific process aspects—provides a more comprehensive solution than a singular approach. Future work will focus on further refining the predictive algorithms and exploring additional real-time communication protocols to meet the requirements of hard real-time systems, if necessary.

Two key strengths of this approach are its flexibility and scalability. The seamless integration of MQTT, Node-RED, and Firebase ensures real-time monitoring, data-driven decision making, and rapid responses to unexpected process deviations. Furthermore, the availability of the developed open resource fosters further advancements in DT applications for metal AM.

Looking ahead, future work will focus on extending the DT’s visualization capabilities through virtual reality (VR) and augmented reality (AR), enabling immersive monitoring and control. Additionally, the cloud-based deployment of the MQTT Node-RED dashboard will facilitate remote supervision and analytics.

The next phase of this research will explore the integration of AI for enhanced predictive analytics and autonomous decision making. Machine-learning models will be applied to diagnose and correct defects in the LMD process, leading to a more resilient and intelligent manufacturing system. By leveraging deep-learning techniques for defect classification and segmentation, the DT will evolve into a self-optimizing system capable of improving product quality and energy efficiency.

Finally, further experimental validation with real metallic components will provide deeper insights into the AM cell’s performance. This will not only refine the process parameters but also establish a data-driven methodology for optimizing robotic AM operations. The advancements outlined in this research mark a significant step toward the convergence of AI, digital twin technology, and smart manufacturing, paving the way for future industrial breakthroughs.