Agenda

Magnetic resonance imaging (MRI) is a cornerstone of diagnostic healthcare and clinical research, providing high-quality structural and functional images without ionizing radiation. Central to MRI is the reconstruction process, which is fundamentally an inverse mathematical problem. In essence, the measured k-space data is modeled as data = Encode × image, meaning that recovering the image involves applying the inverse transformation image = Encode⁻¹ × data.

Historically, MRI has relied on fast Fourier transforms and iterative algorithms because direct inversion was computationally infeasible—the encoding matrix for even a modest 128 × 128 image can reach dimensions of 16,384 × 16,384 (consuming around 2.1 GB in single precision). Today, modern computer hardware and MATLAB^® make direct pseudoinversion feasible. By using optimized linear algebra routines in MATLAB on multicore CPUs and GPUs with ample memory, we compute the pseudoinverse of the encoding matrix and perform image reconstruction in a single matrix-vector multiplication step. This approach is highly flexible, and advanced encoding physics such as parallel imaging, off-resonance, gradient distortions, and motion can all be integrated into the encoding matrix and processed within a unified framework. Moreover, by explicitly computing the pseudoinverse of the encoding matrix, we can extract interpretable metrics such as point-spread function and noise-propagation maps that offer additional insights into the encoding and reconstruction process.

We further extend this algebraic framework toward AI by integrating implicit neural image representations and applying deep learning techniques for tasks like denoising and artifact reduction. These extensions merge the rigor of classical linear algebra with modern AI, further enhancing MRI reconstruction performance.

Die Millimeterwellen-(mmWave)-Kommunikation bietet hohe Bandbreiten und eCiziente Spektrumnutzung, leidet jedoch unter geringer Robustheit gegenüber Blockierungen und Abschattungen infolge von Mobilität. Intelligent Reconfigurable Surfaces (IRS) können diese Probleme zwar mindern, ihre Komplexität erschwert jedoch den praktischen Einsatz. Diese Arbeit untersucht eine praktische Alternative mit geringerer Komplexität – Non-reconfigurable Reflecting Surfaces (NRRS), die feste Signalreflexionen erzeugen und so die Verbindungsstabilität verbessern. Mithilfe eines mmWave-Testbeds zeigen wir, dass NRRS selbst bei begrenztem Einsatz die Zuverlässigkeit der Kommunikation massiv verbessern. Die Messungen und Analysen, durchgeführt mit den WLAN-Toolboxen von MATLAB^®, zeigen zudem, dass NRRS zu starker Frequenzselektivität führen. Dies macht die Verwendung fortschrittlicher physikalischer Schichten wie IEEE 802.11be unerlässlich.

In the field of system engineering for complex product developments, the transition from static diagrams to dynamic, simulatable architectures is crucial for an efficient and reliable system design. This presentation introduces a shift-left approach, using simulatable system architectures instead of traditional static representations. This enables early validation of system concepts, rapid detection of design flaws, and continuous verification throughout development. By using simulation of the system architecture, early error detection leads to an increased maturity of requirements and more informed design decisions, resulting in cost and time savings and enhanced system quality. The integration of CI further enhances this process by supporting complex architecture changes through automated regression tests, ensuring that errors are identified at the earliest possible stage.

The integration of AI into battery management systems (BMS) is revolutionizing the electric vehicle industry. Using large language models (LLMs), such as GPT, and MATLAB^® and Simulink^® allows for more efficient BMS algorithm development. LLMs facilitate advanced natural language processing and rapid analysis of large datasets, which streamlines algorithm creation, implementation, and testing.

MATLAB and Simulink provide a powerful platform for modeling battery behavior, simulating scenarios, and optimizing control strategies, which accelerates the transition from concept to deployment. Central to combining AI technologies with AMTLAB and Simulink is FEV’s GenAI Hub, which aggregates high-quality data from multiple sources and fosters collaboration among experts. This step allows the development of advanced BMS algorithms, enhancing battery performance, safety, and reliability. 

With AI-driven insights and predictive modeling, FEV’s solutions enable accurate estimation of battery health, state-of-charge, and state-of-health—all critical for maintaining EV performance and longevity. These advancements not only benefit individual EV users but also support fleet management and energy storage applications. By maximizing the use of residual battery capacity, FEV’s AI-powered BMS contributes to sustainable energy solutions by reducing environmental impact and total cost of ownership.

The integration of advanced AI technologies into BMS development is a key driver for the future of mobility, paving the way for greener and smarter transportation systems.

Cardiovascular diagnostics rely on spatial information for accurate assessment. This session presents a certified medical software platform for the analysis of vectorelectrocardiography signals in a 3D vector space. Developed entirely in MATLAB^®, this tool enables clinicians and researchers to analyze, visualize, and interpret cardiac electrical activity. 

Attendees will learn how MATLAB was used not only for signal processing and visualization, but also for verification, validation, and certification of a real-world medical product. This includes challenges in aligning development in MATLAB with medical device regulations such as MDR, and successful deployment in a clinical environment.

This project demonstrates how MATLAB and Simulink^® can support the full product life cycle—from prototype to certified solution—and showcases a compelling use case for MATLAB in regulated medical software.

Hero Tech Center Germany showcases an integrated development approach combining model-based systems engineering (MBSE) and model-based software development (MBSD) using a digital vehicle twin for electric powertrain development. This model-centric workflow connects system-level requirements, architecture, behavior modeling, and verification in a single, traceable process.

Traditional motorcycle testing is limited by high costs, long iteration cycles, and late-stage fault detection. Physical prototypes are hard to scale, incomplete early on, and pose risks when validating safety-critical functions. In contrast, our digital twin—a high-fidelity, executable model—enables early, repeatable validation in virtual environments.

Using MATLAB^® and Simulink^®, we integrate system requirements with functional and behavioral models. These requirements are linked and verified across development stages, which improves traceability and ensures consistency from architecture to implementation. Continuous verification through model-, software-, and hardware-in-the-loop testing allows early fault detection and reduces reliance on physical builds.

The digital twin supports scalable testing of control algorithms, system integration, and safety features before hardware is available. Requirements act as both design inputs and validation criteria, enabling automated compliance checks and improved development speed.

This use case in electric powertrain illustrates how a requirements-driven, model-based workflow accelerates development, enhances quality, and ensures robust, scalable design. It also addresses toolchain alignment and integration challenges across MBSE and MBSD environments, boosting agility and system reliability.

Fehlermodellierung ist eine Schlüsselmethode bei der Entwicklung sicherer und zuverlässiger Systeme. Das Entwickeln von Logik für die Erzeugung von verschiedenen Fehlertypen und das parallele Verwalten von fehlerfreien und fehlerbehafteten Modellen stell jedoch eine nicht unerhebliche Herausforderung dar.  In dieser Präsentation geben wir einen praxisnahen Überblick über die Modellierung in Simscape™ und zeigen, wie sich verschiedenste Fehlertypen einfach und systematisch abbilden und analysieren lassen. Nach einer kurzen Einordnung der Fehlermodellierung im Kontext von Model-Based System Engineering, FMEA und Requirements Engineering werden die Vorteile moderner, nicht-invasiver Ansätze gegenüber klassischen „Dirty Models“ erläutert.

Im Mittelpunkt steht die konkrete Umsetzung: Anhand von Beispielen aus den Bereichen Halbleiter, Motoren und Batteriezellen wird gezeigt, wie Fehler in Simscape modelliert, konfiguriert und analysiert werden. Viele Simscape Blöcke bieten bereits die integrierte Möglichkeit, typische physikalische Fehlerfälle abzubilden. Auch eigene Simscape Blöcke können in der ssc-Sprache mit Fehlern beaufschlagt werden. Die Fehler können dann während der Simulation durch zeitliche oder logische Bedingungen ausgelöst und die richtige Reaktion darauf getestet werden. 

Diese Präsentation richtet sich an Entwickler und Ingenieure, die Systeme entwickleln und analysieren, bei denen Robustheit und korrektes Verhalten auch im Fehlerfall bereits in der Simulation abgedeckt werden soll. Sie vermittelt sowohl methodische Grundlagen als auch praktische Tipps für die Umsetzung im Modell.

Künstliche Intelligenz (KI), insbesondere in Form trainierter neuronaler Netze, ist in Bereichen wie Bildklassifizierung, Objekterkennung und natürlicher Sprachverarbeitung nicht wegzudenken. Diese Anwendungsmöglichkeiten von KI-Modellen stoßen auch in sicherheitskritischen Sektoren wie der Luft- und Raumfahrt, der Automobilindustrie und dem Gesundheitswesen auf hohes Interesse. Verifizierung und Validierung dieser Modelle stellen jedoch besondere Herausforderungen dar, da herkömmliche Software-Assurance-Prozesse die inhärenten Komplexitäten von KI-Systemen, wie mangelnde Transparenz und unvorhersehbares Verhalten unter bestimmten Bedingungen, oft nicht berücksichtigen. 

KI-Verifizierung und -Validierung (V&V) ist für die Zertifizierung KI-fähiger Systeme in diesen kritischen Bereichen aber unerlässlich. Es gibt mittlerweile Industriestandards, die diesen Prozess begleiten, darunter ISO/PAS 8800:2024 für die Automobilindustrie, welcher Sicherheitseigenschaften, Risikofaktoren und Verifizierungstechniken für Straßenfahrzeuge definiert. ARP6983 (in Arbeit) ist ein Standard für die Luftfahrtindustrie, welcher Leitlinien für die Entwicklung KI-gesteuerter Flugzeugsysteme mit Schwerpunkt auf Zertifizierung, Sicherheitsbewertung und Einhaltung gesetzlicher Vorschriften bietet. ECSS-E-HB-40-02A schließlich ist ein Handbuch für Raumfahrttechnik, mit Verfahren für zuverlässiges maschinelles Lernen innerhalb einer „Sicherheitskäfig“-Architektur. Diese Standards schaffen robuste Rahmenbedingungen, um die Sicherheit und Zuverlässigkeit KI-gestützter technischer Systeme zu gewährleisten. Durch Methoden wie szenariobasierte Tests, formale Überprüfung von Eigenschaften, Erklärbarkeitstechniken und Laufzeitabsicherung stellt Verifizierung und Validierung (V&V) sicher, dass KI-Modelle bestehende und zukünftige Industriestandards erfüllen. MathWorks bietet Tools, die diese Prozesse unterstützen – kompatibel mit MATLAB^®- und PyTorch^®-Modellen – und so die Zertifizierung von sicherheitskritischen Anwendungen erleichtern.

Having a link-level simulator in MATLAB^® that models our physical layer hardware brings many advantages. We can easily simulate mobile radio scenarios for 4G and 5G and limit expensive lab and field testing. It is not only useful for designing powerful baseband algorithms, but it also allows us to track the link-level performance of our products in detail and compare different hardware generations against each other. For modeling radio channels, we can choose from a variety of standardized channels but also have the possibility to inject channel data collected from field tests.

In this presentation, we investigate eigen-based beamforming with different covariance matrix flavors for DL MU-MIMO. We compare the options both for the widely used 3GPP CDL channel model and for SRS channel data that we collected from field tests. This shows that the comparison between schemes differs for CDL and measured channels.

Model-based systems engineering (MBSE) is a methodology that integrates multiple engineering disciplines to manage and realize complex systems, ensuring the availability of relevant information across the entire system life cycle. The recent development of the SysML v2 API significantly enhances this integration by enabling diverse engineering tools to access and interact with SysML v2 system models. 

In this presentation, we demonstrate how MATLAB^® and Simulink^® support SysML v2 and how users can leverage the SysML v2 standards to streamline their workflows. Specifically, we will showcase how a model created in one tool can be ingested into the MATLAB and Simulink environment, subjected to various analyses—including trade-off studies and requirements completeness and consistency analysis—and subsequently updated based on these findings. The enhanced model can then be committed back to the SysML v2 repository, facilitating a seamless and collaborative MBSE process. 

The capability of electrical devices to contribute to the stability and efficiency of the power grid is referred to as grid supporting. This function is particularly crucial in modern energy systems that increasingly rely on renewable energy sources, which require greater flexibility and resilience. 

In this project, Model-Based Design was employed to integrate the different grid supporting functions into three different devices from the related teams: an inverter, a turbine controller, and a wind farm controller. Model-Based Design allows for the design and simulation of complex systems through models before practical implementation. These models were used to incorporate the grid supporting functions as additional features into the existing devices. The implementation process involved several steps: First, Simulink^® models for the grid supporting functions were developed and tested in simulation. These models were then integrated into the control software of the devices. Finally, the devices were operated in a test environment to evaluate their performance and reliability.

The test results demonstrated that the integration of the grid supporting functions was successful. The devices contributed to grid stability by responding to fluctuations in the power grid and making necessary adjustments. The use of Model-Based Design enabled the rapid and efficient development and implementation of the new functions.

Imagine standing before an impressive machine that produces PET bottles. This machine is not only a marvel of engineering but also a prime example of the integration of cutting-edge technologies. In this presentation, you will learn how continuous integration and continuous delivery seamlessly connect with a machine learning pipeline to develop and provide an intelligent controller.

To cope with the high data rate requirements of 6G, the channel bandwidth for 6G is expected to exceed that of 5G. In 3GPP Release 18, the maximum supported channel bandwidth in FR2 is 2 GHz for certain channel configurations, including the n263 band, which supports up to 2 GHz bandwidth in specific configurations. When it comes to the upper midband, the 7–15 GHz range is considered a potential band for 6G, where the channel bandwidth is expected to exceed 800 MHz.

In this presentation, we address the challenges in the digital front-end (DFE) with respect to increased signal bandwidth in terms of digital signal processing, DFE architecture, and high-speed digital conversion systems. We use MATLAB^® to implement the transmit/receive chains of the DFE, including IFFT/FFT, windowing, frequency conversion, and digital pre-distortion. We optimize the components of the Tx/Rx chains to fulfill the 3GPP base station conformance test requirements, such as modulation quality, ACLR measurements, and emission mask.

Fehlerinjektionstests sind eine entscheidende Methode zur Sicherstellung der Systemrobustheit und werden traditionell auf Hardware-Ebene im Rahmen von Hardware-in-the-Loop-Tests (HiL) durchgeführt.

Durch den Einsatz modellbasierter Entwicklung können Fehlersimulation und Fehlerinjektion jedoch bereits deutlich früher im Entwicklungsprozess, nämlich auf Ebene der Systemarchitektur und der Model-in-the-Loop-Phase (MiL), eingesetzt werden.

Dadurch lässt sich die Lücke zwischen Systemmodellen und HiL-Tests schließen, was zu kürzeren Entwicklungszyklen und einer verbesserten Fehlerabdeckung führt. Durch den Einsatz von Simulink Fault Analyzer™ wird zudem eine systematische Analyse von Fehlereffekten und mehr Sicherheit durch Simulation ermöglicht. Dies schafft eine formale Verbindung zwischen Fehlern, Gefahren, Fehlererkennungs- und -minderungslogik und trägt entscheidend zur Erhöhung der Systemsicherheit bei.

Novel aircraft concepts and technologies like hydrogen drive upcoming systems architectures. The model-based overall systems design (OSD) framework provides a structured approach for conducting trade studies of aircraft systems architectures. 

Initially, the aircraft is designed and top-level aircraft requirements are defined. This step is not directly part of the OSD framework and is typically performed by partners, such as the DLR, providing the aircraft information via a standardized CPACS file.

Next, OBS architectures are defined, evaluated, and down-selected at a functional-logical level with MBSE-driven methods using the in-house SARA methodology. Operations, requirements, functions, and logical architectures are modeled in a traceable manner and the model functions as a single source of truth for integrated safety assessment, performance evaluation, and early verification and validation. Additionally, existing architecture patterns and technology knowledge support the architecture generation step, facilitating decision-making and down-selection of unpromising architecture variants.

Finally, the remaining promising architectures are further assessed using the in-house GeneSys methodology. The system components are positioned geometrically within the aircraft and then preliminarily sized using physical sizing laws derived from detailed systems design methods. The outcome of GeneSys includes component masses, system masses, and component-specific design parameters, such as the design power of an electric generator or the displacement of a hydraulic pump. Moreover, a preliminary quasistatic simulation of the overall system behavior is conducted, ultimately providing information on the shaft power off-take through a reference mission profile. 

The framework is demonstrated on two hydrogen-powered aircraft concepts, showcasing its capability to accelerate parametric architecture trade studies.

Spannungswandler finden sich fast überall – in LED-Leuchtmitteln, Ladegeräten, Computern, Autos oder Zügen. Ihre Regelung basiert aktuell oft auf dem klassischen PID-Regler. Dieser ist leicht zu verstehen und zu implementieren, aber ein rein linearer Regler. Moderne Spannungswandler schalten zwischen verschiedenen Zuständen hin und her. Damit sind die sich ergebenen Systeme weder zeitinvariant noch linear, so dass der PID-Regler nicht in allen Fällen eine optimale Regelung erzeugen kann. Schnelle Lastwechsel und größere Zustandsräume bringen so die PID-Regler an ihre Grenzen. Die Zeit scheint also reif zu sein für Regler, die auf künstlichen neuronalen Netzen basieren und mit Hilfe von Reinforcement Learning trainiert werden. Wie lässt sich jedoch die Güte eines solchen Reglers beurteilen? 

Zunächst ist klar, dass er besser sein muss als ein gut eingestellter PID Regler. Wie gut kann aber ein AI basierter Regler überhaupt werden und wie wird dieser optimal trainiert? Im Rahmen der Präsentation wird eine Klasse von neuronalen Netzen vorgestellt und das entsprechende Reinforcement Learning Problem für deren Training formuliert. Die Reward-Funktion wird so dann in einem Regler basierend auf Model-Predictive-Control eingesetzt. So wird ein Regler mit der (bei gegebenem Konverter) besten möglichen Regelung konstruiert. Jedoch ist diese Regelung konstruktionsbedingt rechnerisch viel zu aufwendig. Im Rahmen der Präsentation werden entsprechende Simulink-Modelle vorgestellt, Benchmark-Probleme und deren Ergebnisse diskutiert und so ein Rahmen bereitgestellt, welcher es ermöglicht, die Güte beliebiger nichtlinearer, AI basierter Regelungen in Kontext zu setzen.

The advent of 6G technology promises to revolutionize communication systems, enabling unprecedented capabilities and applications. In this presentation, hear how powerful workflows in MATLAB® can accelerate your 6G R&D research. MATLAB provides a comprehensive environment for investigating and developing 6G enabling technologies, including waveform exploration, integrated sensing and communication, AI and neural receivers, reconfigurable intelligent surfaces, and non-terrestrial networks.

We will demonstrate how MATLAB facilitates the research and development of these critical technologies, showcasing its robust capabilities in simulation, modeling, and analysis.

In this session, we tackle the integration of predictive maintenance (PdM) systems on industrial controllers, a key advancement in Industry 4.0. Learn how Siemens automation systems and Model-Based Design can optimize industrial operations. We'll explore the integration of Simulink^® components within the TIA Portal and LiveTwin, highlighting practical benefits through a live demo of fault detection using Siemens automation technology. This demo showcases synthetic data generation via a digital twin approach, focusing on an industrial fan application. The project aims to enhance fault classification in industrial equipment by embedding advanced PdM strategies within the Siemens framework. MATLAB^® and Simulink provide deployment strategies that have been successfully optimized maintenance and offer Siemens customers AI-driven applications with significant operational advantages. 

The demo illustrates a PdM framework for Siemens PLCs and edge devices, simulating faults in a physics-based model to create datasets, followed by data preprocessing and feature extraction. A machine learning model is developed for fault classification and tailored for edge computing. Integration with Siemens hardware ensures efficient PdM model deployment for real-time monitoring and swift issue resolution, minimizing downtime and advancing smart manufacturing.