Modular robots are cyber-physical systems composed of elements similar in nature that collaborate for a common global goal. This paper reports on a formal modelling of modular robots using Markov Decision Processes (MDPs) supported by the PRISM tools. The novelty of our modelling approach consists in incorporating empirical data and feedback from energy-dependent components collected thanks to an instrumentation of real modular robots called Blinky Blocks. While using PRISM modules for specifying individual robots or motifs composed of Blinky Blocks in their environment, the collected data are used for probabilistic transitions and non-deterministic choice, as well as for rewards to estimate energy availability that impacts fallback behaviours and actuation success. Afterwards, PRISM allows performing model-based analysis of safety, recovery, and energy-boundness under non-deterministic choice of the environment, and a model-based simulation on scenarios of interest. Comparing the obtained simulation results to those observed on real individual robots and their motifs provides useful insights for the developers, such as anomalous consumption patterns or inter-module variability. The designed trustworthy models can then be used for designing and validating adaptation policies for modular robots.
The terms "consistency" and its counterpart, "inconsistency", are omnipresent in model-driven software engineering. Particularly during model evolution, consistency assurance is a central concern. Technically, consistency can be defined from various viewpoints. However, it is unclear how these definitions of consistency align with practitioners' mental models. Understanding how practitioners assess and respond to different kinds of consistency problems in their own minds is required for realizing successful consistency assurance. An important aspect towards this goal is the investigation of whether there is a shared mental model of consistency among an evenly experienced group of practitioners. The current state of the art gives no answer to this. In this work, we present a model dataset of 62 consistency problems based on simple, compositional models. For this dataset, we collected 452 consistency ratings from a student group and one rating per problem from an expert group. The results show that there is no common understanding, i.e., there is no shared mental model of consistency within a group or between groups. Still, common patterns in the ratings are observable, such as a preference for rating consistency on a fine-grained scale rather than as a boolean property, or that syntactic contradictions are much less relevant to ratings than subjective semantic interpretations. All developed tools, models, and collected data are openly available for future research.
Modeling is inherently a collaborative activity, as it provides stakeholders with a common language for structuring domain-specific knowledge. This makes support for diverse collaboration scenarios a central requirement for modeling tools; however, most existing tools are designed primarily for either real-time or asynchronous collaboration. An emerging workflow that has so far received limited attention, yet is widely valued by practitioners, is hybrid collaborative modeling. In this setting, stakeholders alternate between synchronous and asynchronous work, with real-time subgroups coordinating locally while overall progress is maintained asynchronously across time zones and organizations. This article lays the foundations for the architectural requirements required to support hybrid collaborative modeling. We first introduce a classification framework for collaboration architectures and use it to compare existing collaborative modeling tools, highlighting their limited support for hybrid workflows. Building on this analysis, we present a novel architectural approach: Local-First Collaborative Modeling. Finally, we present a reference implementation approach based on replicated data types, intended to demonstrate feasibility and to guide future implementations.
Unified Modeling Language (UML)-style class diagrams are widely used for conceptual and system modelling because they provide compact and familiar visual representations of domain structure. However, the built-in reasoning support provided by UML tools is generally modest. The Web Ontology Language (OWL), in contrast, is optimised for supporting welldefined logical inference but is less well suited to authoring compact, engineering-oriented model representations comparable to UML-style diagrams. To address these respective weaknesses, we propose endogenous OWL overlays as a lightweight bridge from an authoritative UML-style model to standards-based OWL reasoning services. An overlay is a rule-derived OWL view computed over a triple/fact-based representation of the authoritative model snapshot used by the overlay rules. The host concrete syntax is LML, a UML-style multi-level modelling notation, and the overlay targets selected OWL-style services rather than full UML coverage. Because the overlay is recomputed on demand and consumed by standard OWL tools, the source model remains the source of truth and the derived view stays directly traceable to the modelling declarations that produced it. The approach handles both the model's declared structure and selected modelling idioms that arise naturally in UML-style modelling, using a small set of local OWL axioms only where needed. Using four representative scenarios, we demonstrate subproperty closure, role-membership inference, derived associations, and the detection of logically unsatisfiable classes without maintaining a separately synchronised OWL artefact.
The development of complex and safety-critical systems can benefit from robust and integrated MBSE processes. However, the ecosystem of MBSE methods, languages, and tools is fast-evolving and fragmented. MBSE processes are thus often practiced in isolation. This impedes integration, traceability, and reproducibility. This paper presents a process model for the formal description of transformation processes. We further specify an architecture for an execution environment that uses the process model as input. Together, the process model and the execution environment intend to enable automated, versioned, and traceable MBSE processes. The approach is being evaluated in the cross-domain SQuIRRL project, which concerns IT, railway, automotive, and aviation. The evaluation confirms dependency-consistent execution and reproducibility of heterogeneous MBSE transformation processes, thereby improving comparability and collaboration across domains. The metamodels, specification, and reference architecture are available as open source.
Engineering teams working on complex socio-technical systems frequently rely on tacit knowledge when addressing stakeholderspecific needs, hindering systematic reasoning about design choices. In this paper, we report on a qualitative case study within the EMPOWRD project to observe how best-practice knowledge can be captured using justification diagrams, a safety-inspired argumentation formalism. We explore the organization of these practices as a configurable product line using feature models to manage their variability. By applying well-defined composition operators, we demonstrate how assembling models from selected practices can automate the construction of coherent argumentation structures. Our observations within the EMPOWRD project, specifically regarding interactions with older adults, illustrate how this integrated modelling approach enables the systematic reuse and justification of practices across different experimental scenarios.
Digital Twin (DT) technology creates layered digital representations of physical system, asset or process, denoted as actual twin (AT). DT integrates heterogeneous data, models and semantic technologies, enabling monitoring, simulation, prediction, and optimization to enhance decision-making and efficiency, ensuring an accurate and dynamic representation of their AT. The rise of the inter net of things (IoT) and cyber-physical systems (CPSs) has generated massive volumes of data, which require contextual information about their source and meaning to extract actionable insights. DTs address this need by unifying system data and behavior into coherent, multi-layered heterogeneous models. However, as real-world conditions and AT evolve, semantic drift emerges, leading to a progressive divergence between the DT and its AT, reducing reliability and effectiveness. Semantic drift often arises from inconsistent evolution and misalignment among heterogeneous models within the multi-layered paradigm, causing semantic mismatches, inconsistencies, and synchronization problems. Correcting semantic drift involves structural and behavioral adaptations across heterogeneous models and current methods typically involve manual updates, which are time-consuming, error-prone, and risk compromising data integrity. In this paper, we overcome these challenges and propose a systematic approach to manage semantic drift in multi-layered, model-driven DTs, supporting structural adaptation across heterogeneous models to ensure semantic consistency. We adopt a three-step approach: (i) identification, (ii) evaluation or measurement, and (iii) propagation, to manage semantic drift systematically. First, we identify and assess variants of semantic drift arising from data. We then propagate the changes across heterogeneous models to effectively correct semantic drift. Finally, we employ a case-based generalization strategy to illustrate our approach, showing how insights derived from specific use case support semantic drift management in broader contexts.
Model-Based Systems Engineering (MBSE) uses models for designing, validating and synthesizing complex systems. Different models for large systems are synchronized via transformations, but current Model Management (MoM) approaches lack mechanisms to control or audit changes propagation. Managing such changes is very challenging, especially when the propagated modifications are unauthorized or rejected by downstream teams. This paper introduces Senate, the first fine-grained change propagation control framework based on declarative change policies. Senate provides a declarative domain-specific language to specify policies to provide consistent change propagation in scenarios where changes may be rejected by models. These policies govern which users are permitted to create, modify or delete model fragments based on the affected elements, their types, or if they match given patterns. By enforcing these policies during consistency management, Senate prevents unauthorized or undesired changes from being propagated, thus reducing the delay between introduction of the inconsistency and its discovery during review. We implement and evaluate a prototype of Senate which shows the feasibility and scalability of the approach in an non-intrusive way, without requiring modifications to the models or the consistency management approaches required. We evaluate the design to show that declarative change policies are generalizable, non-intrusive to existing tools, and scalable with respect to the size of models and number of changes, and thus can be integrated into existing consistency management workflows in multi-model collaborative MBSE scenarios.
Architectural models are rich and useful representations of systems' designs, which can be reused and adapted to design new systems. The many publicly available architectural models (e.g., on GitHub) offer the opportunity to reuse domain knowledge to bootstrap the design of new systems, avoiding common mistakes and promoting the reuse of good principles and practices. Discovering related architectures to distill domain knowledge, however, is challenging, and this body of public knowledge often remains underexploited. To address this need, this paper investigates the application of structural-semantic clustering strategies to automatically cluster sets of architectural models. We consider multiple structural model clustering strategies, which represent models as graphs, and multiple semantic model clustering strategies, utilizing embeddings to compare the names and contents of models. To assess the considered strategies, we built a curated dataset of 1,202 manually clustered models collected from GitHub, which is the largest ground truth of clustered architectural models, to the best of our knowledge. The proposed empirical study thoroughly analyses the effectiveness of clustering strategies, since they are the instrument that can enable the discovery of implicit architectural knowledge available in models.
One of the initial objectives of MDE is to obtain a complete software application from models through code generation. A standard such as fUML (Semantics of a Foundational Subset for Executable UML Models) enables to add abstract code on UML and DSL (Domain-Specific Language) models, allowing models to contain the entire definition of an application, from its architecture to its business algorithms. However, this approach raises the issue of the ability to reuse existing code written in a traditional programming language and of its adoption by software engineers. In this paper, we propose an intermediate vision where an application is composed of an executable model defining its behavior and traditional code defining its business operations. We propose a generic solution for weaving business operation signatures onto elements of an executable DSL (xDSL), allowing these operations to be executed automatically during model interpretation. We apply our solution to the implementation of a drone simulator software: a xDSL is used to define the simulation architecture model and the simulation functions are implemented in Java. We show that the approach facilitates the configuration and the setup of the simulator.
In the view-based development of cyber-physical systems, developers with different roles and tasks use specific views to access models of the system. Using projective approaches, these views are derived from consistent, underlying models. The views and transformations from and to the underlying models are defined using domain-specific languages. The problem is that existing view definition languages lack either an easy-to-use syntax or sufficiently expressive transformation operators or incremental model-view transformations. In this paper, we propose a syntax for the view definition language NeoJoin based on the concepts of model queries and meta-model descriptions. We evaluate the usability of the proposed syntax with a user study and apply it to an existing case in the domain of smart electricity grids. With the proposed syntax, we aim to provide an easy-to-use, yet universally applicable language for the definition of views in cyber-physical systems development.
Dynamic processes such as conflict escalation, disease progression, or policy change often involve entities transitioning through structured states over time. Capturing these evolving trajectories in a way that is both formally rigorous and adaptable to unstructured data remains a challenge. In this paper, we present a hybrid approach that integrates formal modeling techniques with large language models (LLMs) to support structured reasoning over dynamic phenomena. Although LLMs offer powerful capabilities for information extraction, their use as standalone analytic tools in sensitive domains is limited, as they are prone to generating unreliable outputs and lack grounding in domain theories. To address this, we propose a framework that combines LLM-based event extraction with ontology-grounded knowledge graph construction, using constraint-preserving graph transformations to ensure semantic and structural validity. We demonstrate the framework in the conflict domain by constructing an ontology based on Glasl's conflict escalation model and using it to guide structured event modeling and trajectory analysis. We implement a prototype system demonstrating the framework in the conflict domain, and present experimental results on an event dataset from ACLED.
The Digital Product Passport (DPP) is an essential conceptual technology enabling circular manufacturing (CM). It provides a mechanism for stakeholders in CM to manage the entire lifecycle of individual products, from material sourcing to their end-of-life, while also promoting reduction, reuse, and recycling strategies. To implement the DPP, many industrial organizations rely on the Asset Administration Shell (AAS) and the ECLASS vocabulary. On the one hand, the combination of these two standards is robust, as they are well-defined, easy to use, and supported by a broad industrial community. On the other hand, many other partners in both public and private sectors are exploring alternative solutions that are open, more distributed, and knowledge-centric, such as ontology-based approaches. To harmonize these two perspectives, this paper proposes a new ontology, called Product Digital Twin Ontology (PDTO), and a method that uses PDTO as the pivot for converting between AAS-based and ontology-based DPP representations.
Artificial Intelligence in general and Machine Learning in particular is a very dynamic field, with evolving technologies and practices. Key decisions, from dataset preparation methods to model architecture and evaluation metrics, are made in an exploratory way, diverging from standard software engineering practices and guided by the data scientist's empirical knowledge and experience. In this context, (i) some practices work, while others do not; (ii) some unexpected approaches perform better than the regular ones; (iii) some anti-patterns are accidentally used, leading to model contamination and biases. Developers and data scientists have to work on code artifacts (e.g., Jupyter notebooks) to identify key differences across multiple variations of the same machine learning pipelines, which is confusing and error-prone because it is only syntactic. In this paper, we defend a model-driven approach that reifies semantic information about machine learning pipelines to improve their understanding. Based on this metamodel, which captures essential steps in a given pipeline and links them to code artifacts, we define a pattern-matching language that supports data scientists in exploring corpora of machine learning artifacts. We validate the approach by identifying real-world use cases in collaboration with data scientists and applying them to the qualitative analysis of 105 Kaggle notebooks (a popular competition platform where participants submit pipelines to solve similar tasks). This work opens the door to transferring program understanding techniques to machine learning while accounting for its intrinsic exploratory nature. By relying on explicit models and a dedicated pattern language, we provide a foundation that supports systematic analysis of ML pipelines-such as pipeline comparison, practices identification, and anti-pattern detection-while remaining robust to the evolution of libraries, frameworks, and implementation technologies.
In Model-Driven Engineering (MDE), metamodeling is a crucial activity and is often the starting point of a full MDE pipeline. A central aspect of this task is extracting domain knowledge from textual documentation and establishing the necessary classes, relationships, and constraints that will be later used to formalize the metamodel. This early stage is known to be demanding, error-prone, and influenced by individual modelers' expertise and bias. Recent advancements in the use of Large Language Models (LLMs), have stimulated research on interpreting textual information to generate models and metamodels. However, there is still limited work exploring the potential of LLM-powered agents, particularly LLM-powered Multi-Agent Systems (LLM-MAS), to support the metamodeling process in a structured manner. In this paper, we present an approach that leverages an LLM-MAS to assist in documentation-driven metamodeling. The proposed approach decomposes the task into multiple specialized agents responsible for activities such as domain analysis, terminology identification, normalization and deduplication, and textual serialization into a PlantUML class diagram. The system operates without human intervention and produces intermediate artifacts that support traceability and inspection. We report on the development of this LLM-MAS and its application in a case study involving the extraction of a draft metamodel from a set of agent framework documentations. We provide an exploratory qualitative evaluation focusing on the feasibility, stability, and structural plausibility of the generated artifacts. The results indicate that LLM-MAS can consistently produce structurally plausible model-like artifacts that may assist modelers in the early stages of metamodel creation. Rather than targeting full automation, the approach positions LLM-MAS as a modeling aid that supports early abstraction and helps modelers initiate metamodel development more systematically.
Maintaining large-scale legacy software systems is complex and time-consuming. Practitioners must integrate diverse artifacts and data, produced by various specialized tools, in order to gain a better understanding of the system and challenges at hand. This process is typically manual and error-prone. In this paper, we propose a solution to support maintenance activities by incrementally merging heterogeneous software models. We build on the existing framework, Unified Data Source (UDS), based on the Single Source of Truth (SST) paradigm. The framework consists of three layers: Probe, SST, and Visualization. The lightweight Probe layer extracts partial models from heterogeneous artifacts and forwards them to the SST layer, which acts as a composition engine that consistently merges the partial models into a unified and coherent representation. The Visualization layer then enables the creation of customized, task-specific model views derived from this unified model. The feasibility and practical value of the method are demonstrated through a systematic, tool-supported approach that applies proven MDE techniques to a concrete system. We showcase this methodology through two real-world use cases: performance troubleshooting and change impact analysis. Both use cases are successfully executed on Spring PetClinic, a widely-used demo project, producing task-specific model views without reconstructing the entire system. These results confirm that task-driven, incremental merging of heterogeneous models is practically achievable, yielding actionable views for practitioners and supporting them in making smarter decisions.
With the rise of Industry 4.0 driven by the integration of Cyber-Physical Systems (CPS) and the Internet of Things (IoT), the use of Digital Twins (DTs) has significantly increased over the past decade, as they provide detailed insights and support well-informed decision-making. However, the lack of standardized methodologies, in addition to the time and resources involved for building them remains an important challenge. Building on the idea that engineering models of the physical twin (PT) are often available, we propose a tool-supported framework that automates the derivation of DTs by leveraging existing structural and behavioral models of the PT and extending them with additional models to build a comprehensive DT. To demonstrate the feasibility of our approach, we applied it to four different use cases, in which we automatically derived DT instances from (1) models of their PT, (2) configuration of our generic framework and (3) minimal ad hoc additional development for connecting the DT to the PT. These experiments illustrate the applicability of our framework for building DTs in contexts that satisfy our assumptions and requirements. By simply configuring the framework, we are able to derive a DT aligned with its operational purpose.
