Open Access

Diese Arbeit beschreibt die Entwicklung eines Systems zur kontinuierlichen Logfile-Analyse von Herz-Lungen-Maschinen und die Umsetzung dessen in Form eines Prototyps. Konkret wird die Frage beantwortet, wie ein System aussieht, welches kontinuierlich Logfiles analysiert und deren Fehler und Ursachen zusammenfasst. Dafür wurde der komplette Softwaredesignprozess von der Anforderungsanalyse bis hin zur Präsentation der Ergebnisse dargestellt. Es entstand ein Softwaresystem, welches in der Lage ist, automatisiert Logfiles von Herz-Lungen-Maschinen einzusammeln. Diese Logfiles werden als Zwischenschritt in Pipelines verarbeitet und anschließend auf einem zentralen Server in einer dateibasierten NoSQL Datenbank abgespeichert. Über ein Webinterface ist es möglich, die gespeicherten Daten explorativ zu untersuchen und mithilfe von Diagrammen und Dashboards zu visualisieren. Als Technologie wurde dabei der Elastic Stack mit den Komponenten Filebeat, Logstash, Elasticsearch und Kibana eingesetzt.

This study aims to analyze a novel indirect photoacoustic sensor (PAS) using Machine Learning techniques. The studies focus on understanding the sensor’s repeatability, the influence of temperature and humidity on microphone output voltage, and the applicability of Machine Learning models to accurately describe the sensor’s behavior. To describe the sensor behavior, two studies are carried out in a controlled setting. With a R2 score of 0.964 between the microphone voltage and the gas concentration in ppm, the first study illustrates the sensor’s repeatability for concentration measurements. The second study looks at how temperature and humidity affect microphone output voltage. For this study, R2 score of 0.948 is obtained. The studies underscore the necessity for further investigation of the sensor under diverse testing conditions. The findings demonstrate that the sensor exhibits consistent behavior and can be effectively modeled using Machine Learning techniques.

Time-Sensitive Networking (TSN) is becoming increasingly important. Especially in the field of industrial applications, the demand for uniform, converged real-time networks is continuously increasing. Furthermore, the request to integrate wireless, mobile, and real-time capable network elements is getting more and more relevant to industrial automation use cases. To address these requests, the 3rd Generation Partnership Project (3GPP) has extended their specifications for mobile telecommunication protocols by descriptions to integrate 5G mobile networks into TSN starting from Release 16 onwards. While the specifications provide a good theoretical overview, there is still a lack of real implementations or even proof of concepts. Therefore, we started an implementation of a 5G network that is ready to be integrated into existing TSN. This work gives an overview of the current work in progress, mainly focusing on the implementation of the TSN Application Function (TSN AF) and the time synchronization features within the TSN Translators (DS-TT and NW-TT). It also shows current limitations and difficulties and how we have overcome them with our setup.

This study introduces EmbeddedTrain, an innovative algorithm optimized for on-device learning in deep neural networks, specifically designed for low-power microcontroller units. EmbeddedTrain refines sparse backpropagation by dynamically adjusting the level of sparity, including the ability to selectively skip training steps. This feature significantly lowers computational effort without substantially compromising accuracy. Our comprehensive evaluation across diverse datasets—CIFAR 10, CIFAR100, Flower, Food, Speech Command, MNIST, HAR, and DCASE2020—reveals that EmbeddedTrain achieves near-parity with full training methods, with an average accuracy drop of only around 1% in most cases. For instance, against full training, EmbeddedTrain’s accuracy drop is minimal, for example, only 0.82% on CIFAR 10 and 1.07% on CIFAR100. In terms of computational effort, EmbeddedTrain shows a marked reduction, requiring as little as 10% of the computational effort needed for full training in some scenarios, and consistently outperforms other sparse training methodologies. These findings underscore EmbeddedTrain’s capacity to efficiently manage computational resources while maintaining high accuracy, positioning it as an advantageous solution for advanced embedded device applications in the IoT ecosystem.

With the expansion of Internet-of-Things (IoT) devices in many aspects of our life, the security of such systems has become an important challenge. Unlike conventional computer systems, any IoT security solution should consider the constraints of these systems such as computational capability, memory, connectivity, and energy consumption limitations. Physical unclonable functions (PUFs) with their special characteristics were introduced as hardware-based solutions to satisfy the security needs while respecting the mentioned constraints. They exploit the uncontrollable and reproducible variations of the underlying components for security applications such as identification, authentication, and secure boot. Since IoT devices are typically low cost, it is important to reuse existing elements in their hardware (for instance, sensors, analog-to-digital converters (ADCs), etc.) instead of adding extra costs for the PUF hardware. Micro-electromechanical system (MEMS) devices are widely used in IoT systems as sensors and actuators. In this work, for the first time, a lightweight MEMS-based circuit with a piezoresistive bridge is introduced as a weak PUF. The piezoresistive PUF leverages the uncontrollable variations in the parameters of the circuit elements to derive secure keys for cryptographic applications. The experimental results show that our proposed piezoresistive PUF is capable of generating enough entropy for a complex key generation, while its responses show stability in different environmental conditions. The manufactured piezoresistive PUF shows a uniqueness of 47.73% and a reliability of 94.19%. Moreover, the generated secret keys passed the National Institute of Standards and Technology (NIST) test suite for randomness.