Refine
Document Type
- Conference Proceeding (522)
- Article (427)
- Part of a Book (138)
- Book (54)
- Doctoral Thesis (15)
- Anthology (11)
- Patent (6)
- Review (5)
- Working Paper (5)
- Journal (1)
Language
- English (1185) (remove)
Institute
- Informatik (391)
- ESB Business School (259)
- Technik (231)
- Angewandte Chemie (183)
- Textil und Design (120)
Semi-automated image data labelling using AprilTags as a pre-processing step for machine learning
(2019)
Data labelling is a pre-processing step to prepare data for machine learning. There are many ways to collect and prepare this data, but these are usually associated with a greater effort. This paper presents an approach to semi-automated image data labelling using AprilTags. The AprilTags attached to the object, which contain a unique ID, make it possible to link the object surfaces to a particular class. This approach will be implemented and used to label data of a stackable box.
The data is evaluated by training a You Only Look Once (YOLO) net, with a subsequent evaluation of the detection results. These results show that the semi-automatically collected and labelled data can certainly be used for machine learning. However, if concise features of an object surface are covered by the AprilTag, there is a risk that the concerned class will not be recognized. It can be assumed that the labelled data can not only be used for YOLO, but also for other machine learning approaches.
The critical process parameters cell density and viability during mammalian cell cultivation are assessed by UV/VIS spectroscopy in combination with multivariate data analytical methods. This direct optical detection technique uses a commercial optical probe to acquire spectra in a label-free way without signal enhancement. For the cultivation, an inverse cultivation protocol is applied, which simulates the exponential growth phase by exponentially replacing cells and metabolites of a growing Chinese hamster ovary cell batch with fresh medium. For the simulation of the death phase, a batch of growing cells is progressively replaced by a batch with completely starved cells. Thus, the most important parts of an industrial batch cultivation are easily imitated. The cell viability was determined by the well-established method partial least squares regression (PLS). To further improve process knowledge, the viability has been determined from the spectra based on a multivariate curve resolution (MCR) model. With this approach, the progress of the cultivations can be continuously monitored solely based on an UV/VIS sensor. Thus, the monitoring of critical process parameters is possible inline within a mammalian cell cultivation process, especially the viable cell density. In addition, the beginning of cell death can be detected by this method which allows us to determine the cell viability with acceptable error. The combination of inline UV/VIS spectroscopy with multivariate curve resolution generates additional process knowledge complementary to PLS and is considered a suitable process analytical tool for monitoring industrial cultivation processes.
Automatic classification of rotating machinery defects using Machine Learning (ML) algorithms
(2020)
Electric machines and motors have been the subject of enormous development. New concepts in design and control allow expanding their applications in different fields. The vast amount of data have been collected almost in any domain of interest. They can be static; that is to say, they represent real-world processes at a fixed point of time. Vibration analysis and vibration monitoring, including how to detect and monitor anomalies in vibration data are widely used techniques for predictive maintenance in high-speed rotating machines. However, accurately identifying the presence of a bearing fault can be challenging in practice, especially when the failure is still at its incipient stage, and the signal-to-noise ratio of the monitored signal is small. The main objective of this work is to design a system that will analyze the vibration signals of a rotating machine, based on recorded data from sensors, in the time/frequency domain. As a consequence of such substantial interest, there has been a dramatic increase of interest in applying Machine Learning (ML) algorithms to this task. An ML system will be used to classify and detect abnormal behavior and recognize the different levels of machine operation modes. The proposed solution can be deployed as predictive maintenance for Industry 4.0.
Power line communications (PLC) reuse the existing power-grid infrastructure for the transmission of data signals. As power line the communication technology does not require a dedicated network setup, it can be used to connect a multitude of sensors and Internet of Things (IoT) devices. Those IoT devices could be deployed in homes, streets, or industrial environments for sensing and to control related applications. The key challenge faced by future IoT-oriented narrowband PLC networks is to provide a high quality of service (QoS). In fact, the power line channel has been traditionally considered too hostile. Combined with the fact that spectrum is a scarce resource and interference from other users, this requirement calls for means to increase spectral efficiency radically and to improve link reliability. However, the research activities carried out in the last decade have shown that it is a suitable technology for a large number of applications. Motivated by the relevant impact of PLC on IoT, this paper proposed a cooperative spectrum allocation in IoT-oriented narrowband PLC networks using an iterative water-filling algorithm.
Our paper gives first answers on a fundamental question: how can the design of architectures of intelligent digital systems and services be accomplished methodologically? Intelligent systems and services are the goals of many current digitalization efforts today and part of massive digital transformation efforts based on digital technologies. Digital systems and services are the foundation of digital platforms and ecosystems. Digtalization disrupts existing businesses, technologies, and economies and promotes the architecture of open environments. This has a strong impact on new value-added opportunities and the development of intelligent digital systems and services. Digital technologies such as artificial intelligence, the Internet of Things, services computing, cloud computing, big data with analytics, mobile systems, and social enterprise networks systems are important enablers of digitalization. The current publication presents our research on the architecture of intelligent digital ecosystems and products and services influenced by the service-dominant logic. We present original methodological extensions and a new reference model for digital architectures with an integral service and value perspective to model intelligent systems and services that effectively align digital strategies and architectures with artificial intelligence as main elements to support intelligent digitalization.
his book highlights new trends and challenges in intelligent systems, which play an important part in the digital transformation of many areas of science and practice. It includes papers offering a deeper understanding of the human-centred perspective on artificial intelligence, of intelligent value co-creation, ethics, value-oriented digital models, transparency, and intelligent digital architectures and engineering to support digital services and intelligent systems, the transformation of structures in digital businesses and intelligent systems based on human practices, as well as the study of interaction and the co-adaptation of humans and systems. All papers were originally presented at the International KES Conference on Human Centred Intelligent Systems 2020 (KES HCIS 2020), held on June 17–19, 2020, in Split, Croatia.
Heat pumps in combination with a photovoltaic system are a very promising option for the transformation of the energy system. By using such a system for coupling the electricity and heat sectors, buildings can be heated sustainably and with low greenhouse gas emissions. This paper reveals a method for dimensioning a suitable system of heat pump and photovoltaics (PV) for residential buildings in order to achieve a high level of (photovoltaic) PV self-consumption. This is accomplished by utilizing a thermal energy storage (TES) for shifting the operation of the heat pump to times of high PV power production by an intelligent control algorithm, which yields a high portion of PV power directly utilized by the heat pump. In order to cover the existing set of building infrastructure, 4 reference buildings with different years of construction are introduced for both single- and multi-family residential buildings. By this means, older buildings with radiator heating as well as new buildings with floor heating systems are included. The simulations for evaluating the performance of a heat pump/PV system controlled by the novel algorithm for each type of building were carried out in MATLAB-Simulink® 2017a. The results show that 25.3% up to 41.0% of the buildings’ electricity consumption including the heat pump can be covered directly from the PV installation per year. Evidently, the characteristics of the heating system significantly influence the results: new buildings with floor heating and low supply temperatures yield a higher level of PV self-consumption due to a higher efficiency of the heat pump compared to buildings with radiator heating and higher supply temperatures. In addition, the effect of adding a battery to the system was studied for two building types. It will be shown that the degree of PV self-consumption increases in case a battery is present. However, due to the high investment costs of batteries, they do not pay off within a reasonable period.
Rising consumption due to a growing world population and increasing prosperity, combined with a linear economic system have led to a sharp increase in garbage collection, general pollution of the environment and the threat of resource scarcity. At the same time, the perception of environmental protection becomes more sensitive as the consequences of neglecting sustainable business and eco-efficiency become more visible. The Circular Economy (CE) could reduce waste production and is able to decouple economic growth from resource consumption, but most of the products currently in use are not designed for the reuse-forms of the CE. In addition, the decision-making process of the End of-Usage (EoU) products regarding the following steps has further weaknesses in terms of economic attractiveness for the participants, which leads to low return rates and thus the disposal is often the only alternative.
This paper proposes a model of the decision-making process, which uses machine learning. For this purpose, a Machine Learning (ML) classification is created, by applying the waterfall model. An artificial neural network (ANN) uses information about the model, use phase and the obvious symptoms of the product to predict the condition of individual components. The resulting information can be used in a downstream economic and ecological evaluation to assess the possible next steps. To test this process comprehensive training data is simulated to train the ANN. The decentralized implementation, cost savings and the possibility of an incentive system for the return of an end-of-usage product could lead to increased return rates. Since electronic devices in particular are attractive for the CE, laptops are the reference object of this work. However, the obtained findings are easily applicable to other electronic devices.
Estimating molar solubility from the Hildebrand-Scott relation employing Hansen solubility parameters (HSP) is widely presumed a valid semi quantitative approach. To test this presumption and to determine quantitatively the inherent accuracy of such a solubility prognosis, l-ascorbic acid (LAA) was treated as an example of a commercially important solute. Analytical calculus and Monte Carlo (MC) simulation were performed for 20 common solvents with total HSP ranging from 14.5 to 33.0 (MPa)0.5 utilizing validated material data. It was found that, due to the uncertainty of the material data used in the calculations, the solubility prediction had a large scattering and, thus, a low precision.
Relationship between a high-performance work system and employee outcomes: a multilevel analysis
(2020)
Although research on high-performance work systems (HPWS) is increasing, there are few studies in which the focus is on whether and how firm-level HPWS affect individual-level employee outcomes. Using social identity theory, we examined the relationship between HPWS and employee outcomes, and the role organizational identification plays as a mediator in this relationship. We used a multilevel research design and collected data at the organizational and individual levels from a sample of 485 employees of 32 companies in Guangdong Province, China. We used Amos 17.0 and hierarchical linear modeling 6.08 software to examine our hypotheses and the theoretical model. Results showed that organizational identification fully mediated the relationship between HPWS and employees’ job performance as well as that between HPWS and their turnover intention. Our findings provide new insights into the relationship between firm-level human resource management and individual-level employee outcomes, and highlight the importance of considering the implementation of HPWS practices to strengthen employees’ identification with the organization and improve their performance.
