Refine
Year of publication
- 2023 (114) (remove)
Institute
- Fachbereich Medizintechnik und Technomathematik (28)
- Fachbereich Elektrotechnik und Informationstechnik (21)
- Fachbereich Luft- und Raumfahrttechnik (19)
- ECSM European Center for Sustainable Mobility (18)
- Fachbereich Chemie und Biotechnologie (16)
- Fachbereich Energietechnik (13)
- INB - Institut für Nano- und Biotechnologien (11)
- IfB - Institut für Bioengineering (9)
- Fachbereich Wirtschaftswissenschaften (8)
- Fachbereich Maschinenbau und Mechatronik (6)
Language
- English (114) (remove)
Document Type
- Article (64)
- Conference Proceeding (35)
- Part of a Book (6)
- Habilitation (2)
- Preprint (2)
- Talk (2)
- Book (1)
- Conference: Meeting Abstract (1)
- Contribution to a Periodical (1)
Keywords
- Information extraction (3)
- Natural language processing (3)
- Associated liquids (2)
- Bacillaceae (2)
- Biotechnological application (2)
- CFD (2)
- Diversity Management (2)
- Engineering Habitus (2)
- Future Skills (2)
- Interdisciplinarity (2)
In recent years, the development of large pretrained language models, such as BERT and GPT, significantly improved information extraction systems on various tasks, including relation classification. State-of-the-art systems are highly accurate on scientific benchmarks. A lack of explainability is currently a complicating factor in many real-world applications. Comprehensible systems are necessary to prevent biased, counterintuitive, or harmful decisions.
We introduce semantic extents, a concept to analyze decision patterns for the relation classification task. Semantic extents are the most influential parts of texts concerning classification decisions. Our definition allows similar procedures to determine semantic extents for humans and models. We provide an annotation tool and a software framework to determine semantic extents for humans and models conveniently and reproducibly. Comparing both reveals that models tend to learn shortcut patterns from data. These patterns are hard to detect with current interpretability methods, such as input reductions. Our approach can help detect and eliminate spurious decision patterns during model development. Semantic extents can increase the reliability and security of natural language processing systems. Semantic extents are an essential step in enabling applications in critical areas like healthcare or finance. Moreover, our work opens new research directions for developing methods to explain deep learning models.
Residential and commercial buildings account for more than one-third of global energy-related greenhouse gas emissions. Integrated multi-energy systems at the district level are a promising way to reduce greenhouse gas emissions by exploiting economies of scale and synergies between energy sources. Planning district energy systems comes with many challenges in an ever-changing environment. Computational modelling established itself as the state-of-the-art method for district energy system planning. Unfortunately, it is still cumbersome to combine standalone models to generate insights that surpass their original purpose. Ideally, planning processes could be solved by using modular tools that easily incorporate the variety of competing and complementing computational models. Our contribution is a vision for a collaborative development and application platform for multi-energy system planning tools at the district level. We present challenges of district energy system planning identified in the literature and evaluate whether this platform can help to overcome these challenges. Further, we propose a toolkit that represents the core technical elements of the platform. Lastly, we discuss community management and its relevance for the success of projects with collaboration and knowledge sharing at their core.
This article describes an Internet of things (IoT) sensing device with a wireless interface which is powered by the energy-harvesting method of the Wiegand effect. The Wiegand effect, in contrast to continuous sources like photovoltaic or thermal harvesters, provides small amounts of energy discontinuously in pulsed mode. To enable an energy-self-sufficient operation of the sensing device with this pulsed energy source, the output energy of the Wiegand generator is maximized. This energy is used to power up the system and to acquire and process data like position, temperature or other resistively measurable quantities as well as transmit these data via an ultra-low-power ultra-wideband (UWB) data transmitter. A proof-of-concept system was built to prove the feasibility of the approach. The energy consumption of the system during start-up was analysed, traced back in detail to the individual components, compared to the generated energy and processed to identify further optimization options. Based on the proof of concept, an application prototype was developed.
Clinical assessment of newly developed sensors is important for ensuring their validity. Comparing recordings of emerging electrocardiography (ECG) systems to a reference ECG system requires accurate synchronization of data from both devices. Current methods can be inefficient and prone to errors. To address this issue, three algorithms are presented to synchronize two ECG time series from different recording systems: Binned R-peak Correlation, R-R Interval Correlation, and Average R-peak Distance. These algorithms reduce ECG data to their cyclic features, mitigating inefficiencies and minimizing discrepancies between different recording systems. We evaluate the performance of these algorithms using high-quality data and then assess their robustness after manipulating the R-peaks. Our results show that R-R Interval Correlation was the most efficient, whereas the Average R-peak Distance and Binned R-peak Correlation were more robust against noisy data.
Digital forensics of smartphones is of utmost importance in many criminal cases. As modern smartphones store chats, photos, videos etc. that can be relevant for investigations and as they can have storage capacities of hundreds of gigabytes, they are a primary target for forensic investigators. However, it is exactly this large amount of data that is causing problems: extracting and examining the data from multiple phones seized in the context of a case is taking more and more time. This bears the risk of wasting a lot of time with irrelevant phones while there is not enough time left to analyze a phone which is worth examination. Forensic triage can help in this case: Such a triage is a preselection step based on a subset of data and is performed before fully extracting all the data from the smartphone. Triage can accelerate subsequent investigations and is especially useful in cases where time is essential. The aim of this paper is to determine which and how much data from an Android smartphone can be made directly accessible to the forensic investigator – without tedious investigations. For this purpose, an app has been developed that can be used with extremely limited storage of data in the handset and which outputs the extracted data immediately to the forensic workstation in a human- and machine-readable format.
In order to reduce energy consumption of homes, it is important to make transparent which devices consume how much energy. However, power consumption is often only monitored aggregated at the house energy meter. Disaggregating this power consumption into the contributions of individual devices can be achieved using Machine Learning. Our work aims at making state of the art disaggregation algorithms accessibe for users of the open source home automation platform Home Assistant.
Due to the decarbonization of the energy sector, the electric distribution grids are undergoing a major transformation, which is expected to increase the load on the operating resources due to new electrical loads and distributed energy resources. Therefore, grid operators need to gradually move to active grid management in order to ensure safe and reliable grid operation. However, this requires knowledge of key grid variables, such as node voltages, which is why the mass integration of measurement technology (smart meters) is necessary. Another problem is the fact that a large part of the topology of the distribution grids is not sufficiently digitized and models are partly faulty, which means that active grid operation management today has to be carried out largely blindly. It is therefore part of current research to develop methods for determining unknown grid topologies based on measurement data. In this paper, different clustering algorithms are presented and their performance of topology detection of low voltage grids is compared. Furthermore, the influence of measurement uncertainties is investigated in the form of a sensitivity analysis.
A method for detecting and approximating fault lines or surfaces, respectively, or decision curves in two and three dimensions with guaranteed accuracy is presented. Reformulated as a classification problem, our method starts from a set of scattered points along with the corresponding classification algorithm to construct a representation of a decision curve by points with prescribed maximal distance to the true decision curve. Hereby, our algorithm ensures that the representing point set covers the decision curve in its entire extent and features local refinement based on the geometric properties of the decision curve. We demonstrate applications of our method to problems related to the detection of faults, to multi-criteria decision aid and, in combination with Kirsch’s factorization method, to solving an inverse acoustic scattering problem. In all applications we considered in this work, our method requires significantly less pointwise classifications than previously employed algorithms.
The increasing share of renewable electricity in the grid drives the need for sufficient storage capacity. Especially for seasonal storage, power-to-gas can be a promising approach. Biologically produced methane from hydrogen produced from surplus electricity can be used to substitute natural gas in the existing infrastructure. Current reactor types are not or are poorly optimized for flexible methanation. Therefore, this work proposes a new reactor type with a plug flow reactor (PFR) design. Simulations in COMSOL Multiphysics ® showed promising properties for operation in laminar flow. An experiment was conducted to support the simulation results and to determine the gas fraction of the novel reactor, which was measured to be 29%. Based on these simulations and experimental results, the reactor was constructed as a 14 m long, 50 mm diameter tube with a meandering orientation. Data processing was established, and a step experiment was performed. In addition, a kLa of 1 h−1 was determined. The results revealed that the experimental outcomes of the type of flow and gas fractions are in line with the theoretical simulation. The new design shows promising properties for flexible methanation and will be tested.