8th ML4CPS 2025 – Machine Learning for Cyber-Physical Systems
Type
Conference
Acronym
ML4CPS
Start Date
March 6, 2025
End Date
March 7, 2025
Location
Berlin
Country
Germany
9 results
Settings
Now showing 1 - 9 of 9
- PublicationOpen AccessWorkshop report: Learning approaches for hybrid dynamical systems(Universitätsbibliothek der HSU/UniBw H, 2025-05-27)
;Plambeck, Swantje; ;Schmidt, Maximilian ;Balzereit, Kaja ;Bracht, Aaron ;Redeker, Magnus ;Arabizadeh, Negar; ;Eickmeier, Jens; Fey, GoerschwinThis report summarizes the workshop on “Learning Approaches for Hybrid Dynamical Systems”, held at the 2025 Conference on Machine Learning for Cyber-Physical Systems (ML4CPS). The workshop aimed to strengthen collaboration and foster exchange between institutions engaged in research on model learning methods for hybrid CPSs. The participating research groups approach the topic from diverse perspectives, for example, from an application perspective, from a tool perspective, or from a fundamental and formal perspective. Accordingly, this paper synthesizes the discussions from the workshop and presents an overview of key perspectives on several central topics, including the taxonomy of hybrid systems, current learning paradigms and techniques, and particularly representative use cases. - PublicationOpen AccessHow to quantify the maturity of production processes(Universitätsbibliothek der HSU/UniBw H, 2025-05-27)
;Arabizadeh, Negar ;Pfrommer, JuliusBeyerer, JürgenQuantifying the properties of interest is a basic need in all fields of science and engineering. We propose formal definitions to quantify the maturity of production processes. The definitions are inspired by the concepts of controllability and observability from control theory. But instead of a binary classification of a system as controllable/uncontrollable and observable/unobservable, we consider a probability distribution for the remaining error. From the proposed maturity quantification, the remaining potential for optimizations can be evaluated. This is shown for the example of an electric arc furnace (EAF), for which high-fidelity simulation models are available. - PublicationOpen AccessMachine learning for cyber physical systems(Universitätsbibliothek der HSU/UniBw H, 2025-05-27)
; ;Beyerer, JürgenKampker, AchimCyber Physical Systems are characterized by their ability to adapt and learn from their environment. Applications include advanced condition monitoring, predictive maintenance, diagnosis tasks, and many other areas. All these applications have in common that Machine Learning and Artificial Intelligence are the key technologies. However, applying ML and AI to CPS poses challenges such as limited data, less understood algorithms, and the need for high algorithm reliability. These topics were a focal point at the 8th ML4CPS–Machine Learning for Cyber-Physical Systems Conference in Berlin, held from March 6th to 7th, where industry and research experts discussed current advancements and new developments. - PublicationOpen AccessAn illumination based backdoor attack against crack detection systems in laser beam welding(Universitätsbibliothek der HSU/UniBw H, 2025-05-27)
;Huo, Wenjie ;Schmies, Lennart ;Gumenyuk, Andrey ;Rethmeier, MichaelWolter, KatinkaDeep neural networks (DNNs) have been wildly used in engineering and have achieved state-of-the-art performance in prediction and measurement tasks. A solidification crack is a serious fault during laser beam welding and it has been proven to be successfully detected using DNNs. Recently, research on the security of DNNs is receiving increasing attention because it is necessary to explore the reliability of DNNs to avoid potential security risks. The backdoor attack is a serious threat, where attackers aim to inject an inconspicuous pattern referred to as trigger into a small portion of training data, resulting in incorrect predictions in the reference phase whenever the input contains the trigger. In this work, we first generate experimental data containing actual cracks in the welding laboratory for training a crack detection model. Then, targeting this scenario, we design a new type of backdoor attack to induce the model to predict the crack as a normal state. Considering the stealthiness of the attack, a common phenomenon during the welding process, illumination, is used as the backdoor trigger. Experimental results demonstrate that the proposed method can successfully attack the crack detection system and achieve over 90% attack success rate on the test set. - PublicationOpen AccessA model learning perspective on the complexity of cyber-physical systems(Universitätsbibliothek der HSU/UniBw H, 2025-05-27)
; ; ;Swantje Plambeck; ;Benndorf, GesaA large palette of models and their corresponding learning algorithms have been applied to time series observed from cyber-physical systems (CPSs). For some use cases, simple linear methods are sufficient, while for others, even sophisticated machine learning approaches fail to extract subtle patterns in system behavior. To date, the literature has not examined this phenomenon adequately and lacks a comprehensive analysis linking the characteristics of CPSs with the suitability of different models and learning algorithms. In this work, after examining the complexity of multiple real-world and artificial CPS use cases, we identify several key aspects that distinguish them: 1) the number of system variables, 2) the degree of interdependence between discrete-event part and continuous part of the system, and 3) the number of unobserved system inputs. By analyzing the approaches successfully applied in the respective use cases, we were able to distill preferred techniques for addressing systems of different complexity levels. - PublicationOpen AccessDeep learning-assisted real-time defect detection and process control for electrode manufacturing of lithium-ion battery cells(Universitätsbibliothek der HSU/UniBw H, 2025-05-27)
;Kampker, Achim ;Heimes, Heiner Hans ;Born, Henrik ;Schmied, Jessica ;Li, Rui YanÖzcan, MertDetecting and preventing defects on electrode surfaces during the manufacturing of lithium-ion battery cells remains a crucial challenge to avoid further cascading effects in subsequent stages of the manufacturing chain. Variations in surface quality or individual contamination can adversely affect battery performance and lifespan, potentially posing safety risks. This paper presents a deep-learning assisted system for detection and classification of coating defects in battery electrodes and subsequent process optimization strategies. Following improvement of product quality and a reduction of the reject rate in the coating and drying process of electrodes, the research contributes to the enhancement of overall efficiency in lithium-ion battery cell manufacturing. To validate the practical application of the system, a case study is conducted in the coating and drying processes of the battery cell pilot line production CELLFAB of the RWTH Aachen University. The results indicate great potential for enhancement of real-time defect detection and further optimization of process parameters. - PublicationOpen AccessChallenges and opportunities in developing INN-based control systems for modular drones(Universitätsbibliothek der HSU/UniBw H, 2025-05-27)
; ;Ludwigs, Robert; ; ;Kampker, AchimAs drone technology evolves, modular drones are increasingly central, offering rapid adaptability through the interchange of sensors, motors, and structural battery modules. However, this flexibility also introduces complex control challenges that traditional Proportional-Integral-Derivative (PID) controllers often struggle to address, particularly under dynamic reconfigurations and nonlinear responses. In this paper, we propose a novel approach integrating Invertible Neural Networks (INNs) and Reinforcement Learning (RL) to enhance adaptability and effectiveness in modular drone control. INNs facilitate precise, reversible command mapping via bijective transformations, ensuring robust handling of changing drone weight, geometry, and functionality. When combined with RL, these networks further enable real-time optimization of flight performance, dynamically responding to shifts in operational conditions. We outline a comprehensive research agenda employing the PX4 simulation framework to benchmark INN- and RL-based methods against standard PID controllers, focusing on improved response times, reduced error rates, and better system resilience. The anticipated findings aim to substantiate the potential of these advanced control systems – particularly in conjunction with emerging structural battery designs – to significantly expand the capabilities and operational scope of next-generation unmanned aerial vehicle (UAVs) in real-world applications. - PublicationOpen AccessAssessing robustness in data-driven modeling of cyber-physical systems(Universitätsbibliothek der HSU/UniBw H, 2025-05-27)
;Schmidt, Maximilian ;Plambeck, SwantjeFey, GoerschwinRobustness is a key factor in the design and analysis of Cyber-Physical Systems (CPS), ensuring that systems function correctly even under perturbations. This paper investigates robustness within the data-driven modeling process, focusing on three core aspects: system robustness, model robustness, and learner robustness. We survey existing notions of robustness and propose unified formal definitions for each aspect, analyzing their interdependencies and their contributions to overall CPS performance. Additionally, we introduce a method for assessing the robustness of models generated by data-driven learning that is independent of both the model’s internal representation and the learning paradigm used. Our approach leverages input perturbations combined with probabilistic analysis to evaluate how well a learned model handles input variations, particularly when formal guarantees are challenging to obtain. To demonstrate the practical application of our method, we conduct a case study on a temperature control system, using decision trees to model system behavior. By perturbing test data and analyzing the resulting model outputs, we identify non-robust regions near decision boundaries, thereby revealing potential vulnerabilities. The proposed framework offers valuable insights for enhancing system design and lays the groundwork for future research into robust machine learning models for CPS. - PublicationOpen AccessTowards adaptive traffic signal control through foundation models and reinforcement learning(Universitätsbibliothek der HSU/UniBw H, 2025-05-27)
;Klein, Lukas ;Müller, ArthurRedeker, MagnusTraffic Signal Control (TSC) is pivotal for managing urban traffic flow and enhancing intersection safety. Traditional TSC systems are rule-based and tailored to specific intersections, requiring substantial training and resources, which restricts their flexibility. This paper proposes a novel adaptive, scalable solution utilizing Foundation Models (FM) and Reinforcement Learning (RL), designed to handle diverse urban intersections efficiently without extensive retraining. The approach leverages advanced neural network architectures, including attention mechanisms, to improve generalization capabilities across different intersection topologies. A safety control mechanism aligned with traffic regulations ensures the safe operation of traffic signals, significantly enhancing the system’s reliability. By systematically classifying intersection types, the method tailors the control strategies to specific traffic scenarios, further reducing implementation times and expertise requirements. This FM- and RL-based approach not only reduces resource demands but also promises more efficient traffic flow and improved safety in various urban settings.