Refine
Year of publication
Document Type
- Conference Proceeding (642) (remove)
Language
- English (492)
- German (149)
- Multiple languages (1)
Keywords
- 360-degree coverage (1)
- 3D Extended Object Tracking (1)
- 3D Extended Object Tracking (EOT) (2)
- 3D shape tracking (1)
- 3D ship detection (1)
- AAL (1)
- ADAM (1)
- AHI (1)
- Abrasive grain material (1)
- Abtragsprinzip (1)
Institute
- Fakultät Bauingenieurwesen (9)
- Fakultät Elektrotechnik und Informationstechnik (10)
- Fakultät Informatik (50)
- Fakultät Maschinenbau (9)
- Fakultät Wirtschafts-, Kultur- und Rechtswissenschaften (8)
- Institut für Angewandte Forschung - IAF (53)
- Institut für Optische Systeme - IOS (19)
- Institut für Strategische Innovation und Technologiemanagement - IST (29)
- Institut für Systemdynamik - ISD (64)
- Institut für Werkstoffsystemtechnik Konstanz - WIK (5)
Standardmäßig werden zur Modellierung magnetischer Systeme für regelungstechnische Anwendungen oder im Bereich der Diagnose und Prognose konzentriert parametrische Modelle verwendet. Falls eine hohe Qualität der Prozessabbildung erforderlich ist, z.B. um Wirbelströme oder Sättigung geeignet zu berücksichtigen, nehmen diese Modelle schnell relativ hohe Ordnungen an. Es ist seit einiger Zeit bekannt, dass verteilparametrische Systeme, die z.B. (Feld-)Diffusionsprozesse beinhalten, durch niederdimensionale Modelle mit nicht ganzzahligen Ableitungen, so genannte fraktionale Modelle, sehr gut abgebildet werden können. Im Bereich der magnetischen Aktuatoren wurden diese vor rund 10 Jahren zum ersten Mal untersucht. Seitdem wird auf diesem Gebiet in verschiedenen Arbeitsgruppen geforscht. Während im Frequenzbereich die Handhabung fraktionaler Systeme einfach ist, sind Anwendungen im Zeitbereich bisher insbesondere bei zeitkritischen Anwendungen kaum anzutreffen. Der Beitrag stellt die prinzipielle Idee dar und zeigt Möglichkeiten zum Einsatz dieser Verfahren im Bereich magnetischer Aktoren auf. In einer konkreten Anwendung wird in Simulation und Experiment demonstriert, wie mit Hilfe dieser Modelle Zustandsschätzung in Magnetaktuatoren erfolgen kann und welche Vorteile sich dadurch ergeben.
A lot of procedures for estimating the spool position in linear electromagnetic actuators using voltage and current measurements only, can be found in the literature. Subject to the accuracy of the estimated spool position some achieve better, some worse results. However, in almost every approach hysteresis has a huge impact on the estimation accuracy that can be achieved. Regardless whether these effects are caused by magnetic or mechanical hysteresis, they will limit the accuracy of the position estimate, if not taken into account. In this paper, a model is introduced which covers the hysteresis effects as well as other nonlinear ities occurring in estimated position-dependent parameters. A classical Preisach model is deployed first, which is then adjusted by using novel elementary preceding Relay-Operators. The resulting model for the estimated position-dependent parameters including the adjusted Preisach model can be easily applied to position estimation tasks. It is shown that the considered model distinctly improves the accuracy for the spool position estimate, while it is kept as simple as possible for real-time implementation reasons.
As fish farming is becoming more and more important worldwide, this ongoing project aims at the simulation and test-based analysis of highly stressed wire contacts, as they are found in off-shore fish farm cages in order to make them more reliable. The quasi-static tensile test of a wire mesh provides data for the construction of a finite element model to get a better understanding of the behavior of high-strength stainless steel from which the cages are made. Fatigue tests provide new insights that are used for an adjustment of the finite element model in order to predict the probability of possible damage caused by heavy mechanical loads (waves, storms, predators (sharks)).
Digitization extends to all areas of people's lives and processes, including public administration and government technology (GovTech for short). However, there are various problems here, such as the inappropriate development of new application systems, that are to be solved efficiently by combining two aspects: methodical digitization according to the process-driven approach and the idea of an app store for processes. This simultaneously fuels a process competition to advance methodical process digitization in the EU. Furthermore, this study explains the target-oriented use of this “firing” within the EU and concludes with a proposal of a new 3-schema architecture standard for successful process digitization within the EU.
This work introduces new signal constellations based on Eisenstein integers, i.e., the hexagonal lattice. These sets of Eisenstein integers have a cardinality which is an integer power of three. They are proposed as signal constellations for representation in the equivalent complex baseband model, especially for applications like physical-layer network coding or MIMO transmission where the constellation is required to be a subset of a lattice. It is shown that these constellations form additive groups where the addition over the complex plane corresponds to the addition with carry over ternary Galois fields. A ternary set partitioning is derived that enables multilevel coding based on ternary error-correcting codes. In the subsets, this partitioning achieves a gain of 4.77 dB, which results from an increased minimum squared Euclidean distance of the signal points. Furthermore, the constellation-constrained capacities over the AWGN channel and the related level capacities in case of ternary multilevel coding are investigated. Simulation results for multilevel coding based on ternary LDPC codes are presented which show that a performance close to the constellation-constrained capacities can be achieved.
It is well known that signal constellations which are based on a hexagonal grid, so-called Eisenstein constellations, exhibit a performance gain over conventional QAM ones. This benefit is realized by a packing and shaping gain of the Eisenstein (hexagonal) integers in comparison to the Gaussian (complex) integers. Such constellations are especially relevant in transmission schemes that utilize lattice structures, e.g., in MIMO communications. However, for coded modulation, the straightforward approach is to combine Eisenstein constellations with ternary channel codes. In this paper, a multilevel-coding approach is proposed where encoding and multistage decoding can directly be performed with state-of-the-art binary channel codes. An associated mapping and a binary set partitioning are derived. The performance of the proposed approach is contrasted to classical multilevel coding over QAM constellations. To this end, both the single-user AWGN scenario and the (multiuser) MIMO broadcast scenario using lattice-reduction-aided preequalization are considered. Results obtained from numerical simulations with LDPC codes complement the theoretical aspects.
With the advancement in sensor technology and the trend shift of health measurement from treatment after diagnosis to abnormalities detection long before the occurrence, the approach of turning private spaces into diagnostic spaces has gained much attention. In this work, we designed and implemented a low-cost and compact form factor module that can be deployed on the steering wheel of cars as well as most frequently touch objects at home in order to measure physiological signals from the fingertip of the subject as well as environmental parameters. We estimated the heart rate and SpO2 with the error of 2.83 bpm and 3.52%, respectively. The signal evaluation of skin temperature shows a promising output with respect to environmental recalibration. In addition, the electrodermal activity sensor followed the reference signal, appropriately which indicates the potential for further development and application in stress measurement.
Deep neural networks have become a veritable alternative to classic speaker recognition and clustering methods in recent years. However, while the speech signal clearly is a time series, and despite the body of literature on the benefits of prosodic (suprasegmental) features, identifying voices has usually not been approached with sequence learning methods. Only recently has a recurrent neural network (RNN) been successfully applied to this task, while the use of convolutional neural networks (CNNs) (that are not able to capture arbitrary time dependencies, unlike RNNs) still prevails. In this paper, we show the effectiveness of RNNs for speaker recognition by improving state of the art speaker clustering performance and robustness on the classic TIMIT benchmark. We provide arguments why RNNs are superior by experimentally showing a “sweet spot” of the segment length for successfully capturing prosodic information that has been theoretically predicted in previous work.
This work proposes a decoder implementation for high-rate generalized concatenated (GC) codes. The proposed codes are well suited for error correction in flash memories for high reliability data storage. The GC codes are constructed from inner extended binary Bose-Chaudhuri-Hocquenghem (BCH) codes and outer Reed-Solomon (RS) codes. The extended BCH codes enable high-rate GC codes. Moreover, the decoder can take advantage of soft information. For the first three levels of inner codes we propose an optional Chase soft decoder. In this work, the code construction is explained and a decoder architecture is presented. Furthermore, area and throughput results are discussed.
This work investigates soft input decoding for generalized concatenated (GC) codes. The GC codes are constructed from inner nested binary Bose-Chaudhuri-Hocquenghem (BCH)codes and outer Reed-Solomon (RS) codes. In order to enable soft input decoding for the inner BCH block codes, a sequential stack decoding algorithm is used. Ordinary stack decoding of binary block codes requires the complete trellis of the code.
In this work a representation of the block codes based on the trellises of supercodes is proposed in order to reduce the memory requirements for the representation of the BCH codes. Results for the decoding performance of the overall GC code are presented.
Furthermore, an efficient hardware implementation of the GC decoder is proposed.
The first part of this work shows the development and application of a new material system using high strength duplex stainless steel wires as net material with environmentally compatible antifouling properties for off-shore fish farm cages. Current net materials from textiles (polyamide) shall be partially replaced by high strength duplex stainless steel in order to have a more environmentally compatible system which meets the more severe mechanical loads (waves, storms, predatores (sharks, seals)). With a new antifouling strategy current issues like reduced ecological damage (e.g. due to copper disposal), lower maintenance costs (e.g. cleaning) and reduced durability shall be resolved.
High strength steel wires are also widely used in geological protection systems, for example rockfall protection or slope stabilisation. Normally hot-dip galvanised carbon steel is used in this case. But in highly corrosive environments like coastal areas, volcanic areas or mines for example, other solutions with a high corrosion resistance and sufficient mechanical properties are necessary. Protection systems made of high strength duplex stainless steel wires enable a significantly longer service life of the portection systems and therefore a higher level of security.
As a result of increasing needs and shrinking resources, aquaculture is gaining progressively significance in the recent years. Ecological issues such as negative effects on the ecological system due to the high fish density in the farms, the use of copper as antifouling strategy etc. are very present, particularly regarding the increasing number of fish going to be produced in farms in the future. Current trends focus on larger farms operated offshore. To make these farms working safe and economical, reliability has to be improved and maintenance costs need to be reduced. Also, alternatives with higher mechanical strength compared to current textile net materials as well as common metal wires might be necessary. In the last years, a new net system made of high strength duplex stainless steel wires with environmentally friendly antifouling properties suitable for offshore applications was developed. The first nets are operating for one year now as predator protection (i.e. seals) for fish farms and show a good performance in cleaning capability and predator protection. But in the real usage, some corrosion effects in the contact points of the net made of duplex stainless steel 1.4362 occur which were not observed in preliminary tests in laboratory and at different test sites around the world. These corrosion effects endanger the sustainable success of the net system. In this work, the observed corrosion effects are investigated. A laboratory test, which simulates the movement in the contact points of the net, was developed. Two pieces of wire are bent in the middle and get stucked into each other. One wire is fixed at both ends and the second wire is fixed on one end. On the other end, a circular movement with 1-2 rps and a 1 cm displacement is applied. The movement generates friction between the wires and the passive layer will be locally damaged. When the movement stops, a repassivation starts. The passivity breakdown and the repassivation were measured with electrochemical techniques. During the friction phase, when the surface will be activated, the open circuit potential breaks down. When the friction stops, the OCP increases. Between the movement phases, measurements of critical pitting potential were done. Thereby the quality of repassivation was investigated. The tests were done in a 3% sodium chloride solution. Different temperatures were tested as well as the influence of air saturation and low oxygen content.
Hot isostatic pressing (HIP) allows the production of complex components geometry. Generally, a high quality of the components is achieved due to the well managed composition of the metal powder and the non-isotropic properties. If a duplex stainless steel is produced, a heat treatment after the HIP-process is necessary to remove precipitations like carbides, nitrides and intermetallic phases. In a new process, the sintering step should be combined with the heat treatment. In this case a high cooling rate is necessary to avoid precipitations in duplex stainless steels. In this work, the influence of the HIP-temperature and the wall thickness on corrosion resistance, microstructure and impact strength were investigated. The results should help to optimize the process parameters like temperature and cooling rate. For the investigation, two HIP-temperatures were tested in a classical HIP-process step with a defined cooling rate. An additional heat treatment was not conducted. The specimens were cut from different sectors of the HIP-block. For investigation of the corrosion resistance, the critical pitting temperature was determined with electrochemical method according to EN ISO 17864. An impact test was used to determine the impact transition temperature. Metallographic investigations show the microstructure in the different sectors of the HIP-block.
Martensitic stainless steels has a wide use, for example for blades, knifes or cutter. The best corrosion resistance of these materials is in hardened condition. For better mechanical properties a tempering is normally applied to increase the durability. The tempering is also reducing the hardness and finally the corrosion resistance. Austempering is meanly used at low alloyed steels and brings a good compromise between durability, hardness and corrosion resistance. For martensitic stainless steels, austempering is normally not a topic because of the very long tempering times.
This work shows first results of austempering of some standard martensitic stainless steels and the influence to corrosion resistance. For reference, hardened and also hardened and tempered specimens were investigated. The corrosions resistance was investigated by electrochemical methods.
Influence of Temperature on the Corrosion behaviour of Stainless Steels under Tribological Stress
(2018)
Posterpräsentation
Deep transformation models
(2021)
We present a deep transformation model for probabilistic regression. Deep learning is known for outstandingly accurate predictions on complex data but in regression tasks it is predominantly used to just predict a single number. This ignores the non-deterministic character of most tasks. Especially if crucial decisions are based on the predictions, like in medical applications, it is essential to quantify the prediction uncertainty. The presented deep learning transformation model estimates the whole conditional probability distribution, which is the most thorough way to capture uncertainty about the outcome. We combine ideas from a statistical transformation model (most likely transformation) with recent transformation models from deep learning (normalizing flows) to predict complex outcome distributions. The core of the method is a parameterized transformation function which can be trained with the usual maximum likelihood framework using gradient descent. The method can be combined with existing deep learning architectures. For small machine learning benchmark datasets, we report state of the art performance for most dataset and partly even outperform it. Our method works for complex input data, which we demonstrate by employing a CNN architecture on image data.
This work is a study about a comparison of survey tools and it should help developers in selecting a suited tool for application in an AAL environment. The first step was to identify the basic required functionality of the survey tools used for AAL technologies and to compare these tools by their functionality and assignments. The comparative study was derived from the data obtained, previous literature studies and further technical data. A list of requirements was stated and ordered in terms of relevance to the target application domain. With the help of an integrated assessment method, the calculation of a generalized estimate value was performed and the result is explained. Finally, the planned application of this tool in a running project is explained.
Ballistocardiography (BCG) can be used to monitor heart rate activity. Besides, the accelerometer should have high sensitivity and minimal internal noise; a low-cost approach was taken into consideration. Several measurements have been executed to determine the optimal positioning of a sensor under the mattress to obtain a signal strong enough for further analysis. A prototype for an unobtrusive accelerometer-based measurement system has been developed and tested in a conventional bed without any specific extras. The influence of the human sleep position for the output accelerometer data was tested. The obtained results indicate the potential to capture BCG signals using accelerometers. The measurement system can detect heart rate in an unobtrusive form in the home environment.
The digital twin concept has been widely known for asset monitoring in the industry for a long time. A clear example is the automotive industry. Recently, there has also been significant interest in the application of digital twins in healthcare, especially in genomics in what is known as precision medicine. This work focuses on another medical speciality where digital twins can be applied, sleep medicine. However, there is still great controversy about the fundamentals that constitute digital twins, such as what this concept is based on and how it can be included in healthcare effectively and sustainably. This article reviews digital twins and their role so far in what is known as personalized medicine. In addition, a series of steps will be exposed for a possible implementation of a digital twin for a patient suffering from sleep disorders. For this, artificial intelligence techniques, clinical data management, and possible solutions for explaining the results derived from artificial intelligence models will be addressed.
The use of deep learning models with medical data is becoming more widespread. However, although numerous models have shown high accuracy in medical-related tasks, such as medical image recognition (e.g. radiographs), there are still many problems with seeing these models operating in a real healthcare environment. This article presents a series of basic requirements that must be taken into account when developing deep learning models for biomedical time series classification tasks, with the aim of facilitating the subsequent production of the models in healthcare. These requirements range from the correct collection of data, to the existing techniques for a correct explanation of the results obtained by the models. This is due to the fact that one of the main reasons why the use of deep learning models is not more widespread in healthcare settings is their lack of clarity when it comes to explaining decision making.
oday many scientific works are using deep learning algorithms and time series, which can detect physiological events of interest. In sleep medicine, this is particularly relevant in detecting sleep apnea, specifically in detecting obstructive sleep apnea events. Deep learning algorithms with different architectures are used to achieve decent results in accuracy, sensitivity, etc. Although there are models that can reliably determine apnea and hypopnea events, another essential aspect to consider is the explainability of these models, i.e., why a model makes a particular decision. Another critical factor is how these deep learning models determine how severe obstructive sleep apnea is in patients based on the apnea-hypopnea index (AHI). Deep learning models trained by two approaches for AHI determination are exposed in this work. Approaches vary depending on the data format the models are fed: full-time series and window-based time series.
The development of automatic solutions for the detection of physiological events of interest is booming. Improvements in the collection and storage of large amounts of healthcare data allow access to these data faster and more efficiently. This fact means that the development of artificial intelligence models for the detection and monitoring of a large number of pathologies is becoming increasingly common in the medical field. In particular, developing deep learning models for detecting obstructive apnea (OSA) events is at the forefront. Numerous scientific studies focus on the architecture of the models and the results that these models can provide in terms of OSA classification and Apnea-Hypopnea-Index (AHI) calculation. However, little focus is put on other aspects of great relevance that are crucial for the training and performance of the models. Among these aspects can be found the set of physiological signals used and the preprocessing tasks prior to model training. This paper covers the essential requirements that must be considered before training the deep learning model for obstructive sleep apnea detection, in addition to covering solutions that currently exist in the scientific literature by analyzing the preprocessing tasks prior to training.
The corporate entrepreneur
(2016)
Corporate entrepreneurship is one tool for established companies to strengthen their capabilities for strategic renewal and innovativeness. The question, however, which factors are influencing the success of a corporate entrepreneurship initiative requires further attention. The corporate entrepreneur who is acting as both the leader of embedded entrepreneurial teams and linking pin to the corporate, is providing one possible perspective. Based on 6 interviews conducted in 6 German organizations this study contributes to the understanding of the role of the corporate entrepreneur and how this role can be distinguished from other roles in the context of innovation.
Today’s markets are characterized by fast and radical changes, posing an essential challenge to established companies. Startups, yet, seem to be more capable in developing radical innovations to succeed in those volatile markets. Thus, established companies started to experiment with various approaches to implement startup-like structures in their organization. Internal corporate accelerators (ICAs) are a novel form of corporate venturing, aiming to foster bottom-up innovations through intrapreneurship. However, ICAs still lack empirical investigations. This work contributes to a deeper understanding of the interface between the ICA and the core organization and the respective support activities (resource access and support services) that create an innovation-supportive work environment for the intrapreneurial team. The results of this qualitative study, comprising 12 interviews with ICA teams out of two German high-tech companies, show that the resources provided by ICAs differ from the support activities of external accelerators. Further, the study shows that some resources show both supportive as well as obstructive potential for the intrapreneurial teams within the ICA.
Corporate venturing has gained much attention due
to challenges and changes that occur because of discontinuous
innovations – which seem to be promoted by digitalization. In this
context, open innovation has become a promising tool for
established companies to strengthen their innovation capabilities.
While the external opening of the innovation process has gained
much attention, the internal opening lacks on investigations.
Especially new organizational forms, such as Internal Corporate
Accelerators, have not been investigated sufficiently. This study,
which is based on 13 interviews from two German tech-companies,
contributes to a better understanding of this new form of corporate
venturing and the resulting effects on the organizational renewal.
Corporate entrepreneurship (CE) supports the strategic renewal of established companies. Corporate venturing represents one key concept of CE that supports companies to strengthen their innovation capabilities. For the successful implementation of corporate ventures dual structures are recommended. The question, how the interface should be designed, plays a crucial role. Although it seems to be an important factor, this aspect requires further attention. One relevant element of the interface design are the different roles of the individuals that are interacting within the interface. This study is based on nine interviews that are representing six internal corporate ventures within one large German corporate from the ICT sector. The results that were mirrored with short case studies of 25 additional companies of the data sample, contribute to a better understanding of the interface design by adding insights about roles in corporate entrepreneurship. This deeper understanding about roles allows to draw conclusions on the interface design from a structural point of view.
In today's volatile world, established companies must be capable of optimizing their core business with incremental innovations while simultaneously developing discontinuous innovations to maintain their long-term competitiveness. Balancing both is a major challenge for companies, since different types of innovation require different organizational structures, operational modes and management styles. Established companies tend to excel in improving their current business through incremental innovations which are closely related to their current knowledge base and competencies. However, this often goes hand in hand with challenges in the exploration of knowledge that is new to the company and that is essential for the development of discontinuous innovations. In this respect, the concept of corporate entrepreneurship is recognized as a way to strengthen the exploration of new knowledge and to support the development of discontinuous innovation. For managing corporate entrepreneurship more effectively, it is crucial to understand which types of knowledge can be created through corporate entrepreneurship and which organizational designs are more suited to gain certain types of knowledge. To answer these questions, this study analyzed 23 semi-structured interviews conducted with established companies that are running such entrepreneurial activities. The results show (1) that three general types of knowledge can be explored through corporate entrepreneurship and (2) that some organizational designs are more suited to explore certain knowledge types than others are.
Corporate entrepreneurship (CE) is experiencing continuously increasing interest from scholars and practitioners. One reason for this seems to be rooted in the organizational structures of established companies, which are cumbersome for implementing organizational agility and for developing radical innovations. In view of the advancing digitalization, however, exactly this is required in order to be successful in the long-term. CE is a promising managerial tool that offers a wide range of options to pursue the creation of new businesses and to support the companies' transformation in order to adapt to changes in the environment. Even though CE offers a broad range of opportunities, the effective management is a challenge. One reason for this is the ambiguity when it comes to the differences between the various CE forms and the objectives that can be achieved by those. This study, which is based on 13 in-depth interviews from eight high-tech companies, contributes to a better understanding of CE by offering a first harmonized set of CE objectives that is suitable to compare and differentiate across the different forms. In addition to that, three CE types, offering a new perspective on how to differentiate CE forms, are identified and give implications for a more effective management.
In diesem Beitrag wird eine Methode des maschinellen Lernens entwickelt, die die Schlafstadienerkennung untersucht. Übliche Methoden der Schlafanalyse basieren auf der Polysomnographie (PSG). Der präsentierte Ansatz basiert auf Signalen, die ausschließlich nicht-invasiv in einer häuslichen Umgebung gemessen werden können. Bewegungs-, Herzschlags- und Atmungssignale können vergleichsweise leicht erfasst werden aber die Erkennung der Schlafstadien ist dadurch erschwert. Die Signale werden als Zeitreihenfolge strukturiert und in Epochen überführt. Die Leistungsfähigkeit von maschinellem Lernen wird der Polysomnographie gegenübergestellt und bewertet.
The citizen-centered health platform project is intended to provide a platform that can be used in EU cross-border regions, where social and economic exchange occurs across national borders. The overriding challenges are: (a) social: improving citizen-centered health and care provision; (b) technical: providing a digital platform for networking citizens, service providers, and municipal actors; (c) economic: developing long-term successful (sustainable) business models/value chains. The platform should strengthen and expand existing networks and establish new regional networks. Each network addresses particular challenges and apply them in a region-specific manner. Here, the national boundary conditions and the interregional needs play an essential role. These objectives require sufficient participation of civil society representatives. Furthermore, the platform will establish an overarching, sustainable, and knowledge-based network of health experts. The platform is to be jointly developed and implemented in the regions and follow an open-access approach. Therefore, synergies will be shared more quickly, strengthening competencies and competitiveness. In addition to practice partners, scientific and municipal institutions and SMEs are involved. The actors thus contribute to scientific performance, innovative strength, and resilience.
Autism spectrum disorders (ASD) affect a large number of children both in the Russian Federation and in Germany. Early diagnosis is key for these children, because the sooner parents notice such disorders in a child and the rehabilitation and treatment program starts, the higher the likelihood of his social adaptation. The difficulties in raising such a child lie in the complexity of his learning outside of children's groups and the complexity of his medical care. In this regard, the development of digital applications that facilitate medical care and education of such children at home is important and relevant. The purpose of the project is to improve the availability and quality of healthcare and social adaptation at home of children with ASD through the use of digital technologies.
Sleep is a multi-dimensional influencing factor on physical health, cognitive function, emotional well-being, mental health, daily performance, and productivity. The barriers such as time-consuming, invasiveness, and expense have caused a gradual shift in sleep monitoring from traditional and standard in-lab approach, e. g., polysomnography (PSG) to unobtrusive and noninvasive in-home sleep monitoring, yet further improvement is required. Despite an increasing interest in fiberoptic-based methods for cardiorespiratory estimation, the traditional mechanical-based sensors consist of force-sensitive resistors (FSR), lead zirconate titanate piezoelectric (PZT), and accelerometers yet serve as the dominant approach. The part of popularity lies in reducing the system’s complexity, expense, easy maintenance, and user-friendliness. However, care must be taken regarding the performance of such sensors with respect to accuracy and calibration.
The development of home health systems can provide continuous and user-friendly monitoring of key health parameters. This project aims to create a concept for such a system, implement it on a test basis, and evaluate it. Three health areas were selected for this purpose:
Sleep, Stress, and Rehabilitation. Appropriate devices were installed in the homes of test subjects and used by them for two weeks. Besides, relevant questionnaires were completed to obtain a complete picture. Finally, the implemented system was evaluated, and the results of the conducted study showed that home health systems have great potential. However, it is necessary to consider some points to increase the usability of the system and the motivation of the users. Among others, ease of use of the equipment is of extreme importance.
Long-term sleep monitoring can be done primarily in the home environment. Good patient acceptance requires low user and installation barriers. The selection of parameters in this approach is significantly limited compared to a PSG session. The aim is a qualified selection of parameters, which on the one hand allow a sufficiently good classification of sleep phases and on the other hand can be detected by non-invasive methods.
Assistive environments are entering our homes faster than ever. However, there are still various barriers to be broken. One of the crucial points is a personalization of offered services and integration of assistive technologies in common objects and therefore in a regular daily routine. Recognition of sleep patterns for the preliminary sleep study is one of the health services that could be performed in an undisturbing way. This article proposes the hardware system for the measurement of bio-vital signals necessary for initial sleep study in a non-obtrusive way. The first results confirm the potential of measurement of breathing and movement signals with the proposed system.
Present demographic change and a growing population of elderly people leads to new medical needs. Meeting these with state of the art technology is as a consequence a rapidly growing market. So this work is aimed at taking modern concepts of mobile and sensor technology and putting them in a medical context. By measuring a user’s vital signs on sensors which are processed on a Android smartphone, the target system is able to determine the current health state of the user and to visualize gathered information. The system also includes a weather forecasting functionality, which alerts the user on possibly dangerous future meteorological events. All information are collected centrally and distributed to users based on their location. Further, the system can correlate the client-side measurement of vital signs with a server-side weather history. This enables personalized forecasting for each user individually. Finally, a portable and affordable application was developed that continuously monitors the health status by many vital sensors, all united on a common smartphone.
Für die Überwachung des Schlafs zu Hause sind nichtinvasive Methoden besonders gut anwendbar. Die Signale, die häufig überwacht werden, sind Herzfrequenz und Atemfrequenz. Die Ballistokardiographie (BCG)ist eine Technik, bei der die Herzfrequenz aus den mechanischen Schwingungen des Körpers bei jedem Herzzyklus gemessen wird. Kürzlich wurden Übersichtsarbeiten veröffentlicht. Die Untersuchung soll in einem ersten Ansatz bewerten, ob die Herzfrequenz anhand von BCG erkannt werden kann. Die wesentlichen Randbedingungen sind, ob dies gelingt, wenn der Sensor unter der Matratze positioniert wird und kostengünstige Sensoren zum Einsatz kommen.
Measuring cardiorespiratory parameters in sleep, using non-contact sensors and the Ballistocardiography technique has received much attention due to the low-cost, unobtrusive, and non-invasive method. Designing a user-friendly, simple-to-use, and easy-to-deployment preserving less errorprone remains open and challenging due to the complex morphology of the signal. In this work, using four forcesensitive resistor sensors, we conducted a study by designing four distributions of sensors, in order to simplify the complexity of the system by identifying the region of interest for heartbeat and respiration measurement. The sensors are deployed under the mattress and attached to the bed frame without any interference with the subjects. The four distributions are combined in two linear horizontal, one linear vertical, and one square, covering the influencing region in cardiorespiratory activities. We recruited 4 subjects and acquired data in four regular sleeping positions, each for a duration of 80 seconds. The signal processing was performed using discrete wavelet transform bior 3.9 and smooth level of 4 as well as bandpass filtering. The results indicate that we have achieved the mean absolute error of 2.35 and 4.34 for respiration and heartbeat, respectively. The results recommend the efficiency of a triangleshaped structure of three sensors for measuring heartbeat and respiration parameters in all four regular sleeping positions.
Differences in the pitting resistance between cold worked CrNi and CrNiMnN metastable austenites
(2015)
Fachvortrag auf dem Kongress CORROSION 2015, 15-19 March, Dallas, Texas, USA. NACE International
Effect of cold working on the localized corrosion behavior of CrNi and CrNiMnN metastable austenites
(2015)
This paper aims to apply the basics of the Service-Dominant Logic, especially the concept of creating benefits through serving, to the stationary retail industry. In the industrial context, the shift from a product-driven point of view to a service-driven perspective has been discussed widely. However, there are only few connections to how this can be applied to the retail sector on a B2C-level and how retailers can use smart services in order to enable customer engagement, loyalty and retention. The expectations of customers towards future stationary retail develop significantly as consumers got used to the comfort of online shopping. Especially the younger generation—the Generation Z—seems to have changed their priorities from the bare purchase of products to an experience- and service-driven approach when shopping over-the-counter. To stay successful long-term, companies from this sector need to adapt to the expectations of their future main customer group. Therefore, this paper will analyse the specific needs of Generation Z, explain how smart services contribute to creating benefit for this customer group and how this affects the economic sustainability of these firms.
Ein neuzeitlicher Pfahlbau
(2016)
Bauhäusler am Bodensee
(2016)
Probabilistic data association for tracking extended targets under clutter using random matrices
(2015)
The use of random matrices for tracking extended objects has received high attention in recent years. It is an efficient approach for tracking objects that give rise to more than one measurement per time step. In this paper, the concept of random matrices is used to track surface vessels using highresolution automotive radar sensors. Since the radar also receives a large number of clutter measurements from the water, for the data association problem, a generalized probabilistic data association filter is applied. Additionally, a modification of the filter update step is proposed to incorporate the Doppler velocity measurements. The presented tracking algorithm is validated using Monte Carlo Simulation, and some performance results with real radar data are shown as well.
Small vessels or unmanned surface vehicles only have a limited amount of space and energy available. If these vessels require an active sensing collision avoidance system it is often not possible to mount large sensor systems like X-Band radars. Thus, in this paper an energy efficient automotive radar and a laser range sensor are evaluated for tracking surrounding vessels. For these targets, those type of sensors typically generate more than one detection per scan. Therefore, an extended target tracking problem has to be solved to estimate state end extension of the vessels. In this paper, an extended version of the probabilistic data association filter that uses random matrices is applied. The performance of the tracking system using either radar or laser range data is demonstrated in real experiments.
Magnetic effects on austenitic stainless steels, formed during a low temperature carburizing depending on the alloy composition are discussed in this paper. Samples of different austenitic stainless steel alloys have been subjected to a multiple low-temperature carburization. Layer characteriszation with light microscope and hardness profiles show a growth of the layer thickness. The formation of an expanded austenite layer (lattice expansion) could be detected by X-ray diffraction (XRD). Feritscope was used to determine the magnetizability, whereby not all austenitic alloys form a magnetizability after treatment. Furthermore, test procedures were developed to visualize the magnetizability. For this purpose, magnetic force microscope measurements and investigations with ferrofluid were carried out and a fir tree ferromagnetic layer strucure could be proven.
Investigation of magnetic effects on austenitic stainless steels after low temperature carburization
(2018)
Fachvortrag auf der 10th International European Stainless Steel Conference and 6th European Duplex Stainless Steel Conference (ESSC & DUPLEX 2019), 30.09. – 02.10.2019, Vienna, Austria
Systematic Generation of XSS and SQLi Vulnerabilities in PHP as Test Cases for Static Code Analysis
(2022)
Synthetic static code analysis test suites are important to test the basic functionality of tools. We present a framework that uses different source code patterns to generate Cross Site Scripting and SQL injection test cases. A decision tree is used to determine if the test cases are vulnerable. The test cases are split into two test suites. The first test suite contains 258,432 test cases that have influence on the decision trees. The second test suite contains 20 vulnerable test cases with different data flow patterns. The test cases are scanned with two commercial static code analysis tools to show that they can be used to benchmark and identify problems of static code analysis tools. Expert interviews confirm that the decision tree is a solid way to determine the vulnerable test cases and that the test suites are relevant.
Many secure software development methods and tools are well-known and understood. Still, the same software security vulnerabilities keep occurring. To find out if new source code patterns evolved or the same patterns are reoccurring, we investigate SQL injections in PHP open source projects. SQL injections are well-known and a core part of software security education. For each common part of SQL injections, the source code patterns are analysed. Examples are pointed out showing that developers had software security in mind, but nevertheless created vulnerabilities. A comparison to earlier work shows that some categories are not found as often as expected. Our main contribution is the categorization of source code patterns.
We present source code patterns that are difficult for modern static code analysis tools. Our study comprises 50 different open source projects in both a vulnerable and a fixed version for XSS vulnerabilities reported with CVE IDs over a period of seven years. We used three commercial and two open source static code analysis tools. Based on the reported vulnerabilities we discovered code patterns that appear to be difficult to classify by static analysis. The results show that code analysis tools are helpful, but still have problems with specific source code patterns. These patterns should be a focus in training for developers.
We compared vulnerable and fixed versions of the source code of 50 different PHP open source projects based on CVE reports for SQL injection vulnerabilities. We scanned the source code with commercial and open source tools for static code analysis. Our results show that five current state-of-the-art tools have issues correctly marking vulnerable and safe code. We identify 25 code patterns that are not detected as a vulnerability by at least one of the tools and 6 code patterns that are mistakenly reported as a vulnerability that cannot be confirmed by manual code inspection. Knowledge of the patterns could help vendors of static code analysis tools, and software developers could be instructed to avoid patterns that confuse automated tools.
To get a better understanding of Cross Site Scripting vulnerabilities, we investigated 50 randomly selected CVE reports which are related to open source projects. The vulnerable and patched source code was manually reviewed to find out what kind of source code patterns were used. Source code pattern categories were found for sources, concatenations, sinks, html context and fixes. Our resulting categories are compared to categories from CWE. A source code sample which might have led developers to believe that the data was already sanitized is described in detail. For the different html context categories, the necessary Cross Site Scripting prevention mechanisms are described.
We investigated 50 randomly selected buffer overflow vulnerabilities in Firefox. The source code of these vulnerabilities and the corresponding patches were manually reviewed and patterns were identified. Our main contribution are taxonomies of errors, sinks and fixes seen from a developer's point of view. The results are compared to the CWE taxonomy with an emphasis on vulnerability details. Additionally, some ideas are presented on how the taxonomy could be used to improve the software security education.
This paper summarizes the trends in metallization and interconnection technology in the eyes of the participants of the 8th Metallization and Interconnection Workshop. Participants were asked in a questionnaire to share their view on the future development of metallization technology, the kind of metal used for front side metallization and the future development of interconnection technology. The continuous improvement of the screen-printing technology is reflected in the high expected percentage share decreasing from 88% in three years to still 70% in ten years. The dominating front side metal in the view of the participants will be silver with an expected percentage share of nearly 70% in 2029. Regarding interconnection technologies, the experts of the workshop expect new technologies to gain significant technology shares faster. Whereas in three years soldering on busbars is expected to dominate with a percentage share of 71% it will drop in ten years to 35% in the eyes of the participants. Multiwire and shingling technologies are seen to have the highest potential with expected percentage shares of 33% (multiwire) and 16% (shingling) in ten years.
Corrosion
(2016)
The paper investigates an innovative actuator combination based on the magnetic shape memory technology. The actuator is composed of an electromagnet, which is activated to produce motion, and a magnetic shape memory element, which is used passively to yield multistability, i.e. the possibility of holding a position without input power. Based on the experimental open-loop frequency characterization of the actuator, a position controller is developed and tested in several experiments.
Think BIQ: Gender Differences, Entrepreneurship Support and the Quality of Business Idea Description
(2023)
Entrepreneurship support, its influencing factors and female entrepreneurship are recently discussed topics with great relevance for society and politics. However, research on the subject has been divergent in its results and lacks a focus on the impact of support programs’ characteristics concerning different types of entrepreneurs. Thus, we conduct a fuzzy-set Qualitative Comparative Analysis on entrepreneurship support characteristics aiming to shed light on possible gender differences occurring in respective programs. We investigate the quality of business idea descriptions, as a predecessor for a high-potential business model, operationalized using inter alia causation and effectuation theory and social role theory as possible explanations. In our fuzzy-set Qualitative Comparative Analysis on a sample of 911 Norwegian ventures, we find a variety of differences related to the entrepreneurs’ gender. For instance, that financial support combined with a well described key contribution or careful planning seem to be more important antecedents for female entrepreneurs’ business idea quality than for males. Moreover, it seems a well-described key contribution has a positive effect on the outcome variable in most cases. Another interesting finding concerns the entrepreneurs’ network partners, where we found evident gender differences in our combinations. Female entrepreneurs seemingly benefitted from rather small networks, and males from big networks, although the former possess larger networks in the sample. In conclusion, we find that gender differences in combinations of entrepreneurship support for high business idea quality still occur even in a country like Norway, calling for an adaption of the provided support and environment.
Szenografie ist eine Universaldisziplin, die mit Raum, Licht, Grafik, Ton und digitalen Medien Erlebnisräume schafft, in denen Wissen vermittelt und Besuchern verborgene Zusammenhänge intuitiv zugänglich gemacht werden können. Hierbei spielen mediale Vermittlungsstrategien eine zentrale Rolle. Eine dieser Strategien ist Information on demand: kontextualisierende Informationen werden hier nicht permanent angeboten, sondern nur dann, wenn der Besucher sie abruft. Dies ermöglicht sowohl eine auratische Objektpräsentation als auch gleichermaßen Kontextualisierung und Informationsvermittlung. Eine weitere Strategie sind
personalisierte Devices, die zudem eine zielgruppenorientierte Ansprache von Besuchern ermöglichen. Darüber hinaus bieten VR- und AR-Technologien die Möglichkeit zur publikumswirksamen Darstellung komplexer wissenschaftlicher Zusammenhänge.
Die Überwindung des Bruchs (Seam) beim Lernen im Studium zwischen dem Hochschulkontext und der beruflichen Praxis ist durch die zeitlich, räumlich und organisatorisch bedingte Trennung der relevanten Akteure (u. a. Lehrende, Lernende, Unternehmensvertreter) eine sehr große Herausforderung (Milrad et al., 2013). Eine seamless-learning-basierte Konzeption einer Lehrveranstaltung auf Basis agiler Werte und Methoden (u. a. inkrementelles Vorgehen, Fokus auf lernendenzentrierte Veranstaltungen, individualisiertes Lernenden-Feedback) kann bei der Überwindung dieses bedeutenden Bruchs helfen. In dem Poster wird das grundsätzliche Design eines derartigen agilen SL-Konzepts auf Basis eines iterativ, inkrementellen Vorgehens innerhalb eines Semesterzyklus von 15 Wochen in drei Lernsprints erörtert. Darüber hinaus wird über erste Lehrerfahrungen der Dozierenden sowohl aus der Hochschule als auch aus dem industriellen Umfeld und Lernerfahrungen der Studierenden aus den vergangenen zwei Jahren berichtet.
The magneto-mechanical behavior of magnetic shape memory (MSM) materials has been investigated by means of different simulation and modeling approaches by several research groups. The target of this paper is to simulate actuators driven by MSM alloys and to understand the MSM element behavior during actuation, which shall lead to an increased performance of the actuator. It is shown that internal and external stresses should be taken into consideration using numerical computation tools for magnetic fields in an efficient way.
Stress is a recognized as a predominant disease with growing costs of treatment. The approach presented here is aimed to detect stress using a light weighted, mobile, cheap and easy to use system. The result shows that stress can be detected even in case a person’s natural bio vital data is out of the main range. The system enables storage of measured data, while maintaining communication channels of online and post-processing.
The principal objective of this study is to investigate the impact of perceived stress on traffic and road safety. Therefore, we designed a study that allows the generation and collection of stress-relevant data. Drivers often experience stress due to their perception of lack of control during the driving process. This can lead to an increased likelihood of traffic accidents, driver errors, and traffic violations. To explore this phenomenon, we used the Stress Perceived Questionnaire (PSQ) to evaluate perceived stress levels during driving simulations and the EPQR questionnaire to determine the personality of the driver. With the presented study, participants can categorised based on their emotional stability and personality traits. Wearable devices were utilised to monitor each participant's instantaneous heart rate (HR) due to their non-intrusive and portable nature. The findings of this study deliver an overview of the link between stress and traffic and road safety. These findings can be utilised for future research and implementing strategies to reduce road accidents and promote traffic safety.
The purpose of this paper is to examine the effects of perceived stress on traffic and road safety. One of the leading causes of stress among drivers is the feeling of having a lack of control during the driving process. Stress can result in more traffic accidents, an increase in driver errors, and an increase in traffic violations. To study this phenomenon, the Stress Perceived Questionnaire (PSQ) was used to evaluate the perceived stress while driving in a simulation. The study was conducted with participants from Germany, and they were grouped into different categories based on their emotional stability. Each participant was monitored using wearable devices that measured their instantaneous heart rate (HR). The preference for wearable devices was due to their non-intrusive and portable nature. The results of this study provide an overview of how stress can affect traffic and road safety, which can be used for future research or to implement strategies to reduce road accidents and promote traffic safety.
Methods based exclusively on heart rate hardly allow to differentiate between physical activity, stress, relaxation, and rest, that is why an additional sensor like activity/movement sensor added for detection and classification. The response of the heart to physical activity, stress, relaxation, and no activity can be very similar. In this study, we can observe the influence of induced stress and analyze which metrics could be considered for its detection. The changes in the Root Mean Square of the Successive Differences provide us with information about physiological changes. A set of measurements collecting the RR intervals was taken. The intervals are used as a parameter to distinguish four different stages. Parameters like skin conductivity or skin temperature were not used because the main aim is to maintain a minimum number of sensors and devices and thereby to increase the wearability in the future.
The investigation of stress requires to distinguish between stress caused by physical activity and stress that is caused by psychosocial factors. The behaviour of the heart in response to stress and physical activity is very similar in case the set of monitored parameters is reduced to one. Currently, the differentiation remains difficult and methods which only use the heart rate are not able to differentiate between stress and physical activity, without using additional sensor data input. The approach focusses on methods which generate signals providing characteristics that are useful for detecting stress, physical activity, no activity and relaxation.
The perception of the amount of stress is subjective to every person, and the perception of it changes depending on many factors. One of the factors that has an impact on perceived stress is the emotional state. In this work, we compare the emotional state of 40 German driving students and present different partitions that can be advantageous for using artificial intelligence and classification. Like this, we evaluate the data quality and prepare for the specific use. The Stress Perceived Questionnaire (PSQ20) was employed to assess the level of stress experienced by individuals while participating in a driving simulation for 5 and 25 min. As a result of our analysis, we present a categorisation of various emotional states into intervals, comparing different classifications and facilitating a more straightforward implementation of artificial intelligence for classification purposes.
Stress is recognized as a predominant disease with raising costs for rehabilitation and treatment. Currently there several different approaches that can be used for determining and calculating the stress levels. Usually the methods for determining stress are divided in two categories. The first category do not require any special equipment for measuring the stress. This category useless the variation in the behaviour patterns that occur while stress. The core disadvantage for the category is their limitation to specific use case. The second category uses laboratories instruments and biological sensors. This category allow to measure stress precisely and proficiently but on the same time they are not mobile and transportable and do not support real-time feedback. This work presents a mobile system that provides the calculation of stress. For achieving this, the of a mobile ECG sensor is analysed, processed and visualised over a mobile system like a smartphone. This work also explains the used stress measurement algorithm. The result of this work is a portable system that can be used with a mobile system like a smartphone as visual interface for reporting the current stress level.
Stress and physical activities are important aspects of life of people. Body reactions on stress and on physical activities can be very similar but long-term stress leads to diseases and damages the body. Currently there is no method to differentiate easily and clearly between these two aspects in a time slot. We have confronted this problem while developing a mobile system for detection and analysis of stress. This paper presents an approach, which uses a long-term monitor with ECG/EKG capabilities and analysis of the heart rate data that is extracted from the device. The focus of the work is to find characteristics that are useful for differentiation between physical activity and stress.
Stress is recognized as a factor of predominant disease and in the future the costs for treatment will increase. The presented approach tries to detect stress in a very basic and easy to implement way, so that the cost for the device and effort to wear it remain low. The user should benefit from the fact that the system offers an easy interface reporting the status of his body in real time. In parallel, the system provides interfaces to pass the obtained data forward for further processing and (professional) analyses, in case the user agrees. The system is designed to be used in every day’s activities and it is not restricted to laboratory use or environments. The implementation of the enhanced prototype shows that the detection of stress and the reporting can be managed using correlation plots and automatic pattern recognition even on a very light-weighted microcontroller platform.
In previous studies, we used a method for detecting stress that was based exclusively on heart rate and ECG for differentiation between such situations as mental stress, physical activity, relaxation, and rest. As a response of the heart to these situations, we observed different behavior in the Root Mean Square of the Successive differences heartbeats (RMSSD). This study aims to analyze Virtual Reality via a virtual reality headset as an effective stressor for future works. The value of the Root Mean Square of the Successive Differences is an important marker for the parasympathetic effector on the heart and can provide information about stress. For these measurements, the RR interval was collected using a breast belt. In these studies, we can observe the Root Mean Square of the successive differences heartbeats. Additional sensors for the analysis were not used. We conducted experiments with ten subjects that had to drive a simulator for 25 minutes using monitors and 25 minutes using virtual reality headset. Before starting and after finishing each simulation, the subjects had to complete a survey in which they had to describe their mental state. The experiment results show that driving using virtual reality headset has some influence on the heart rate and RMSSD, but it does not significantly increase the stress of driving.
Cardiovascular diseases are directly or indirectly responsible for up to 38.5% of all deaths in Germany and thus represent the most frequent cause of death. At present, heart diseases are mainly discovered by chance during routine visits to the doctor or when acute symptoms occur. However, there is no practical method to proactively detect diseases or abnormalities of the heart in the daily environment and to take preventive measures for the person concerned. Long-term ECG devices, as currently used by physicians, are simply too expensive, impractical, and not widely available for everyday use. This work aims to develop an ECG device suitable for everyday use that can be worn directly on the body. For this purpose, an already existing hardware platform will be analyzed, and the corresponding potential for improvement will be identified. A precise picture of the existing data quality is obtained by metrological examination, and corresponding requirements are defined. Based on these identified optimization potentials, a new ECG device is developed. The revised ECG device is characterized by a high integration density and combines all components directly on one board except the battery and the ECG electrodes. The compact design allows the device to be attached directly to the chest. An integrated microcontroller allows digital signal processing without the need for an additional computer. Central features of the evaluation are a peak detection for detecting R-peaks and a calculation of the current heart rate based on the RR interval. To ensure the validity of the detected R-peaks, a model of the anatomical conditions is used. Thus, unrealistic RR-intervals can be excluded. The wireless interface allows continuous transmission of the calculated heart rate. Following the development of hardware and software, the results are verified, and appropriate conclusions about the data quality are drawn. As a result, a very compact and wearable ECG device with different wireless technologies, data storage, and evaluation of RR intervals was developed. Some tests yelled runtimes up to 24 hours with wireless Lan activated and streaming.
Nowadays, the inexpensive memory space promotes an accelerating growth of stored image data. To exploit the data using supervised Machine or Deep Learning, it needs to be labeled. Manually labeling the vast amount of data is time-consuming and expensive, especially if human experts with specific domain knowledge are indispensable. Active learning addresses this shortcoming by querying the user the labels of the most informative images first. One way to obtain the ‘informativeness’ is by using uncertainty sampling as a query strategy, where the system queries those images it is most uncertain about how to classify. In this paper, we present a web-based active learning framework that helps to accelerate the labeling process. After manually labeling some images, the user gets recommendations of further candidates that could potentially be labeled equally (bulk image folder shift). We aim to explore the most efficient ‘uncertainty’ measure to improve the quality of the recommendations such that all images are sorted with a minimum number of user interactions (clicks). We conducted experiments using a manually labeled reference dataset to evaluate different combinations of classifiers and uncertainty measures. The results clearly show the effectiveness of an uncertainty sampling with bulk image shift recommendations (our novel method), which can reduce the number of required clicks to only around 20% compared to manual labeling.
Increasing robustness of handwriting recognition using character N-Gram decoding on large lexica
(2016)
Offline handwriting recognition systems often include a decoding step, that is retrieving the most likely character sequence from the underlying machine learning algorithm. Decoding is sensitive to ranges of weakly predicted characters, caused e.g. by obstructions in the scanned document. We present a new algorithm for robust decoding of handwriting recognizer outputs using character n-grams. Multidimensional hierarchical subsampling artificial neural networks with Long-Short-Term-Memory cells have been successfully applied to offline handwriting recognition. Output activations from such networks, trained with Connectionist Temporal Classification, can be decoded with several different algorithms in order to retrieve the most likely literal string that it represents. We present a new algorithm for decoding the network output while restricting the possible strings to a large lexicon. The index used for this work is an n-gram index with tri-grams used for experimental comparisons. N-grams are extracted from the network output using a backtracking algorithm and each n-gram assigned a mean probability. The decoding result is obtained by intersecting the n-gram hit lists while calculating the total probability for each matched lexicon entry. We conclude with an experimental comparison of different decoding algorithms on a large lexicon.
Offline handwriting recognition systems often use LSTM networks, trained with line- or word-images. Multi-line text makes it necessary to use segmentation to explicitly obtain these images. Skewed, curved, overlapping, incorrectly written text, or noise can lead to errors during segmentation of multi-line text and reduces the overall recognition capacity of the system. Last year has seen the introduction of deep learning methods capable of segmentation-free recognition of whole paragraphs. Our method uses Conditional Random Fields to represent text and align it with the network output to calculate a loss function for training. Experiments are promising and show that the technique is capable of training a LSTM multi-line text recognition system.
Multi-Dimensional Connectionist Classification is amethod for weakly supervised training of Deep Neural Networksfor segmentation-free multi-line offline handwriting recognition.MDCC applies Conditional Random Fields as an alignmentfunction for this task. We discuss the structure and patterns ofhandwritten text that can be used for building a CRF. Since CRFsare cyclic graphical models, we have to resort to approximateinference when calculating the alignment of multi-line text duringtraining, here in the form of Loopy Belief Propagation. This workconcludes with experimental results for transcribing small multi-line samples from the IAM Offline Handwriting DB which showthat MDCC is a competitive methodology.
Recent years have seen the proposal of several different gradient-based optimization methods for training artificial neural networks. Traditional methods include steepest descent with momentum, newer methods are based on per-parameter learning rates and some approximate Newton-step updates. This work contains the result of several experiments comparing different optimization methods. The experiments were targeted at offline handwriting recognition using hierarchical subsampling networks with recurrent LSTM layers. We present an overview of the used optimization methods, the results that were achieved and a discussion of why the methods lead to different results.
Visualization-Assisted Development of Deep Learning Models in Offline Handwriting Recognition
(2018)
Deep learning is a field of machine learning that has been the focus of active research and successful applications in recent years. Offline handwriting recognition is one of the research fields and applications were deep neural networks have shown high accuracy. Deep learning models and their training pipeline show a large amount of hyper-parameters in their data selection, transformation, network topology and training process that are sometimes interdependent. This increases the overall difficulty and time necessary for building and training a model for a specific data set and task at hand. This work proposes a novel visualization-assisted workflow that guides the model developer through the hyper-parameter search in order to identify relevant parameters and modify them in a meaningful way. This decreases the overall time necessary for building and training a model. The contributions of this work are a workflow for hyper-parameter search in offline handwriting recognition and a heat map based visualization technique for deep neural networks in multi-line offline handwriting recognition. This work applies to offline handwriting recognition, but the general workflow can possibly be adapted to other tasks as well.
Digital cameras are subject to physical, electronic and optic effects that result in errors and noise in the image. These effects include for example a temperature dependent dark current, read noise, optical vignetting or different sensitivities of individual pixels. The task of a radiometric calibration is to reduce these errors in the image and thus improve the quality of the overall application. In this work we present an algorithm for radiometric calibration based on Gaussian processes. Gaussian processes are a regression method widely used in machine learning that is particularly useful in our context. Then Gaussian process regression is used to learn a temperature and exposure time dependent mapping from observed gray-scale values to true light intensities for each pixel. Regression models based on the characteristics of single pixels suffer from excessively high runtime and thus are unsuitable for many practical applications. In contrast, a single regression model for an entire image with high spatial resolution leads to a low quality radiometric calibration, which also limits its practical use. The proposed algorithm is predicated on a partitioning of the pixels such that each pixel partition can be represented by one single regression model without quality loss. Partitioning is done by extracting features from the characteristic of each pixel and using them for lexicographic sorting. Splitting the sorted data into partitions with equal size yields the final partitions, each of which is represented by the partition centers. An individual Gaussian process regression and model selection is done for each partition. Calibration is performed by interpolating the gray-scale value of each pixel with the regression model of the respective partition. The experimental comparison of the proposed approach to classical flat field calibration shows a consistently higher reconstruction quality for the same overall number of calibration frames.
Algorithms for calculating the string edit distance are used in e.g. information retrieval and document analysis systems or for evaluation of text recognizers. Text recognition based on CTC-trained LSTM networks includes a decoding step to produce a string, possibly using a language model, and evaluation using the string edit distance. The decoded string can further be used as a query for database search, e.g. in document retrieval. We propose to closely integrate dictionary search with text recognition to train both combined in a continuous fashion. This work shows that LSTM networks are capable of calculating the string edit distance while allowing for an exchangeable dictionary to separate learned algorithm from data. This could be a step towards integrating text recognition and dictionary search in one deep network.
Die Entstehung von Radsatz-Torsionsschwingungen kann aufgrund der zunehmenden Ausnut- zung des Kraftschlusses im Rad-Schiene-Kontakt nicht vollends verhindert werden. Während analytische Untersuchungen zeigen, dass bei der Überschreitung eines bestimmten Dämp- fungswerts die Entstehung von Radsatz-Torsionsschwingungen vollständig vermieden wird, zeigen Simulationen, dass die Schwingungsamplitude des dynamischen Torsionsmoments in- folge des betragsmäβig sinkenden Kraftschlussgradienten und einer gewissen Antriebsstrang- dämpfung bei höheren Gleitgeschwindigkeiten begrenzt ist. Im Gegensatz zur analytischen Formel (17) ist damit ein maximales, dynamisches Torsionsmoment berechenbar.
Durch eine ganzheitliche Betrachtung des Gesamtsystems, bestehend aus dem Kraftschluss zwischen Rad und Schiene, dem Radsatz inkl. Lagerung, sowie Antriebsstrang können ver- schiedene Einflussfaktoren untersucht werden, die eine Radsatz-Torsionsschwingung beein- flussen. Die damit verbundenen Berechnungsmodelle sind aufgrund der Anzahl von Massen und Freiheitsgraden nicht mehr analytisch lösbar. Durch den Einsatz von Mehrkörperdynamik-Software können die Einflussfaktoren identifiziert und deren Einfluss auf die Torsionsneigung des Systems und das dynamische Torsionsmoment quantifiziert werden.
Radsatz-Torsionsschwingungen verursachen Torsionsamplituden, die das quasistatische Torsionsmoment um ein Vielfaches überschreiten können. Dies führt zu enormen Beanspruchungen von Radsatzwelle, Pressverbänden und Radsatzgetrieben. Durch die Auswertung und Analyse zahlreicher Messfahrten verschiedener Fahrzeuge konnten einige Merkmale bezüglich der Verteilung der Torsionsbelastung innerhalb eines Triebdrehgestells identifiziert werden. Da die Messresultate stark von der individuellen Antriebsregelung den vorliegenden Kraftschlussbedingungen abhängig sind lassen sich für Fahrzeuge mit unterschiedlicher Antriebskonfiguration nur bedingt vergleichen. Während im Regelbetrieb von Schienenfahrzeugen das Auftreten von Radsatz- Torsionsschwingungen vermieden werden soll, sind insbesondere bei Zulassungsfahrten Fahrzustände anzustreben in denen bei aktiver Rollier- und Schleuderschutzsoftware maximale Torsionsschwingungen erzeugt werden können. Basierend auf den durchgeführten Analysen und gesammelten Erfahrungen haben sich für die Durchführung von Versuchsfahrten folgende Bedingungen zielführend herausgestellt: (1) Herstellung möglichst identischer Kraftschlussverhältnisse aller im Gruppenantrieb elektrisch gekoppelten Radsätze durch gemeinsame Bewässerung in beiden Fahrrichtungen. (2) Sicherstellung einer maximalen Zugkraftausnutzung durch hohe Fahrwiderstände, geeignete Streckwahl mit konstanter Steigung oder Einsatz einer Bremslokomotive.