Flowcean — Model Learning for Cyber-Physical Systems

Maximilian Schmidt, Swantje Plambeck, Markus Knitt, Hendrik Rose, Goerschwin Fey, Jan Christian Wieck, Stephan Balduin · 2026

Read Flowcean — Model Learning for Cyber-Physical Systems by Maximilian Schmidt, Swantje Plambeck, Markus Knitt, Hendrik Rose, Goerschwin Fey, Jan Christian Wieck, Stephan Balduin (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-12015.txt.

Introduction

Schmidt and co-authors present Flowcean, a framework that learns models of cyber-physical systems from data. Building such models by hand is slow, since these systems mix software, sensors, and moving parts. Flowcean wraps the full task in a modular design that users can adapt. It offers varied learning strategies, data handling steps, and scoring metrics in one place. The design also lets teams plug in diverse learning libraries and tools with little friction.

Catalog record

CREATORMaximilian Schmidt, Swantje Plambeck, Markus Knitt, Hendrik Rose, Goerschwin Fey, Jan Christian Wieck, Stephan Balduin
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE27561
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.12015

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Flowcean — Model Learning for Cyber-Physical Systems Maximilian Schmidt, Swantje Plambeck, Markus Knitt, Hendrik Rose, Goerschwin Fey, Jan Christian Wieck, Stephan Balduin Abstract Effective models of Cyber-Physical Systems (CPS) are crucial for their design and operation. Constructing such models is difficult and time consuming due to the inherent complexity of CPS. As a result, data-driven model generation using machine learning methods is gaining popularity. In this paper, we present Flowcean, a novel framework designed to automate the generation of models through data-driven learning that focuses on modularity and usability. By offering various learning strategies, data processing methods, and evaluation metrics, our framework provides a comprehensive solution, tailored to CPS scenarios. Flowcean facilitates the integration of diverse learning libraries and tools within a modular and flexible architecture, ensuring adaptability to a wide range of modeling tasks. This streamlines the process of model generation and evaluation, making it more efficient and accessible. [orcid=0009-0005-4532-7669, [email protected], ] [orcid=0000-0002-4875-5280, [email protected], ] [orcid=0000-0002-4051-2268, [email protected], ] [orcid=0000-0002-7850-5071, [email protected], ] [orcid=0000-0001-6433-6265, [email protected], ] [orcid=0000-0003-0330-2939, [email protected], ] [orcid=0000-0002-2018-1078, [email protected], ] keywords Cyber-Physical Systems , Data-Driven Modeling , Industry 4.0 † † copyrightyear: 2024 † † copyright: Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0). † † venue: 4th Italian Workshop on Artificial Intelligence and Applications for Business and Industries — AIABI — co-located with AI*IA 2024 † † address: Hamburg University of Technology, Hamburg, Germany † † address: Fraunhofer — Center for Maritime Logistics and Services CML, Hamburg, Germany † † address: OFFIS — Institute for Information Technology, Oldenburg, Germany 1 Introduction Cyber-Physical Systems (CPS) are crucial in many industrial sectors, including energy, mobility, and logistics 16 . These systems integrate physical components such as sensors and actuators with digital logic 17 . As a result, CPS pose a range of interdisciplinary challenges; ensuring their reliability and security are only two. In response to these challenges, we introduce a novel machine learning framework for CPS. The following sections outline the motivation for this development, provide a relevant background from related work, and highlight the main contributions of this paper. 1.1 Motivation CPSs represent a broad category of systems, often encompassing expensive components or critical infrastructure, such as specialized hardware or large-scale systems like the electrical power grid. Despite their diversity, CPSs share common characteristics: they are highly complex, often not fully understood, and typically unsuitable for direct use in development processes. Consequently, CPS models are essential for tasks such as design, verification, and testing, as they provide a clearer understanding of system behavior. The diversity of CPS domains brings additional modeling challenges. Conventional physics-based modeling approaches require significant manual effort, extensive domain knowledge, and computational expertise. All of which scale with the complexity of the system 2 . Moreover, the absence of standardized modeling frameworks, along with the wide range of CPS applications, means that each system has a unique design and configuration. This variation complicates the reuse of models and learning processes in different CPSs, making manual modeling time consuming and difficult 6 . For some systems, manual modeling may not even be feasible due to their complexity or the lack of necessary domain expertise. Given these challenges, data-driven modeling has emerged as a promising alternative, as it reduces the need for manual effort and specialized knowledge. Data-driven modeling involves the automatic generation of models from data collected from the system. Figure 1 provides a high-level overview of this process. Observe System Preprocess Data Learn Model A B C Figure 1: Abstract workflow of data-driven modeling The process begins with system observation (A), where data is collected from various sources, such as precollected datasets, online simulations, or interactions with real-world CPS. The data is then pre-processed (B) through tasks such as standardization or feature selection, which determine how the learning algorithm interprets the information and establishes an abstraction between the system and its model. In the model learning phase (C), a learning algorithm processes the data to generate a model that captures the behavior of the system. Since each CPS is unique, the data-driven learning pipeline must be customized to the specific characteristics of the system. As a result, these pipelines often produce highly specialized solutions, tailored to individual CPS. 1.2 Background Machine learning frameworks such as PyTorch and TensorFlow have made significant strides in modeling the behavior of CPS. The flexibility of these frameworks has been effectively used in tasks such as anomaly detection within digital twins 15 and the development of learning-based adversarial agents in CPS environments 5 . These frameworks support the processing of large-scale datasets and the design of complex neural network architectures, making them suitable for a wide range of CPS applications 4 . For time-discrete CPS, traditional approaches like automata learning are commonly used to model system behavior. Automata can be learned through active system interaction 14 ; 13 or passive observation of system behavior 8 . These approaches are particularly effective for systems where the state space is discrete and the dynamics are governed by rules or transitions between states. For nondeterministic systems, decision trees have also been employed as a simple yet powerful approach to model behavior 12 ; 1 . Although effective, these techniques often depend on domain-specific algorithms and typically require custom implementations tailored to each new application. Libraries such as Keras and scikit-learn have been popularizing machine learning by providing accessible and efficient implementations of common algorithms. Keras, known for its user-friendly interface, enables rapid prototyping and experimentation with neural networks. scikit-learn offers a wide array of tools for data mining and analysis, including regression, classification, and clustering algorithms, which have been applied in CPS for tasks such as fault detection 10 and predictive maintenance 7 . Although these frameworks excel at general-purpose machine learning tasks, they are less equipped to meet the specific demands of CPS, such as real-time data processing or the integration of heterogeneous data sources. 1.3 Contribution Although existing machine learning frameworks are powerful, they are typically designed for specific learning strategies, such as passive learning on static datasets or reinforcement learning in controlled environments. This specialization limits their flexibility in accommodating different approaches or generalizing across various CPS applications. Adapting the same system for different learning strategies typically requires modifying the learning library and restructuring the data pipelines. For example, a framework optimized for passive learning may not efficiently handle incremental learning scenarios, where models must be continuously updated as new data become available. Similarly, deep learning frameworks may not support automata-based approaches or decision trees, which are better suited for certain types of CPS. Due to the complexity of CPS behavior, it is often difficult to predict the optimal approach and framework before modeling begins, leading to duplicated effort if a change in framework becomes necessary mid-process. To address these limitations and challenges, we introduce Flowcean , a modular framework designed specifically to generate CPS models. Flowcean offers several key contributions to existing machine learning frameworks by addressing the unique needs of CPS modeling: • Re-usability of learning processes and components: Flowcean provides a flexible learning pipeline that defines reusable paradigms and interfaces, enabling modules to be applied across different CPS applications without significant reconfiguration. • Integration and combination of

Read the complete text