Refine
Publication Type
- Conference Proceeding (12)
- Article (10)
- Part of a Book (2)
- Report (1)
Keywords
- Analysis (1)
- Battery State Estimation (1)
- Deep Learning (1)
- Development and implementation (1)
- Diamond (1)
- End of charge (1)
- GNSS (1)
- Highly Sensitive (1)
- Hochintegrierte Mikro- und Nanosysteme (1)
- Interval analysis · Particle filtering · Kalman filtering · Bayesian filtering · Sequential Monte Carlo simulation · Bounded-error estimation (1)
Faculty
This paper deals with the issue of automating the
process of machine learning and analyzing bio-datasets. For this
a user-friendly website has been developed for the interaction
with the researchers. On this website it is possible to upload
datasets and to share them, if desired, with other scientists. The
uploaded data can also be analyzed by various methods and
functions. The signals inside these datasets can also be visualized.
Furthermore several algorithms have been implemented to create
machine learning models with the uploaded data. Based on these
generated models new data can be classified or calculated. For all
these applications the simplest possible handling was
implemented to make the website available to all interested
researchers.
State of Charge estimation of zinc air batteries using electrochemical impedance spectroscopy
(2018)
Metal air batteries provide a high energy density as the ca-thodic reaction uses the surrounding air. Different metals can be usedbut zinc is very promising due to its disposability and nontoxic behav-ior. State estimation is quite complicated as the voltage characteristicof the battery is rather flat. Especially estimating the state of chargeis important as a secondary electrolysis process during overcharging canlead to an unsafe state. Another technique for state estimation is theelectrochemical impedance spectroscopy. Therefore, this paper describesthe process of setup and measuring a time series of impedance spectraat known states of charge. Then these spectra are used to derive anequivalent circuit. Finally the development of the circuit’s parameter areanalyzed to extract most important parameters.
The main task of battery management systems is to keep the working area of the battery in a safe state. Estimation of the state of charge and the state of health is therefore essential. The traditional way uses the voltage level of a battery to determine those values. Modern metal air batteries provide a flat voltage characteristic which necessitates new approaches. One promising technique is the electrochemical impedance spectroscopy, which measures the AC resistance for a set of different frequencies. Previous approaches match the measured impedances with a nonlinear equivalent circuit, which needs a lot of time to solve a nonlinear least-squares problem. This paper combines the electrochemical impedance spectroscopy with neural networks to speed up the state estimation using the example of zinc air batteries. Moreover, these networks are trained with different subsets of the spectra as input data in order to determine the required number of frequencies.
Oxygen consumption of zinc-air batteries and theirperformance at low oxygen concentration levels
(2018)
Already existing primary Zinc-air batteries providea high energy density. Due to new secondary cells, its tech-nology can become an alternative for energy storage. Sincethese applications require a big amount of storable energy, theoxygen consumption has to be taken into account. This articledetermines the oxygen consumption of zinc-air batteries duringdischarging. Furthermore the performance of zinc-air batteries atlow oxygen concentrations is analyzed. Both aspects are validatedby practical experiments.
Accurate self-localisation is a fundamental ability of any mobile robot. In Monte Carlo localisation, a probability distribution over a space of possible hypotheses accommodates the inherent uncertainty in the position estimate, whereas bounded-error localisation provides a region that is guaranteed to contain the robot. However, this guarantee is accompanied by a constant probability over the confined region and therefore the information yield may not be sufficient for certain practical applications. Four hybrid localisation algorithms are proposed, combining probabilistic filtering with non-linear bounded-error state estimation based on interval analysis. A forward-backward contractor and the Set Inverter via Interval Analysis are hybridised with a bootstrap filter and an unscented particle filter, respectively. The four algorithms are applied to global localisation of an underwater robot, using simulated distance measurements to distinguishable landmarks. As opposed to previous hybrid methods found in the literature, the bounded-error state estimate is not maintained throughout the whole estimation process. Instead, it is only computed once in the beginning, when solving the wake-up robot problem, and after kidnapping of the robot, which drastically reduces the computational cost when compared to the existing algorithms. It is shown that the novel algorithms can solve the wake-up robot problem as well as the kidnapped robot problem more accurately than the two conventional probabilistic filters.