ABSTRACT. The first part of this tutorial will cover the key differences between EN 50716 and EN 50128. Emphasis will be placed on advancements in software development methodologies, including the recognition of Model-based techniques, Agile development, and an expanded focus on formal methods. Additionally, we will touch upon AI and Machine Learning, highlighting how the guidance aligns with the rail industry's careful, safety-first approach while still fostering innovation.

ABSTRACT. The second part of the tutorial will use two representative rail systems, Slip-Slide Detection and Interlocking, to illustrate how EN 50716 can be implemented using SCADE. This part will start with a quick introduction to understand the fundamentals of SCADE and its applications in rail systems and will cover agile Model-Based Design, integration in DevOps workflows and leveraging formal verification. Each topic will be illustrated on the examples and the compliance with the standard clearly explained.

ABSTRACT. Since its introduction as an academic project at the RSSRAIL Conference 2017, the CLEARSY Safety Platform has undergone significant advancements, achieving SIL4 certification and successful deployment in operational environments. This session will provide a comprehensive overview of the journey from an academic platform to a robust industrial solution. We will delve into the technical enhancements, certification challenges, and real-world applications that have shaped the CLEARSY Safety Platform into a reliable tool for the railway industry. Attendees will gain practical insights into the platform's capabilities, its role in enhancing railway safety, and the lessons learned from its implementation in industrial settings.

ABSTRACT. Relay-based Railway Interlocking Systems (RIS) control railway components like signals and turnouts safely but are still analysed manually through relay diagrams, a process prone to errors. Previous works proposed formal methods and industrial tools for RIS analysis, but digitalizing and automatically transforming these diagrams into formal specifications remained challenging. This paper presents a proof of concept using an existing Multimodal Language Model to analyse relay diagrams and automatically generate formal specifications in propositional logic. Our approach adapts a prompt based on prior methodologies to orient the model expected outcomes, which are then applied to new diagrams. The results confirm that automatic formalization is feasible and accessible. This work opens promising perspectives for further improving correctness through dedicated prompt engineering or fine-tuning, advancing automation in the formal verification of railway systems.

ABSTRACT. Each planning phase of ETCS-compliant railway tracks at Deutsche Bahn (DB) prescribes a concluding review, now performed by manually inspecting printed diagrams and tables. This is time-consuming and bears the risk to overlook critical mistakes. We present a concept and a tool for fully automated formal verification of railway plannings against ETCS planning rules. The approach is based on a modular translation of track models, as well as planning rules, to the SMT-LIB language understood by Satisfiability Modulo Theories (SMT) solvers, which are used as a backend. Track models are assumed to be available in the standardized object-oriented PlanPro format and are automatically translated to SMT-LIB constraints. The planning rules themselves are given in natural language in rule books and cannot be translated automatically. Instead, we provide a translation schema that lets a planning engineer render planning rules almost one-to-one as first-order formulas. No specific knowledge of logic or SMT solver internals is required to perform this task, and it is sufficient to do it once and for all for each planning rule. Subsequent verification of a track model against a planning rule is fully automatic. Deviations are visually highlighted for manual inspection. To this end, we integrated rule verification with an existing track visualization tool into a GUI. Our approach was evaluated with real DB infrastructure data, showing that it is easy to use and sufficiently powerful to be integrated into existing planning workflows.

ABSTRACT. This paper was originally published in MDPI MAKE in 2025. This version is presented at RSSRail 2025. Autonomous trains require reliable and accurate environmental perception to take over safety-critical tasks from the driver. This paper investigates the application of N-version architectures to rail track detection using Deep Neural Networks (DNNs) as a means to improve the safety of machine learning (ML)-enabled perception systems. We combine three different neural network architectures (WCID, VGG16-UNet, MobileNet–SegNet) in a 3M1I configuration. In this configuration, we apply two fusion methods to increase accuracy and to enable error detection: Maximum Confidence Voting (MCV), combining the DNN predictions at the image level, and Pixel Majority Voting (PMV), a novel approach for combining the predictions at the pixel level. In addition, we implement a new method for evaluating and combining prediction confidence values in the N-version architecture during runtime. We adjust the overall prediction confidence according to the conformity of all individual predictions, which is not possible with an individual network. Our results show that the N-version architecture not only enables a detection of erroneous predictions by utilizing those adjusted confidence values, but it can also partially improve the predictions by using the PMV combination algorithm. This work emphasizes the importance of model diversity and appropriate thresholds for an accurate assessment of prediction safety. These approaches can significantly improve the practical applicability of ML-based systems in safety-critical domains such as rail transportation.

ABSTRACT. Machine learning (ML) offers promising capabilities for predicting rail infrastructure failure and enabling a shift from diagnostic to prognostic railway maintenance. However, the real-world adoption of high-performing ML models in safety-critical domains such as railway systems hinges on their trustworthiness, particularly their interpretability and transparency. This study, based on a case study in track geometry management, explores the trade-off between ac-curacy and interpretability in predicting track alignment failures by comparing six ML classifiers: Logistic Regression, Random Forest, Gradient Boosting, XGBoost, Support Vector Machine (SVM), and a Neural Network (NN). The models were trained on railway defect datasets using features such as operating speed, train traffic, total gross tonnage, and defect length. Performance was evaluated using recall as the primary metric, given the high cost of false negatives in rail safety contexts. Results showed that SVM and NN models achieved the highest recall, but at the cost of lower interpretability. To address this, post hoc Explainable AI (XAI) techniques, includ-ing SHAP and LIME were applied. These methods collectively enhance both local and global interpretability and support model transparency, stakeholder trust, and the bridging of the gap between predictive performance and decision-making needs. While XAI is increasingly applied in other sectors, its use in asset management and particularly railway predictive maintenance remains limited. This work fills that gap by demonstrating how XAI can foster more informed and confident adoption of ML models in rail infrastructure management. These explainability techniques help domain experts and end users understand why a model produced a specific result and what key factors influenced that decision, while also supporting data scientists and develop-ers in refining model performance.

ABSTRACT. Rail industry increasingly uses software tools, for example, to design scheme plans. However, how can such tools be tested? The manual design of rail artifacts such as scheme plans is laborious, costly, and in itself an error-prone process. Thus, there is a demand in rail industry for synthetic test data.

In our paper, we present a technique and a tool to generate artificial scheme plans. To this end, we utilise Genetic Algorithms with controlled fault injection. Which faults ought to be injected is steered by decision tables, which allow for a systematic case analysis for different design rules. The generated scheme plans can then be supplied to rail design tools such as the Siemens’ data checker, and we can evaluate if the rail design tool find the injected faults and, possibly, report faults that are not there.

Generation, test execution, and also test evaluation can be automatised. However, fault analysis is a manual process, requiring rail engineers to determine which scenarios to consider. When decision tables are given, our approach leads to a automatic quality evaluation of checking tools for scheme plans.

Our paper considers considers all ingredienst for such a testing approach for the simple case of, what we call it, linear scheme plans. Already for this simple subclass of scheme plans, our tool generates scheme plans that come as a suprise to industrial partner: Our tool created designs they had not thought of and therefore were challenging their tools.