As software modeling repositories expand, the need for semantic retrieval of conceptual models becomes increasingly important for effective model management and reuse. In this paper, we propose a classification-based semantic retrieval scenario where the classifier acts as the core component enabling automated and intelligent model organization. Each metamodel, expressed in Ecore, is transformed into a semantic attributed graph that captures entities, attributes, associations, and multiplicities. To effectively exploit semantic information, we employ a fine-tuned BERT embedding as the primary representation method. Additionally, we use a TF-IDF-based embedding as a baseline to compare semantic and non-semantic approaches and to demonstrate the impact of semantic representations. These embeddings, together with node and edge labels, are used as input features to a graph neural network (GNN) that classifies conceptual models into 24 predefined categories. Since the models are represented as semantic graphs, our experiments confirm that GNNs significantly outperform traditional baseline classifiers. Specifically, the BERT-based GNN achieved the highest accuracy (around 96%), while the TF-IDF-based GNN also performed strongly (above 95%). In contrast, non-graph baseline models such as MLP (84.3%), SVM (78.6%), and C5.0 (85.4%) exhibited notably lower performance. These findings validate the proposed classification-based semantic retrieval scenario and highlight the advantages of graph-based semantic learning for automated, scalable, and semantically aware retrieval.
Biodiversity monitoring is concerned with keeping track of different species in an ecosystem over time, with respect to their abundance, distribution and diversity. Environmental digital twins used for biodiversity monitoring share characteristics with industrial digital twins, but face additional challenges in connecting data and models: Biodiversity data is often not livestreamed, interventions are slow and require human interaction, and the scientific knowledge about species and their habitats constantly evolves. Today, environmental digital twins offer little automation support, or any support to help scientists link species observations to assumptions about biodiversity. This paper presents an application of structural self-adaptation, originally developed in the industrial domain, to environmental digital twins. We show how structural self-adaptation enables to autonomously adapt monitored assumptions to changes in the available data sources, and further discuss how digital twins can adapt to changes in the domain knowledge. A first evaluation is given based on underwater cameras in the Oslo Fjord.
Digital Twin-enabled Systems (DTeS) are increasingly relevant in practice, particularly in the context of large, complex, and long-lived industrial facilities, buildings, and infrastructure. These systems are typically composed by orchestrating existing artefacts using a wide range of technologies and artefacts originating from diverse engineering disciplines. Due to continuous adaptation and evolution of the underlying systems, DTeS are subject to significant change over time, making both their development and long-term maintenance highly challenging.This paper presents TwinLang, a domain-specific language (DSL) designed to support the modelling and management of DTeS throughout their lifecycle. In TwinLang, Digital Twins can be modelled as code, enabling abstraction, reusability, and integration with software engineering workflows. The DSL supports model transformations as a central concept, which enables refinement from abstract representations to concrete implementations. These transformations also facilitate the projection of models into different views, supporting tasks such as deployment, monitoring, and visualisation. TwinLang is realised as an internal DSL in Scala, leveraging the language’s expressiveness and strong type system. To illustrate and evaluate our approach, we present the development of a Digital Twin for our AI Lab as a concrete use case.
Modern logistics systems aim to leverage digital technologies and may integrate autonomous components to increase efficiency and flexibility. In so-called mixed-fleet systems, human workers, manually operated machines, and autonomous machines collaboratively work towards a common goal. The subsystems are loosely coupled and can be reconfigured flexibly, leading to a change in behavior. Modeling this behavior requires flexible designs for model-driven systems engineering. Serviceoriented architectures can help focus on defining the expected behavior, independently of the involved actors. Additionally, eventbased communication mechanisms can decouple interactions between subsystems. This paper explores the use of SysML v2 for modeling a service-oriented architecture of mixed-fleet systems. Based on an available set of requirements, suitable SysML v2 modeling elements are identified that can describe services, events, and service choreographies. We use the described concepts to create a SysML v2 model of a mixed-fleet harbor logistics use case. Based on this model, we demonstrate how business processes can be composed of reusable services and how requirements and verification can be integrated to ensure correctness of behavior. The results show that SysML v2 meets key requirements for service-oriented architectures and enables separating service definitions, actors, and verification elements. Reusable modeling patterns were applied to support scalability, and enable traceability within the model across actors and services. Furthermore, domain-specific constraints and requirements were composed into modeling elements using formal mechanisms to ensure that they are not only documented, but actively connected to the model.
While Digital Twins are traditionally associated with domains of human-made systems, Earth Science Digital Twins (ESDTs) have been rising in popularity. An ESDT represents a twin of a natural entity or ecosystem at a local, regional, or global scale, such as a river, a forest, an ocean, or the Earth itself. In this work, we conduct a systematic literature review of the existing ESDT systems. We identify the main ESDT domains as hydrology, geology, ecology, Earth, and atmosphere. We analyse the purposes of ESDTs, describe the software development tools that are often employed in ESDT development, and discuss the challenges and limitations that are specific to ESDTs. We conclude that data collection and computational limitations are among the most prominent, and maintaining the real-time mirror of the physical entity remains a challenge.
