A Reference Architecture for Digital Transformation of SMEs in the Manufacturing Domain

A Reference Architecture for Digital Transformation of SMEs in the Manufacturing Domain Konstantinos Tsitseklis1, Lukas Morand2, Miguel A. Mateo-Casali7, Georgia Stavropoulou1, Lydia Mavraidi1, Stefanos Voikos1, Yoav Nahshon2, Matthias B¨ uschelberger2, Pablo de Andres2, Artemis Lavasa3, Anastasios Karakostas3, Ioannis Papadimitriou4, Leonardo Cosma5, Javier Gomez6, Harrison de la Rosa Ram´ ırez7, Anastasios Zafeiropoulos1, Francisco Fraile7, Andr´ es Boza7, Ciprian Cˆ andea8and Symeon Papavassiliou1 1School of Electrical and Computer Engineering, National Technical University of Athens, Athens, Zografou, 15780, Greece e-mails: {ktsitseklis, gstavr, lmavraidi, stefvoikos}@netmode.ntua.gr, [email protected], papav[email protected] 2Fraunhofer Institute for Mechanics of Materials IWM, Freiburg, Germany, e-mails: {lukas.morand, yoav.nahshon, matthias.bueschelberger, pablo.de.andres}@iwm.fraunhofer.de 3DRAXIS Environmental S.A., Thessaloniki, Greece, e-mails: {alavasa, akarakos}@draxis.gr 4Information Technology Institute, Centre for Technology Hellas, 6th km Charilaou Thermis, Thessaloniki, 57001, Greece Thessaloniki, Greece, e-mail: [email protected] 5CETMA Brindisi, Italy, e-mail: [email protected] 6ADVANCED MATERIAL SIMULATION SL, Bilbao, Spain e-mail: javier[email protected] 7Research Center on Production Management and Engineering, Universitat Polit` ecnica de Val` encia, Spain, e-mail: [email protected].es 8ROPARDO S.R.L., Sibiu, Romˆ ania, e-mails: {ciprian.candea}@ropardo.ro Abstract—The Industry 4.0 paradigm has led to a revolution in the manufacturing domain, driving improvements in product quality, reducing time-to-market, and lowering environmental costs through digitization and advanced analytics. In spite of this, many Small and Medium Enterprises (SMEs) face significant barriers in adopting such solutions that could assist them in the competition with their larger rivals, including limited resources, expertise, and access to cutting-edge software. Aiming to mitigate this gap and empower SMEs, the DiMAT framework is developed offering a number of open-source, scalable tools organized into three core Suites. This paper deals with the description of DiMAT’s architecture, providing an in-depth view of its strong aspects. Beginning with a high-level overview of the architecture and then delving in more detail on each toolkit’s structure, this article demonstrates the manner in which the toolkits are designed following state-of-the-art practices and how they can be deployed. Index Terms—Materials Manufacturing, Industry 4.0, Digitization, Reference Architecture, Small and Medium Enterprises, SMEs This research work has received funding from the European Union’s Horizon Europe research and innovation programme under the project DiMAT with grant agreement No 101091496. I. INTRODUCTION Today, rapid changes in the Information and Communication Technologies (ICT) sector present new challenges and significant opportunities for stakeholders in the manufacturing domain [1]. Technological advances such as the integration of Internet of Things (IoT) devices, artificial intelligence (AI), and cloud computing enable manufacturers to optimize production processes, reduce costs, and accelerate innovation cycles [2]. Cutting-edge software tools and IoT applications enable the easier and more rapid production of novel materials at a reduced cost. In addition, events such as the climate crisis and others, such as the recent Covid-19 pandemic, further highlight the importance of digitization. Competition among companies involved in the manufacturing industry leads even Small and Medium Enterprises (SMEs) to adopt new technologies to survive against the more agile and resourceful larger companies. Despite the acknowledged need for the adoption of new technologies by SMEs, significant challenges pose barriers that must be overcome [3]. First of all, the limited budget available for purchasing and maintaining expensive closedsource software is oftentimes a prohibiting factor towards adopting a more modern approach to their business processes. In addition, the lack of experienced personnel and resistance to change have been identified as potential obstacles for SMEs. [4]. This is particularly critical in manufacturing, where many SMEs rely on legacy systems and traditional workflows that lack the agility and scalability required to compete with larger firms equipped with advanced tools and infrastructures. In this landscape, the European Union (EU) acknowledges the crucial role of modern and innovative SMEs to ensure their competitiveness in the global market [5] and acts as a major funding agency in relevant research areas through programs such as the “Horizon Europe” research and innovation program1. DiMAT2, a project funded by the EU, aims to address the need for a holistic digital transformation of SMEs in the manufacturing industry through the provision of open source, affordable, and scalable tools targeted to the needs of SMEs helping them in their efforts to bridge the gap with the competition, successfully enter the Industry 4.0 era, and produce market leading and high quality products. The DiMAT framework is organized into three core software Suites, covering different aspects of SME operation: data management and lifecycle assessment, materials modeling and design, and simulation and optimization of manufacturing processes. Each Suite consists of three specialized toolkits covering more specific areas. Combining these toolkits can generate a versatile ecosystem of possible solutions that aim to modernize and transform traditional SME workflows. Aiming to develop a flexible, modern and scalable software solution, DiMAT’s architecture follows closely the Industrial Internet Reference Architecture (IIRA) [6] which defines four viewpoints (i.e., Business, Usage, Functional and Implementation), ensuring an ambitious design that covers the needs of stakeholders while, at the same time, using state-of-theart software tools for developing its components. In this article, the architectural schema of DiMAT will be presented in both a high-level overview and greater detail, offering useful insights on its advantages and capability of addressing various requirements of numerous industrial business processes. The remainder of the paper is structured as follows. In Section II, the overall architecture of the DiMAT framework is presented alongside the common vocabulary adopted by the consortium. Then, Sections III to V delve into each Suite and present a detailed overview of each toolkit’s architecture. Finally, Section VI concludes the paper. II. OVERALL ARCHITECTURE The DiMAT architecture is designed to provide a robust, scalable, and user-focused solution tailored specifically for SMEs in the materials manufacturing sector. By adhering to the Industrial Internet Reference Architecture (IIRA) [6], the framework ensures a holistic and multi-faceted design approach that aligns technical excellence with practical business 1https://research-and-innovation.ec.europa.eu 2https://dimat-project.eu/ needs by incorporating analysis from different viewpoints (i.e., Business, Usage, Functional and Implementation viewpoints). This alignment is critical, as SMEs often operate under constraints that larger enterprises do not face, such as limited budgets, smaller IT teams, and a lack of in-house expertise to deploy and maintain complex solutions. The high-level architectural schema of DiMAT can be seen in Fig. 1. In this figure, the three distinct Suites with their respective toolkits can be seen. In Sections III-V, a more indepth analysis will be provided for each of them. DiMAT architecture addresses user fatigue in digital solutions [7], [8], which refers to the cognitive and physical exhaustion experienced when interacting with complex digital systems. A well-structured user interface, combined with integrated authentication mechanisms, significantly reduces frustration and supports seamless interaction [9], [10]. DiMAT architecture incorporates a robust Identity and Access Management (IAM) solution [11], implemented through Keycloak 3. This integration ensures a secure and streamlined environment by managing authentication and access control, allowing only authorized users to access specific toolkits and functionalities. Keycloak provides features such as Single Sign-On (SSO) and Role-Based Access Control (RBAC). Integration reduces friction and eliminates the need for multiple logins across toolkits, streamlining the user experience while maintaining high-security standards. Each Suite has its own front-end page that leads to each toolkit. Each user of DiMAT has access exclusively to the relevant (i.e., purchased) Suites and toolkits. Each toolkit has its own frontand back-end as well as its own dedicated storage space. In this way, the toolkits can be deployed either in the cloud and be provided as Software-as-a-Service (SaaS) or separately on-premise, depending on the specific customers’ needs. DiMAT’s cloud-based architecture uses containerization to enable both vertical and horizontal scalability on demand, ensuring that computing resources can be provisioned for varying workloads. By packaging system components within lightweight Docker4containers, SMEs can seamlessly deploy and manage independent services across diverse infrastructures (with Kubernetes potentially integrated in the future). Infrastructure as Code (IaC) methodologies can be employed using Terraform [12] to provision cloud environments declaratively, while Ansible automates configuration management tasks, thereby minimizing manual intervention and reducing operational overhead. In addition to containerized services, DiMAT’s architecture includes mandatory supporting components such as domain name and SSL configuration for secure communication, a reverse proxy for managing incoming requests, and a VPN for secure remote access—all provisioned and maintained automatically through Ansible. This integrated cloud approach enables cost-effective deployments, reducing the need for complex manual configurations and minimizing 3https://www.keycloak.org/ 4https://www.docker.com/ Fig. 1. Overall Architecture of the DiMAT framework. operational expenses. SMEs can easily set up new cloud instances, deploy applications, and manage their infrastructure with minimal effort. The ability to scale resources dynamically optimizes cost-efficiency. By leveraging containerization, automation, and cloud provisioning, DiMAT’s architecture ensures seamless scalability and low-cost cloud adoption, making it an ideal choice for SMEs seeking a robust digital infrastructure with minimal investment. Taking into account that DiMAT consists of three different Suites and nine toolkits in total, the aspect of interoperability gains significant importance for their interconnection. To this end, DiMAT adopts the use of standardized communication protocols (e.g., HTTP, MQTT, etc.) and elaborates on the use of a unified terminology, creating a common language that promotes clarity and eases the toolkits’ interactions. For instance, a user working on life-cycle assessments and simulations can rely on the same standardized terms and data formats, minimizing integration complexities. A crucial aspect of this effort is the generation of the DiMAT’s vocabulary, described in the next subsection. DiMAT’s architecture investigates an AI layer to handle GPU-intensive tasks, thereby preparing the DiMAT cloud environment for AI technology stack that can be deployed either on a smaller scale in on-premise settings or on AI provider clouds. In this layer, different configurations are evaluated for how the Backend and GPU components interact—ranging from a direct, synchronous function call that waits for the GPU’s response to more advanced, asynchronous approaches that integrate a queue (either on the GPU side or, preferably, on the Backend side) to manage task scheduling and resource allocation. By incorporating these AI-focused design considerations, DiMAT also examines how its various toolkits need to be adapted in terms of architecture, modules, and deployment processes to effectively leverage GPU acceleration and advanced machine learning capabilities, ensuring efficient performance and broad applicability across different SME scenarios. By combining modularity, interoperability, scalability, and security, the DiMAT architecture provides SMEs with a powerful yet accessible framework to embrace digital transformation. Its careful alignment with real-world SME constraints and requirements ensures that it not only introduces cutting-edge technologies but does so in a way that is practical, sustainable, and impactful for its users. A. DiMAT’s vocabulary One of the key challenges in ensuring seamless interoperability across toolkits and systems is the inconsistency in the terminology used to refer to the same elements (e.g., physical properties, machinery equipment, etc.). The partners of the DiMAT consortium collaborated in establishing a common vocabulary, adopting a common terminology for various concepts relevant primarily to the materials manufacturing industry and the scope of the DiMAT project in general. This common terminology not only standardizes references within the materials manufacturing domain but also supports the broader objectives of the DiMAT project by streamlining communication between its toolkits and third-party systems. Through an iterative process, the DiMAT consortium members systematically collected the necessary terms employed and decided on standardized names (ontological IDs) following certain conventions (e.g., capitalization, camelCase writing). The sources for the decided names are based on publicly defined standards including ISO 2062:2009, ISO 1133, DIN EN ISO 6892-1 among others. In the cases where no specific standards were identified, the published scientific literature and manufacturer’s manuals were used. As of December 2024, DiMAT’s vocabulary encompasses an extensive list of 230 terms. Developing the vocabulary allows making toolkit interconnections easier, since queries to databases and API requests can be made without facing errors that arise from the use of different aliases for the same entities (e.g. toolkit A refers to ‘Density’ and toolkit B refers to ’density’ for the same physical property). Moreover, beyond internal consistency, the DiMAT vocabulary facilitates integration with external applications and thirdparty software. By aligning with standardized terms, DiMAT toolkits can easily connect with business and manufacturing software, ensuring that the interoperability benefits of the framework extend beyond its internal ecosystem. This standardized approach ultimately ensures a more cohesive and efficient experience for SMEs adopting DiMAT, reducing technical barriers and enhancing the framework’s usability as a whole. III. DATA AND ASSESSMENT SUITE Addressing the SMEs’ needs for secure data storage, traceability, and extraction of useful insights from their data on materials and processes, a set of tools is developed. The toolkits of this Suite have been developed to increase the workforce’s digital skills and lead to the production of more environmental-friendly, less costly products, while also reducing time-to-market. The three toolkits that comprise this Suite are described in detail in the following sections. A. Cloud Materials Database - CMDB The CMDB provides a semantics-based data management platform for storing and linking heterogeneous materials and processing data in a FAIR (Findable, Accessible, Interoperable, Reusable) manner. It necessitates the use of three database types: •Relational Database: Storage of tabular data (e.g., time series from experiments) •Graph Database. Storage of semantically annotated data and metadata •Object Storage: Storage of raw data files These storage solutions are orchestrated via a central backend, allowing users (both human and machine) to interact with the CMDB through a user-friendly GUI (web frontend) or a Python SDK5. The latter provides programmatic access, enabling users to automate workflows and integrate CMDB functionalities into their applications seamlessly. The CMDB utilizes a microservice architecture deployed on Docker containers, ensuring scalability, flexibility, and ease of deployment. Each microservice is responsible for specific functionalities, promoting a modular and maintainable architecture. Built on the Dataspace Management System (DSMS) developed at Fraunhofer IWM [13], the CMDB is a web platform that enhances data management by using semantic technologies to adhere to the FAIR data principles and enable automation of data processing workflows. Users can create and publish applications and data processing scripts that access stored data through the GUI, a REST API, or the Python SDK. A vital component of CMDB is the data2rdf tool6, which facilitates data integration by converting raw files (e.g., measurement data, sensor data, simulation files) into RDF format. The data2rdf tool processes input files and outputs RDF files while referencing an SQL database that stores time-series data from raw files. Overall, the CMDB plays a pivotal role within the DiMAT architecture, acting as a central data repository for various toolkits. To facilitate this, a vocabulary service has been integrated into the CMDB used to manage DiMAT’s vocabulary. This integration allows the CMDB to streamline access and 5https://pypi.org/project/dsms-sdk/ 6https://pypi.org/project/data2rdf/ data sharing, utilizing standardized vocabulary to annotate data for enhanced interoperability. B. Knowledge Acquisition Framework - KAF The KAF toolkit organizes information about materials and business processes in a structured format, represented as a Knowledge Graph (KG). The KAF toolkit facilitates data exploration, providing visualizations of subgraphs that highlight the interconnection among different entities. Carefully designed dashboards allow personnel to have a clear overview of the stored data. Moreover, through the incorporated analytics, useful insights can be extracted, acting as guides in the design process of the company. The main architectural components of KAF are the graph database and the front-end and back-end services. Each of these services is organized as a separate docker container. Communication among the various containers is made possible through the use of REST APIs. The centerpiece of KAF’s architecture is its graph database, which is used to store the data in a structured manner defined by its appropriately designed schema. The designed knowledge graph (KG) schema captures information about manufacturing materials and industrial business processes. Around this KG, various functionalities have been developed. Through the front end, users can interact with the toolkit and access the available services. These services are hosted in the backend container. They are independent of each other and can be deployed in different equipment if necessary. The querying mechanisms, offer the opportunity for users to interact with the stored data in the KG either via using a Large Language Model (LLM) or through various predefined menus. Employing these methods, the users can gain insights on the stored datasets and the manner in which different data types depend on one another. To aid the latter, helpful visualizations are also generated that depict parts of the knowledge graph, providing a better understanding of the business’ processes captured by the toolkit. Last but not least, beyond data storage, KAF provides powerful analytic functionalities helping its customers gain helpful insights on their data. Capitulating on its KG structure, a number of graph based analytics can be supported, such as similarity extraction, community detection and recommendations. The analysis of the data stored in KAF can be employed during the design phases of the manufacturing procedure and aid in cost reducing as well as making the onboarding of new employees easier. C. Materials Environmental and Cost Life Cycle Assessment - MEC-LCA The Materials Environmental and Cost Life Cycle Assessment (MEC-LCA) toolkit provides a high-level graphical overview of the environmental and economic impact of industrial processes. Product designers, engineers and decisionmakers can assess environmental and cost indicators through dedicated dashboards allowing them to identify environmental Fig. 2. DiMAT Data and Assessment Suite Architecture. and financial hotspots, evaluate process performance, and conduct benchmarking. DiMAT MEC-LCA is built on a Docker-based architecture consisting of three main containers, namely Grafana7, Laravel8 and MySQL9. Each container performs a specific role in managing and processing data for Life Cycle Assessment (LCA) and Life Cycle Costing (LCC) analysis. Grafana serves as the user interface for visualization and analysis. It connects directly to the MySQL database, allowing administrators and users to retrieve and analyze data through queries. Integrated within Grafana are the Infinity plugin, which supports direct uploads of CSV files for rapid graph generation without having to query the database; the Variable plugin, which enables dynamic calculations for specific scenarios based on the user’s selection; the Data Manipulation plugin, which allows users to update existing data in the MySQL database directly from the interface. Laravel, a PHP-based framework, is used by administrators for database management tasks. It enables the creation of seeders, which automate the insertion of datasets into MySQL. This process ensures that data is organized and ready for querying and analysis. Finally, MySQL acts as the central database system, providing a relational framework for storing and retrieving data. The data flow begins with data insertion, either through Laravel’s seeders or direct uploads using Grafana’s Infinity 7https://grafana.com/ 8https://laravel.com/ 9https://www.mysql.com/ plugin. Once stored in MySQL, the data can be queried and visualized in Grafana. User authentication and access control are managed through Keycloak, which operates outside the Docker stack. This ensures that access to the platform’s resources is regulated and secure. Administrators can configure user permissions and determine dashboard access levels, ensuring secure and regulated access to platform resources. A key goal of MEC-LCA is to address the fact that traditional LCA software is typically used by experts, and the results can be difficult to interpret without specialized knowledge. MEC-LCA builds on the rigorous, ISO-compliant LCA assessments conducted outside the system, focusing on making these results more accessible and easier to understand for a broader range of users, ultimately supporting more sustainable choices across the value chain. IV. MODELING AND DESIGN SUITE The DiMAT Modeling and Design suite aims to guide material design, predict material behavior, and improve material performance before manufacturing. This suite addresses the needs of material designers by supporting the discovery of correlations between processing and performance, providing tools to analyze data and patterns, and allowing the virtual evaluation of different material configurations. The suite uses a combination of traditional numerical methods and AI technologies, adopting multiple approaches to analyze and solve problems from different perspectives. Fig. 3. DiMAT Modeling and Design Suite Architecture. A. Materials Design Framework - MDF The MDF is an ontology-based knowledge-system designed to assist engineers in materials modeling and design tasks. The MDF comprises three main applications: 1) Search app: Facilitates in-depth searches for materials and process data. 2) Correlation app: Analyzes correlations within the data. 3) Material relations app: Manages and identifies material models to simulate observed material behavior. The MDF features a custom front-end built with Angular, offering an intuitive and user-friendly interface for accessing the above-mentioned applications. Deployed using Docker containers, the MDF ensures scalability, flexibility, and ease of deployment. Each application (search app, correlation app, and material relations app) is containerized to promote modularity and maintainability. The MDF operates on top of the CMDB, which serves as both, a source for materials and process data and a repository for materials modeling data. The search app utilizes keyword-based free-text search to identify relevant materials and process data in the CMDB for specific applications at hand. This functionality is implemented via the Python SDK to interact with CMDB, with plans to enhance the user experience and depth of searches through integration with a large language model like LLaMA [14]. Once relevant datasets are identified, they can be analyzed using the correlation app. The correlation app is a lightweight implementation built with Streamlit [15], providing tools for analyzing and visualizing correlations within and between different datasets stored in the CMDB. It enables users to gain deeper insights into materials and process data, facilitating data-driven decisionmaking. This app offers specialized analytical capabilities tailored to the needs of materials engineers. The material relations component helps users to explore the landscape of material models. This service identifies relevant relationships that describe material behavior, fostering innovation and informed decision-making in materials modeling and design. The material relations app features a user-friendly frontend developed with Angular and a backend based on Python FastAPI. It leverages the CMDB to store semantically annotated material models, with annotations derived from the continuum modeling ontology created during the EU project Marketplace10. B. Materials Modeler - MM The Materials Modeler (MM) toolkit is designed to provide robust AI-driven analysis and prediction capabilities for material properties under various manufacturing conditions. The architecture of MM follows a modular approach with distinct layers handling different aspects of the data processing and analysis pipeline. At the core of MM’s architecture is a multilayered system comprising communication, data preprocessing, AI analysis, and prediction layers. The communication layer interfaces with other components, facilitating seamless data exchange with the Cloud Materials Database (CMDB) and other DiMAT toolkits. A message bus handles the internal communication between components, with a security layer 10https://www.the-marketplace-project.eu/ ensuring data protection throughout the process. The data preprocessing layer employs Python libraries such as Pandas [16] for handling various data formats, particularly CSV files containing material properties and manufacturing parameters. This layer performs essential tasks including data cleaning, handling missing values, normalization, and preparation of data structures optimal for AI algorithm consumption. Special attention is given to raw data processing, employing techniques like imputation for missing values and various transformation methods to ensure data quality. The AI analysis layer leverages a suite of machine learning algorithms, including both traditional ML approaches through scikit-learn [17] and advanced techniques like XGBoost [18]. This layer is responsible for uncovering latent relationships between material structures and their properties, facilitating the prediction of material behaviour under different conditions. The analysis includes feature importance calculations, helping identify the most significant parameters affecting material properties. The prediction and inference layer generates insights and recommendations based on the AI analysis results. The layer supports both univariate and multivariate prediction models, offering flexible approaches to material property prediction. A key feature of this layer is its hybrid optimization approach that combines trained AI models with a genetic algorithm optimization strategy, allowing for efficient exploration of the complex parameter space while respecting physical constraints provided by the user. This hybrid approach enables the system to handle multiple competing objectives and suggest optimal material compositions and process parameters based on desired target properties. These predictions and optimization results are visualized through an interactive dashboard built using Streamlit [15], which provides an intuitive interface for parameter optimization and result visualization. The system presents comprehensive statistics including parameter ranges, confidence levels, and relative errors, enabling users to make informed decisions about material design. All results can be exported in standardized formats for further analysis or documentation. This approach ensures that complex relationships and predictions are not only accurate but also presented in an accessible and actionable manner to users. The entire solution is containerized using Docker, ensuring consistency across different deployment environments and facilitating scalability. The MM toolkit can be deployed either as a cloud-based service or on-premises, depending on specific user requirements. For optimal performance, the system requires substantial computing resources, particularly for handling ML workloads, with recommendations including ample RAM and modern CPU capabilities, with optional GPU acceleration for enhanced model training and inference tasks. The toolkit exposes several API endpoints for core functionalities including data upload, exploratory data analysis, feature importance analysis, and property prediction, supporting standard CSV data format. This architectural design ensures that MM can efficiently serve its role in the materials modeling and design process while maintaining flexibility and scalability for future enhancements. C. Materials Designer - MD The DiMAT Materials Designer (DiMD) is a tool that provides the user with the capability to define complex materials, such as composite materials, in terms of their base component and internal microstructure and obtain the mechanical properties of the new material. Properties are calculated via a virtual testing approach, which consists of building a Finite Element Method (FEM) model and solving its equations to extract the requested properties from the results. DiMAT MD is implemented using a microservice architecture and is deployed on Docker containers. The front-end consists of a single container serving the User Interface (UI) implemented using the NiceGUI [19] Python framework. Its main function is to collect user requests for defining new materials or retrieving calculated properties. The backend of the toolkit is articulated in different containers, each responsible for a specific task: •The main web server, implemented using the FastAPI [20] framework, responsible for the definition of new materials. •The database, a PostgreSQL [21] instance responsible for data storage. •The simulation scheduler, a simple web server in FastAPI responsible for defining and scheduling the simulations, using an auxiliary internal database service. An external calculation code, responsible for executing the FEM simulations, is accessible from the toolkit containers. The typical data flow starts when the user defines a new material in the toolkit frontend. The UI component sends all data to the backend which saves them into the main database and notifies the simulation scheduler of the newly received request. The simulation scheduler defines the necessary virtual tests to be executed, prepares the input files for the FEM simulations and starts their execution via the calculation code. When the results are available, the scheduler parses them and sends the calculated properties back to the main web server, which saves them in the toolkit database and provides the results to the UI when asked by the user. Communications between the front-end, the main web server and the scheduler happen via RESTful APIs. The toolkit can be used via the UI interface or via the exposed API of the main web server of the backend. Authentication and authorization are provided by a Keycloak server outside of the toolkit stack. The toolkit also offers the possibility to define parametric studies as well as a fast analytical approach for the initial screening of material properties. The DiMAT MD toolkit enables users to define complex materials and explore their behavior and performance through a simple web interface, eliminating the need for software installation. This reduces the cost barrier to accessing advanced simulation capabilities for virtual testing. V. SIMULATION AND OPTIMIZATION SUITE This suite aims to build a set of digital tools for material manufacturing simulation and material properties and behavior prediction, together with a set of digital twins for process monitoring, control, and optimization. The main objective of the tools is to reduce the costs, time, materials, and energy consumption associated with industrial trials, by providing a set of functionalities tailored to create efficient simulation processes and determine the behavior of mechanical characterization models for use in AI training and prediction. This suite is dedicated to supporting different pilot use cases, such as the development of new polymer material and determining the best processing condition for the glass-forming process. A. Materials Mechanical Properties Simulator - MMS The Materials Mechanical Properties Simulator (MMS) toolkit is a numerical simulation tool designed to predict the mechanical properties of materials and the mechanical performance of components. This toolkit plays a crucial role in optimizing material design and ensuring high-performance outcomes. MMS integrates multiscale simulations by combining micro-, meso-, and macro-scale models, incorporating atomistic simulations with LAMMPS [22] as well as finite element models performed with CALCULIX [23]. LAMMPS is an advanced molecular dynamics solver that allows for precise and flexible modeling and analysis of material behavior at the atomic scale. CALCULIX is a robust open-source finite element code that facilitates efficient simulation of structural mechanics problems across a wide range of scales. LAMMPS and CALCULIX offer high customizability through userdefined potentials or subroutines, allowing users to adapt material models or element behavior to meet specific simulation requirements. This flexibility enables the integration of advanced properties or behaviors, such as nonlinear material responses or specialized boundary conditions, expanding its applicability to complex engineering problems. MMS toolkit combines also physical models with AI and data-driven approaches, resulting in a powerful hybrid simulation framework. This combination enables both real-time simulations and AIenhanced physical models. Additionally, the integration of AI with physical models ensures explainability, expanding the applicability of data-driven components and enhancing their reliability. The backend of the MMS toolkit offers two modes of functionality: physical simulation and real-time calculations. In physical simulations, users input the necessary data, which is then sent to the solver, and the output is retrieved upon completion. Depending on the complexity and size of the model, these simulations may require a significant amount of time to process. For real-time predictions, the physical problem is pre-solved to generate a large database used to train a surrogate machine learning model. The surrogate model replicates the input-output behavior of the original physical model while providing the added capability of delivering results in real time. The backend can handle complex modeling workflows, which include preprocessing (data validation, mesh generation, and boundary condition assignments), solving, and postprocessing at various multiscale levels. The process begins with data input, where material properties and simulation parameters are specified through an intuitive user interface, and concludes with graphical visualization of the results. The MMS front-end provides a user-friendly interface that enables users to define and execute simulations without requiring extensive technical expertise. It features interactive visualizations and analytical tools to interpret simulation results, ensuring accessibility for engineers and scientists. MMS is containerized using Docker, offering deployment flexibility and compatibility across a variety of environments. Its API endpoints facilitate seamless integration with other toolkits within the DiMAT framework, enabling comprehensive workflows that span multiple suites. This robust architecture establishes MMS as a versatile and powerful tool for advancing material modeling and design. B. Materials Processing Simulation - MPS The Materials Processing Simulation (MPS) is a toolkit for determining manufacturing conditions and concepts by recreating manufacturing processes to predict materials’ behavior when modifying the process parameters and material properties while simulating their application, results, and requirements in each of the materials, processes, and processing conditions in the DiMAT Open Cloud Materials Database. Fundamentally, it works on the simulation of polymer extrusion processes, curing resin cycle for composite materials, and glass bending process. The simulation of materials processing using the MPS toolkit is generated by specific CAD geometry and boundary conditions using finite element method (FEM) and computational fluid dynamics (CFD), which uses powerful software such as Prepomax and Open-Foam Prepomax is an open-source graphical user interface (GUI) designed to facilitate pre-processing, post-processing, and solving tasks for finite element analysis (FEA). The software provides an intuitive and user-friendly front end for the CalculiX solver, with a streamlined environment to perform complex simulations. Prepomax11 offers a cohesive simulation workflow, using integrated models during the different steps, an analysis setup, and results visualization. The GUI provides the users with the possibility to import complex CAD geometries, customize materials, assign boundary conditions, and create accurate finite element meshes. Once the simulation is defined, CalculiX solver executes the analysis, and results can be visualized in the same environment using post-processing tools the software offers. The software offers different types of calculations, structural and thermal analysis, and the opportunity to develop thermos-mechanical ones. OpenFOAM12 is a powerful open-source software suite designed for computational fluid dynamics (CFD). The software offers high flexibility and scalability, providing an extended library of solvers and utility models, such as complex physical solvers on fluid dynamics, heat transfer, and more. The software offers a code open, which brings the 11https://prepomax.fs.um.si/ 12https://www.openfoam.com/ Fig. 4. DiMAT Simulation and Optimization Suite Architecture. community the opportunity to develop new uses (as was the case in the MPS toolkit, in which an existing solver was modified for better performance in the curing resin degree determination, applicable for composite materials). The users create simulations by defining boundary conditions, mesh geometries, material properties, and numerical settings in the case directory. OpenFOAM employs a finite volume method to discretize equations and solve them iteratively. The simulation is executed through a command line interface, with the optional use of tools for visualization results like ParaView. The versatility of this software allows different kinds of calculations, such as fluid dynamics, heat transfer, solid mechanics, and specialized applications defined by users. MPS uses data from material transformation process, geometry, processing conditions, and materials properties for recreating a virtual material processing, giving output data that is important for the correct processing of the materials, such as pressure, temperature, etc. This output will be used to increase the database and optimize the transformation process by taking the new manufacturing parameters and feeding them back into the process to obtain new data. The functionality of MPS is based on the combination of a set methodology in which a fair number of simulations are carried out using specific materials and manufacturing processes. The simulation results are then compiled to generate a backend which can be a reduced model or a script that automates the calculation completely. These steps are followed by the front-end development, offering a digital interface with 3D graphical contours to allow the simulation results to be analyzed. Finally, through Dockerizing, the system application is encapsulated to save, distribute, and install the application. MPS also uses data population mechanisms to manipulate the different kinds of input and output to develop the simulations properly and store them on the DiMAT database. Querying mechanisms are needed for the user to manipulate the toolkit correctly. Functionalities of analytics and visualization simulation results will bring users a proper understanding of the process simulation and the changes needed. The different functionalities are connected to Database APIs to provide proper data to the user and continue feeding the DiMAT database. C. Digital Twin for Process Control - DTPC In the DTPC toolkit, digital twins (DTs) are created as virtual abstractions of real manufacturing devices or materials. The developed DTs communicate with their physical counterpart with the help of IoT devices. The DTs provide access to simulation/emulation algorithms as well as functionalities for process control based on specific rules that are either explicitly mentioned or inferred from historical data. The architecture of the DTPC toolkit adopts a modular three-layer approach. On the bottom level, Virtual Objects (VOs) acting as digital abstractions of physical devices are defined. The design of virtual objects is accomplished by using the Nephele VO protocol stack, which is aligned with the Web-of-Things standard. The developed Virtual Objects can communicate over various protocols with each other and thus enable the operation of “higher level” VOs, composite VOs