With increasing maturity, their functions range from state description to analysis and prediction, and even to control interventions in production and logistics processes [4-5]. With a maturity model, digital twins of production and logistics systems can be systematically classified according to their expansion stage (maturity) using indicators by evaluating the degree to which previously defined requirements are fulfilled [6].<\/p>\n\n\n\n
Maturity model for digital twins<\/h2>\n\n\n\n
The maturity model developed, specifically designed for small and medium-sized enterprises (SMEs), is the result of a systematic literature review and analysis, in which 31 maturity models for digital twins from the Web of Science database were identified and analyzed using the search string \u201cdigital twin* AND (level OR maturity)\u201d (see e. g. [5, 7-8]). A detailed documentation is available in the research data repository of the University of Kassel [9]. <\/p>\n\n\n\n
The new maturity model is based on four dimensions, each comprising four indicators:<\/p>\n\n\n\n
The first dimension<\/strong>, \u201cconvergence,\u201d assesses the degree of synchronicity between the real and digital systems using the indicators connection mode <\/em>(1.1), update frequency <\/em>(1.2), level of detail <\/em>(1.3), and data interpretation <\/em>(1.4).<\/p>\n\n\n\n The second dimension<\/strong>, \u201ccapability,\u201d describes the functional performance of digital twins using the indicators cognition level <\/em>(2.1), degree of autonomy <\/em>(2.2), simulation capability <\/em>(2.3), and learning capability <\/em>(2.4).<\/p>\n\n\n\n The third dimension<\/strong>,\u201cintegration,\u201d assesses the depth and breadth to which digital twins are embedded in existing systems, processes, and life cycle phases using the indicators integration breadth <\/em>(3.1), integration depth <\/em>(3.2), life cycle integration <\/em>(3.3), and trustworthiness <\/em>(3.4).<\/p>\n\n\n\n In addition to the three technical core dimensions, the fourth dimension<\/strong>, \u201cstrategy and organization,\u201d evaluates the prerequisites for the successful use of digital twins using the indicators strategic anchoring <\/em>(4.1), personnel competence <\/em>(4.2), collaborative network <\/em>(4.3), and process governance <\/em>(4.4). Further details can be found in [9].<\/p>\n\n\n\n The three technical core dimensions span a three-dimensional assessment space in which the expansion stage of a digital twin can be located for each dimension (Fig. 1<\/strong>). The fourth dimension acts as a normative guiding framework for the sustainable implementation of digital twins along the three core dimensions (sphere radius = maturity stage).<\/p>\n\n\n\n The expansion stage of a dimension is determined by the rounded average of its associated indicators. Stages 1 to 4 can be assigned for each dimension (example: convergence = 2, capability = 1, integration = 1, radius (strategy and organization) = 2). The technical maturity of the digital twin should not exceed the organizational prerequisites at the dimension stage, but individual indicators may have a higher maturity stage depending on the specific application purpose.<\/p>\n<\/div>\n<\/div>\n\n\n\n In order to iteratively test different expansion stages for digital twins, the fischertechnik\u00ae Learning Factory 4.0 [10] is available at the Department of Production Organization and Factory Planning at the University of Kassel. The decision to implement a fischertechnik\u00ae Learning Factory 4.0 is based on a systematic literature review and analysis [9]; in particular, the modular structure, the consistent use of standardized data interfaces, the detailed documentation for rapid expandability, and the transferability of knowledge to real production and logistics systems were decisive selection criteria.<\/p>\n\n\n\n Profiles are available for 17 different learning factories related to digital twins, which include a general description of the learning factory, its scope and system boundaries, and the components used. An analysis of the identified learning factories shows that the fischertechnik\u00ae Learning Factory 4.0 forms a suitable basis for a physical test environment for digital twins without significant initial hurdles.<\/p>\n\n\n\n The fischertechnik\u00ae Learning Factory 4.0 comprises the modules \u201cDelivery and Pickup Station\u201d (DPS), \u201cVacuum Gripper Robot\u201d (VGR), \u201cHigh-Bay Warehouse\u201d (HBW), \u201cMulti-Processing Station with Oven\u201d (MPO), and \u201cSorting Line with Color Detection\u201d (SLD). These modules represent a realistic production and logistics flow, enabling experimental testing of digital twins along real event and state sequences.<\/p>\n\n\n\n The further development of the standard kit into a test and development environment for digital twins is based on a system architecture with a multi-layered logic structure (Fig. 2<\/strong>). At the lowest layer, \u201cdatabase,\u201d data models are used for the structured acquisition of sensor, state, or operating data, for example. The data is stored persistently and therefore available in a consistent form for further processing steps. The Asset Administration Shell [11] serves as an information model and forms the semantic basis for describing plant resources, products, and processes. <\/p>\n\n\n\n The \u201cmethods and models\u201d layer provides digital representations that map system behavior synchronously with the real system. Methods for data acquisition, preprocessing and analysis, visualization, simulation, and optimization access the available data, information, and digital representations as required and include both classical approaches and AI-based methods.<\/p>\n\n\n\n At higher expansion stages, these methods and models can be used to generate recommendations for operational control. An application manager orchestrates data and process flows based on defined workflows and uses knowledge models to map domain-specific relationships. The top layer comprises \u201cservices\u201d such as monitoring, forecasting, or optimization, which are provided by digital twins using the application manager.<\/p>\n\n\n\n To represent the system behavior of the fischertechnik\u00ae Learning Factory 4.0, a simulation model implemented in Tecnomatix Plant Simulation by Siemens is available. This model serves as an integral component of a digital twin and can be used for forecasts and improvements based on real-time data. The simulation model is also connected to NVIDIA Omniverse, enabling 3D visualization and, in the future, the coupling of additional digital twins (Fig. 3<\/strong>). The combination of real system operation, semantic structuring, and simulation models enables both the use of the digital twin at lower maturity stages and the expansion, e.g., of predictive and AI-based functionalities.<\/p>\n\n\n\n In the following, three selected use cases serve to illustrate the possibilities for experimental testing digital twins in a learning factory. Dimension 4 is not considered due to the laboratory environment.<\/p>\n\n\n\n In the first use case (Fig. 4<\/strong>: Use Case 1 <\/strong>\u2013 Real-time monitoring of the learning factory), the digital twin is used exclusively for recording real-time state and for providing transparency regarding material flow, plant utilization, and throughput times. Communication is unidirectional (Indicator (I) 1.1, Stage (S) 2) via the network protocol Message Queuing Telemetry Transport (MQTT), whereby sensor states, material flow events, and machine statuses are transmitted in standardized topics (e.g., ft\/<station>\/<state>) and persistently stored in a SQL database. The update frequency is near real time (I 1.2, S 4). The Asset Administration Shell describes, e.g., basic system and product information, but without semantic interpretation logic (I 1.4, S 2).\u00a0<\/p>\n\n\n\n The simulation model is merely synchronized, without forecasting and without intervention in the real process (I 2.3, S 1). In the fischertechnik\u00ae Learning Factory 4.0, the events relate to real process chains via DPS, VGR, HBW, MPO, and SLD. The Digital twin has neither cognitive abilities (I 2.1, S 1) nor autonomy (I 2.2, S 1), but it provides the necessary basis for all subsequent expansion stages. The integration comprises the relevant system modules in production and logistics (I 3.1, S 2).<\/p>\n\n\n\n The first use case with the expansion stages (2-1-1-2) addresses key challenges that also limit the initial implementation of digital twins in real factories (see also [12]). In industrial environments, many implementation projects fail due to insufficient data quality, undocumented interfaces, lack of semantic structuring, and heterogeneous data formats from sensors and control technology.<\/p>\n\n\n\n Seemingly simple requirements (e.g., stable data flow, harmonized topics, unique identification of resources, unambiguous timestamps, and consistent data persistence) often represent the largest cost and effort factor in practice. In addition, the introduction of a standardized information model such as the Asset Administration Shell requires significant competence building. Limitations arise where existing systems do not offer retrofit solutions or where proprietary protocols make interoperability difficult.<\/p>\n\n\n\n At the same time, there are also significant opportunities; transparency, data validation, and traceable real-time visualization facilitate earlier detection of process errors, fact-based decision-making, and trust in data-driven methods. Thus, the first use case provides the technical and organizational prerequisites for ensuring that higher maturity stages, such as forecasting and AI-based optimization, are robust and low-risk.<\/p>\n\n\n\n The second use case (Fig. 4<\/strong>: Use Case 2 <\/strong>\u2013 Forecasting) encompasses predictive capabilities. The forecasting functions are performed on the basis on real measured process data from all fischertechnik\u00ae modules. Communication between the real system and the simulation model is bidirectional (I 1.1, S 3) with continuous updates (I 1.2, S 3). Data is interpreted context-sensitively, as the ontology-based knowledge model (for ontologies, see [13]) links system entities with process chains (I 1.4, S 3). <\/p>\n\n\n\n The simulation performs predictive calculations and makes them available in the order manager to support decision-making (I 2.3, S 2). Thus, the digital twin gains predictive cognitive capabilities (I 2.1, S 3) while still remaining an assistant without operational autonomy (I 2.2, S 2). Integration takes place at the system stage (I 3.1, S 2).<\/p>\n\n\n\n From the perspective of real factory environments, this use case addresses key challenges in predictive production control with the expansion stages (3-2-2-3). Forecasts of arrival times, bottleneck probabilities, and expected throughput rates create reliable decision spaces in which scheduling variants and alternative sequences can be evaluated using simulations before operational interventions are made. This corresponds to the implementation of a production or logistics control center, where decisions are prepared on the basis of data and finalized with human assistance.<\/p>\n\n\n\n The benefits result from improved human decision-making capabilities, reduced throughput times, and enhanced usability of simulation results for operational planning. At the same time, methodological limitations arise in the synchronization of simulation and real-world behavior, continuous model calibration, interoperability between the Asset Administration Shell, MQTT, and simulation environment, and in relation to robust validation schemes for forecasts. Consequently, the implementation also requires active change management. This use case bridges the gap between monitoring and autonomous decision-making.<\/p>\n\n\n\n In the third use case (Fig. 4<\/strong>: Use Case 3 <\/strong>\u2013 AI-based optimization), AI agents are trained in the simulation environment using reinforcement learning [14] (not yet technically implemented). The training environment is based on the real process chains of the fischertechnik\u00ae modules and is additionally connected to NVIDIA Omniverse. Communication is context-dependent and bidirectional (I 1.1, S 4) with near real-time data updates (I 1.2, S 4).<\/p>\n\n\n\n Digital information is interpreted cognitively (I 1.4, S 4) and the digital twin has prescriptive cognitive abilities (I 2.1, S 4). The simulation is used both for decision-making and serves as a training basis (I 2.3, S 4). The integration encompasses several domains within a factory (I 3.1, S 3). This results in learning, prescriptive digital twins that generate optimized proposals for action.<\/p>\n\n\n\n For technical implementation, simulation and AI agents are integrated to develop learning-based control strategies. First, the \u201cstate space<\/em>\u201d of the learning factory is modeled, which includes buffer levels, machine availability, and process states, among other things. On this basis, the agent selects suitable action from the \u201caction space<\/em>\u201d, e.g., the prioritization of orders or the selection of routing options. The \u201creward function<\/em>\u201d is formulated as a weighted objective function consisting of throughput maximization and energy efficiency. <\/p>\n\n\n\n The learned strategies are initially trained in the simulation environment and subsequently transferred to the real system for validation and later application. In real industrial applications, this expansion stage with the expansion stages (4-3-3-4) corresponds to highly digitized production systems with integrated models, standardized interface architectures, and established data quality processes. Through the connection with NVIDIA Omniverse, there is also the prospective possibility of coordinating multiple digital twins of different assets within a shared environment, enabling cross-system evaluation of strategies.<\/p>\n\n\n\n The benefits of this stage lie primarily in data-driven process innovation, continuous improvement, and increased efficiency in ongoing operations. At the same time, technical complexity and requirements for data management, validation, IT security, governance, and ethics increase significantly. Acceptance becomes a critical success factor, as AI decisions must remain traceable and controllable before trust in autonomous control decisions can be established. Thus, the third use case marks the transition from data-driven decision-making to data-driven action, initiating the transformative phase of digital twins.<\/p>\n\n\n\n![\"Figure](\"https:\/\/industry-science.com\/wp-content\/uploads\/2026\/04\/Gliem_I4S-26-2_Figure-1-1024x813.webp\")
Test and development environment for digital twins<\/h2>\n\n\n\n
![\"Figure](\"https:\/\/industry-science.com\/wp-content\/uploads\/2026\/04\/Gliem_I4S-26-2_Figure-2-e1775567722949-1024x758.webp\")
![\"Figure](\"https:\/\/industry-science.com\/wp-content\/uploads\/2026\/04\/Gliem_I4S-26-2_Figure-3-1024x683.webp\")
Discussion of selected use cases\u00a0<\/h2>\n\n\n\n
1. Real-time monitoring of the learning factory<\/h3>\n\n\n\n
![\"Figure](\"https:\/\/industry-science.com\/wp-content\/uploads\/2026\/04\/Gliem_I4S-26-2_Figure-4-1024x671.webp\")
2. Forecasting<\/h3>\n\n\n\n
3. AI-based optimization<\/h3>\n\n\n\n
Added value of learning factories<\/h2>\n\n\n\n