Informatik im Maschinenbau

Cyber 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 7th ML4CPS - Machine Learning for Cyber-Physical Systems Conference in Berlin, held from the 20th to the 21st, where industry and research experts discussed current advancements and new developments.

Kubeflow integrates a suite of powerful tools for Machine Learning (ML) software development and deployment, typically showcased independently. In this study, we integrate these tools within an end- to-end workflow, a perspective not extensively explored previously. Our case study on anomaly detection using telemetry data from the International Space Station (ISS) investigates the integration of various tools—Dask, Katib, PyTorch Operator, and KServe—into a single Kubeflow Pipelines (KFP) workflow. This investigation reveals both the strengths and limitations of such integration in a real-world context. The insights gained from our study provide a comprehensive blueprint for practitioners and contribute valuable feedback for the open source community developing Kubeflow.

This paper examines the integration of Continuous-Time Neural Networks (CTNNs), including Neural ODEs, CDEs, Neural Laplace, and Neural Flows, into engineering practices, particularly in dynamical system modeling. We provide a detailed introduction to CTNNs, highlighting their underutilization in engineering despite similarities with traditional Ordinary Differential Equation (ODE) models. Through a comparative analysis with conventional engineering approaches, using a spring-mass-damper system as an example, we demonstrate both theoretical and practical aspects of CTNNs in engineering contexts. Our work underscores the potential of CTNNs to harmonize with traditional engineering methods, exploring their applications in Cyber- Physical Systems (CPS). Additionally, we review key open-source software tools for implementing CTNNs, aiming to facilitate their broader integration into engineering practices.

In many use cases cyber-physical systems are employed to produce products of small batch sizes as efficiently as possible. From an engineering standpoint, a major drawback of this flexibility is that the architecture of the cyber-physical system may change multiple times over its lifetime to accommodate new product variants. To keep a cyber-physical system working normally it has become common to employ fault diagnosis algorithms. These algorithms partly rely on physical first-principles models that need to be updated when the architecture of the system changes which usually has to be done manually. In this article we present a practical approach to obtain such a first-principles model through evaluating piping and instrumentation diagrams (P&IDs) with visual questions answering (VQA) models. We demonstrate that it is possible to leverage VQA models to construct physical equations which are a preliminary stage for the creation of models suitable for fault diagnosis. We evaluate our approach on OpenAIs GPT-4 Vision Preview model using a P&ID we created for a benchmark water tank system. Our results show that VQA models can be used to create physical first-principles models.

Cyber-physical systems are a class of technical systems that integrate mechanical components with intelligent, adaptable control devices and software. Nowadays, this integration enables high-performance, modular, and parameterized systems with high complexity, but low operating cost. Typical examples of cyber-physical systems are production machinery, cars, aeroplanes, and smart home appliances. In this thesis, the focus is on diagnosing faults within cyber-physical systems used in industrial production contexts. Faults occurring during production quickly lead to degrading product quality or production stops, which can be costly and may endanger human lives. Existing approaches to automated fault diagnosis are mostly defined on narrow use-cases or require a significant amount of expert knowledge. In this thesis, three different algorithms to automatically identify faults in cyber-physical systems are presented to mitigate these drawbacks. Therefore, this thesis makes four main contributions: (i) It introduces a novel diagnosis algorithm HySD to find faults in cyber-physical systems. (ii) It presents a new uninformed algorithm DDRC to learn diagnosis models from process data, using correlations in time-series data. (iii) It presents the new algorithm DDGD, which learns diagnosis models from time-series data supervised, using Granger Causality. (iv) It provides a novel theory to describe fault propagation in cyber-physical systems. More precise, the algorithm HySD uses satisfiability modulo linear arithmetic to combine process data with traditional symbolic consistency-based diagnosis algorithms. However, the algorithm heavily relies on models formulated by experts. Therefore, the algorithms DDRC and DDGD are introduced to learn diagnosis models from process data automatically. All algorithms build on the foundation of the theory of fault propagation. The algorithms were evaluated on internationally accepted benchmarks of tank systems, the well-known Tennessee Eastman Process, and two industrial use-cases. Throughout all empirical results, the algorithms exhibit good performance in learning suitable models and in diagnosing faults in synthetic and real fault scenarios.

Hybrid systems are characterized by a combination of discrete and continuous behavior. Such hybrid systems are becoming more and more widespread, while constantly developing and growing in size because they create countless possibilities for modeling real-world systems. Let a tank, filled with water and connected to a sink through a binary valve, be a simple example; a model of such a system must follow the laws of fluid dynamics, representing the continuous aspect of the tank behavior, but the behavior also contains a discrete aspect, the binary valve. This tank, as all hybrid systems operating in the real-world are, is susceptible to faults, e.g., a crack in the casing creating a leak or a blockage in the valve. Such faults may lead to the system no longer behaving as desired and its control becoming invalid. In such cases, reconfiguration becomes necessary: the automated adaptation of a system to restore the desired behavior by exploiting redundancies in the system. Reconfiguration has been in the scope of research on Artificial Intelligence (AI) for many years now. However, reconfiguration of hybrid systems is still an open research gap for several reasons. From the perspective of AI, the reconfiguration problems are typically solved using qualitative techniques such as search and logical reasoning, but continuous behavior and control are usually not considered. For practical applications, reconfiguration is often solved using naive search techniques or brute force. But as the number of discrete combinations rises exponentially with the number of options for adaptation, these approaches quickly come to their limitations. Hence, reconfiguration for hybrid systems requires dedicated and enhanced solution algorithms. This thesis addresses this research gap with a new definition of reconfiguration for hybrid systems and two novel algorithms, which make use of the efficiency of AI approaches but also consider continuous aspects. The main contributions are as follows: (i) A new definition of reconfiguration for hybrid systems is presented. This definition builds upon related definitions for reconfiguration of discrete systems from AI and extends these to also consider continuous characteristics. It thus forms the basis for the development of novel solution algorithms. (ii) Two novel solution algorithms for reconfiguration of hybrid systems are presented. The first algorithm, AutoConf, is based on encoding the reconfiguration problem in propositional logic. It is applicable to partially observable systems and is very intuitive due to the use of logical calculi that mimic human reasoning. Furthermore, as it relies on well-established satisfiability solving, it allows for efficiently solving reconfiguration problems. For these reasons, AutoConf can be applied to a broad range of hybrid systems. The second algorithm, BFReconf, is very explicit and directly operates on hybrid automata. It creates a problem representation in search and uses greedy best-first search to identify a solution. Even though the algorithm requires complete observability of the hybrid system, it rewards this with a clarity that allows for thorough theoretical examinations. (iii) Soundness and completeness are formalized in the context of reconfiguration for hybrid systems for the first time. These properties allow about a formal and theoretical analysis of the effectiveness of a reconfiguration algorithm. Furthermore, the search-based algorithm BFReconf is shown to be sound and complete for stable, weakly-interconnected, linear hybrid systems. (iv) Experimental evaluation shows the effectiveness of the presented algorithms in practical applications. It comprises experiments on representative simulations and real-world systems and examines statistical accuracy and runtime. The algorithms are shown to efficiently handle many reconfiguration scenarios. Furthermore, the runtimes of the algorithms are observed to often stay below an exponential blowup, which would usually appear when increasing the problem size. To summarize, the presented formal definition of reconfiguration provides a basis for the development of reconfiguration approaches for hybrid systems. And indeed, two such approaches are formalized and shown to be capable of efficiently solving reconfiguration problems for representative systems. And thus, the presented work paves the way for hybrid systems to act autonomously and handle faults on their own.

One possibility to extend the service life of engi-neering structures is to provide adequate maintenance based on Structural Health Monitoring (SHM). Typically, SHM involves a sensor network which is spatially distributed at the surface or within the structure to be monitored. Each sensor measures at least one physical quantity, the data of all sensors then have to be properly evaluated to derive the health state and to predict the remaining service life. Health issues may be detected by machine learning methods by looking for anomalous behaviour in sensor data. Hereby the problem is that malfunctions differ excessively in the representation of the data collected by sensors such that specialisation of methods on anomaly types is required. The current contribution suggests the simulation of sensor malfunction based on established criteria by creating different types of artificial anomalous data indicating different types of issues. Several proposed autoencoder approaches are verified for different anomaly representations, which are artificially introduced in a set of data. The final solutions are different autoencoder specialized on different types of simulated anomaly data, making the conclusions drawn from the measured data more reliable. As a case study, data of a numerical experiment of fibre pull-out are considered.