The landscape of modern computing systems is shifting towards architectures built by combining available services under the “everything as a service” paradigm. These architectures are deployed on distributed cloud-edge infrastructures, aiming to provide innovative services to a wide range of users. However, it is crucial for these systems to address environmental sustainability concerns. This poses challenges in operating such systems in open, dynamic, and uncertain environments while minimizing their energy consumption. To tackle these challenges, we propose a decentralized service assembly approach that ensures the assembly is energetically self-sustainable by relying on locally harvested and stored energy. In our contribution, we introduce a general service selection template that enables the derivation of different selection policies. These policies guide the construction and maintenance of the service assembly. To evaluate their effectiveness in meeting the sustainability requirements, we conduct a comprehensive set of simulation experiments, providing valuable insights.

As software systems grow in complexity, traditional development methods increasingly struggle to meet demands for scalability, agility, and efficiency. In this context, software architecture plays a central role in maintaining system quality and long-term adaptability. However, SA development faces persistent challenges such as bridging loosely defined requirements with design, managing architectural knowledge across distributed teams, and preserving structural integrity in agile settings. At Volvo Construction Equipment, Artificial Intelligence is being investigated as a means to support and enhance architectural work. Specifically, AI is applied to automate labour-intensive tasks, support design decision-making, and document architectural rationale to mitigate erosion. This paper presents four use cases at Volvo Construction Equipment, including generating Architecture Decision Records, mapping requirements to modules, supporting component design, and evaluating generative AI for module classification. These efforts illustrate AI’s potential to act as a co-pilot in the software architecture process—improving consistency, traceability, and decision support. By sharing our approach and early experiences, this paper contributes to the broader discourse on integrating AI into industrial software architecture practice.

Context: Ensuring high levels of dependability in modern computer-based systems has become increasingly challenging due to their complexity. Although systems are validated at design time, their behavior can be different at runtime, possibly showing control-flow anomalies due to "unknown unknowns".

Objective: We aim to detect control-flow anomalies through software monitoring, which verifies runtime behavior by logging software execution and detecting deviations from expected control flow.

Methods: We propose a methodology to develop software monitors for control-flow anomaly detection through Large Language Models (LLMs) and conformance checking. The methodology builds on existing software development practices to maintain traditional V&V while providing an additional level of robustness and trustworthiness. It leverages LLMs to link design-time models and implementation code, automating source-code instrumentation. The resulting event logs are analyzed via conformance checking, an explainable and effective technique for control-flow anomaly detection.

Results: We test the methodology on a case-study scenario from the European Railway Traffic Management System/European Train Control System (ERTMS/ETCS), which is a railway standard for modern interoperable railways. The results obtained from the ERTMS/ETCS case study demonstrate that LLM-based source-code instrumentation can achieve up to 82.849% control-flow coverage of the reference design-time process model, while the subsequent conformance checking-based anomaly detection reaches a peak performance of 95.957% F1-score and 93.669% AUC.

Conclusion: Incorporating domain-specific knowledge to guide LLMs in source-code instrumentation significantly allowed obtaining reliable and quality software logs and enabled effective control-flow anomaly detection through conformance checking.

Software-as-a-Service solutions are increasingly being adopted when developing software applications, as they are scalable, cost-effective, and facilitate rapid deployment while providing high availability and flexibility. However, the impact of Software-as-a-Service in terms of carbon emissions is not yet adequately addressed as a design concern, and most of the existing efforts revolve around measuring and containing the carbon impact after the deployment. Our work proposes a model-driven reasoning framework that integrates UML-based software architecture modeling with carbon-aware concerns. Architectural elements are supplemented with sustainability and performance properties of interest through a dedicated Domain Specific Language; then, a model-driven transformation generates a simulation model to evaluate multiple architectural designs according to their Software Carbon Intensity and performance metrics. The results guide decision-making by assessing and comparing the trade-offs between performance and carbon intensity for the analyzed designs. In this way, the reasoning framework provides an automated, tool-supported approach to designing environmentally responsible Software-as-a-Service applications.

Cyber-physical systems (CPS) are challenging to control due to the complex uncertainties arising from physical and virtual sources. Enhancing CPS with self-adaptation is beneficial in addressing these uncertainties. While reactive adaptation often struggles with reliability, proactive adaptation could be more advantageous by preparing systems to make informed decisions, considering the consequences of changes before they occur. CPS and their execution environment usually exhibit timevarying or non-linear dynamics, which are more complex to predict than linear systems, while recent proposals of proactive self-adaptation methods have focused on linear systems. This work bridges this gap by proposing a method for Proactive self-Adaptation for Nonlinear Cyber-physical Systems (PANCS). PANCS is developed through a ground vehicle running example, leveraging MAPE-K loop, and its strengths and limitations are discussed.

Business process management deals with the administration of the chains of events, activities, and decisions that add value to an organization. The ever-growing complexity and inner uncertainty of business-related operations call for a paradigm shift toward new technologies, such as digital twins, to coordinate and optimize the processes belonging to an organization. In this paper, we propose an extended version ofthe business process lifecycle that includes a Digital Twin of the Organization, in charge of consistently ensuring the process’s alignment with the required performance goals and facilitating recovery from failures. We also contribute to defining the model-driven components that transform the business process model, in BPMN notation, into a compatible process representation for the digital twin, enabling simulation capabilities and scenario analyses. Through a manufacturing use case, we demonstrate how the digital twin optimizes processes pre-implementation and facilitates rapid recovery and reconfiguration following critica lfailures, underscoring its potential to improve efficiency and reliability in process management.

As simulation applications become vital for understanding and predicting complex systems, analyzing data from repeated simulation runs is essential to gauge model uncertainty and identify optimal parameter settings. This paper presents a visualization tool for analyzing time series ensemble data generated by discrete-event simulations, focusing on feature importance within clustering results. The tool combines dimensionality reduction, clustering, and SHapley Additive exPlanations (SHAP) to highlight influential features and identify trends within clustered simulation data, advancing previous approaches focusing solely on visualization or clustering without analyzing specific feature contributions. By analyzing a manufacturing use case, we show how the visualization supports decision-makers by depicting the main features driving cluster formation and displaying time intervals critical to characterizing distinct system behaviors.

Digital Twin of the Organization (DTO) is a relatively new concept that emerged to help managers have a full understanding of their organization and realize their objectives. Indeed, DTO enables connecting all the elements of an organization virtually by providing monitoring, optimization, prediction, and other capabilities through continuous simulations. Creating a flexible and evolvable DTO that covers and supports the organization's business strategies is a complex and time-consuming task that requires engineering best practices. In this context, this paper presents and evaluates the EA Blueprint Pattern, which serves as an architectural reference for the development of a DTO by allowing for mapping well-known Enterprise Architecture concepts into software components defining the DTO software architecture. The evaluation is carried on by showing how to use the pattern for creating the DTO for a given organization. Then, a thorough discussion is conducted to analyze how the developed DTO should evolve to deal with vertical and horizontal integration. The lessons learned highlight the strengths and weaknesses along with practical implications for organizations that are eager to develop their DTO according to the EA Blueprint Pattern.

Digitization of the manufacturing and assembly sector is important to set up Industry 4.0. In this process, one of the key factors is the channel of sharing and distributing information on the shop floor. This study highlights the implementation of digital work instructions in the manufacturing and assembly sectors and finds the benefits that it could bring to the industry. The study was conducted in a large production plant with over five hundred workers in Malaysia and was carried out for almost a year. Whereas, most of the existing studies have been conducted in a controlled environment with a group of inexperienced workers in manufacturing and assembly tasks. In this article, the benefits and challenges of digital work instructions are studied over paper-based textual representation of assembly instructions. The study was conducted among groups of people with different roles, such as electrical assembly, mechanical assembly, and final quality check. The qualitative analysis is carried out based on the survey conducted among operators with different roles. Results show that digitalization eases the work for the quality inspection group. In contrast, people with other tasks are either neutral or find it more difficult to work with digitalized versions over paper-based instructions. In addition to this, some data-driven facts are presented, which help in improving the plant operations. This includes recording material shortages, optimizing working hours, and having real-time updates on production status which leads to effective production planning. At last, with the collected information, manufacturing plants can also optimize power utilization that impacts the environment in a positive direction.

Architecture refactoring is a big challenge and requires thorough analysis and labor-intensive, error-prone activities to restructure functionalities from a legacy architecture to a new intended one. Indeed, source code should be adapted to match the new structure. In this context, automatically mapping source code to the intended architecture would significantly reduce manual work and prevent technical debt. To this end, in this paper, we aim to map methods to architectural modules solely defined by textual descriptions, i.e., formulated as a machine learning text classification problem. Methods are mapped into modules using different approaches. We apply the proposed approach to an open-source software system, results show that vectorizing text and code using large language models outperforms other modern methods. The different applied machine learning classifiers perform comparably well, where the best attain accuracy of around 40% and F1-score of around 30%.