Search

The combination of physics and engineering information with data-driven methods like machine learning (ML) and deep learning is gaining attention in various research fields. One of the promising practical applications of such hybrid methods is for supporting maintenance decision making in the form of condition-based and predictive maintenance. In this paper we focus on the potential of physics-informed data augmentation for ML algorithms. We demonstrate possible implementations of the concept using three use cases, differing in their technical systems, their algorithms and their tasks ranging from anomaly detection, through fault diagnostics up to prognostics of the remaining useful life. We elaborate on the benefits and prerequisites of each technique and provide guidelines for future practical implementations in other systems.

Patient monitoring in intensive care units, although assisted by biosensors, needs continuous supervision of staff. To reduce the burden on staff members, IT infrastructures are built to record monitoring data and develop clinical decision support systems. These systems, however, are vulnerable to artifacts (e.g. muscle movement due to ongoing treatment), which are often indistinguishable from real and potentially dangerous signals. Video recordings could facilitate the reliable classification of biosignals using object detection (OD) methods to find sources of unwanted artifacts. Due to privacy restrictions, only blurred videos can be stored, which severely impairs the possibility to detect clinically relevant events such as interventions or changes in patient status with standard OD methods. Hence, new kinds of approaches are necessary that exploit every kind of available information due to the reduced information content of blurred footage and that are at the same time easily implementable within the IT infrastructure of a normal hospital. In this paper, we propose a new method for exploiting information in the temporal succession of video frames. To be efficiently implementable using off-the-shelf object detectors that comply with given hardware constraints, we repurpose the image color channels to account for temporal consistency, leading to an improved detection rate of the object classes. Our method outperforms a standard YOLOv5 baseline model by +1.7% mAP@.5 while also training over ten times faster on our proprietary dataset. We conclude that this approach has shown effectiveness in the preliminary experiments and holds potential for more general video OD in the future.

We examine the paradigm of data-centric artificial intelligence (DCAI) as a solution to the obstacles that small and medium-sized enterprises (SMEs) face in adopting AI. While the prevalent model-centric approach emphasizes collecting large amounts of data, SMEs often suffer from small datasets, data drift, and sparse ML knowledge, which hinders them from implementing AI. DCAI, on the other hand, emphasizes to systematically engineer the data used to build an AI system. Our contribution is to provide a concrete, transferable implementation of a DCAI development process geared towards industrial application, specifically in machining and manufacturing, and demonstrate how it enhances data quality by fostering collaboration between domain experts and ML engineers. This added value can place AI at the disposal of more SMEs. We provide the necessary background for practitioners to follow the rationale behind DCAI and successfully deploy the provided process template.

The fairness of machine learning-based decision support systems has become a critical issue, also in the field of predicting the success of venture capital investment startups. Inappropriate allocation of venture capital, fueled by discriminatory biases, can lead to missed investment opportunities and poor investment decisions. Despite numerous studies that have addressed the prevalence of biases in venture capital allocation and decision support models, few have addressed the importance of incorporating fairness into the modeling process. In this study, we leverage invariant feature representation learning to develop a startup success prediction model using Crunchbase data, while satisfying group fairness. Our results show that discriminatory bias can be significantly reduced with minimal impact on model performance. Additionally, we demonstrate the versatility of our approach by mitigating multiple biases simultaneously. This work highlights the significance of addressing fairness in decisionsupport models to ensure equitable outcomes in venture capital investments.