For the purpose of refined fault detection in DC grids, linear models of components, cabling and possible faults can be applied in order to model the first milliseconds after a possible sudden event. Using a semi-analytical model, a factorial analysis of signals at voltage and current sensors is set up - specific signal patterns representing either fault events or changes between normal operation modes are estimated. With these results, refined fault detection methods avoiding false tripping can be implemented and parametrized.

This paper presents a novel and comprehensive approach for electrical safety in DC microgrids, where different protective devices are combined to an overall protection system with central monitoring. It also describes the challenges that must be faced when applying typical safety elements to DC.

A specific refinement of fault detection in LVDC-microgrids requires model-based knowledge about the system due to possible variations in topology, component parameters and line lengths. In order to gain this information, passive system identification methods based on current and voltage measurement data are applied. Further using machine learning tools for parameter validation, refined and fast model-based fault identification methods can be set up.

Experimental systems are necessary to validate and improve theoretical molecular communication models. We propose an analytical model for a biological modulator based on genetically modified Escherichia coli (E. coli) bacteria which convert an electrically-controlled light signal into a chemical response. Techniques from system identification and electrical networks synthesis are used to model the complex biological system. Compared to a simple reference model, a better fit to the observed dynamics in the experimental data is obtained. I. PROBLEM DESCRIPTION Molecular communication (MC) is motivated by biomedical applications such as targeted drug delivery. Moreover, the interest in these applications has driven significant progress in theoretical models and spawned first experimental systems [1], [2]. These experimental setups are important to guide the further development of theoretical models, which are necessary for efficient analysis and design of MC systems. In this work, methods from system identification [3] are used to obtain an accurate representation of the measured system responses of the experimental testbed proposed in [4]. The identified system is modeled as a second-order system and an electrical network is used to represent the experimental process. Furthermore, the individual network components are given a biological meaning. First, the first-order reference model from [4] is reviewed, which is not sufficient to capture all signal dynamics. Then, a general framework for the derivation of a reference network from measurement data by system identification and network synthesis is presented. The benefits of the proposed model are shown based on simulations of the proton concentration. II. EXPERIMENTAL TESTBED For MC systems, the dynamics of the chemical signal needs to be fast and easy to control. Fig. 1 shows an experimental interface for MC, based on protons as chemical signal carriers as proposed in [4]. The signal modulator is a bacterial cell expressing Gloeorhodopsin (GR). The membrane protein GR is a light-dependent proton pump. In the natural environment of a cell, a proton is pumped into the periplasmic space by the energy of one photon within 40 μs and the regeneration back to the ground state takes around 15 ms. In the experimental testbed, the bacteria are used without the outer membrane layer. These cells are illuminated by green light driving a proton translocation to the surrounding medium. The increase of the proton concentration can then be measured by a pH electrode. Fig. 1. Biological Modulator model. Benchtop experimental setup [4]. During illumination the proton concentration increases until saturation and turning the light off results in a decrease of the concentration. The kinetics of illumination and regeneration in the dark are dependent on cellular mechanisms. Pumping protons out of the cell is an active and energy driven process. During illumination the pressure of the rising proton gradient drives protons passively inward. During regeneration in the dark, only the passive, and therefore slower influx of protons reaches saturation with the decreasing proton gradient until the natural pH environment of the cell is fully recovered.

Direct current microgrids in the low voltage range require specific system protection. In addition to the basic functionality of conventional protective devices, a refined and model-based analysis of measured current and voltage signals is necessary. Signal processing, system identification and machine learning methods are helpful to identify, classify and localize faults and gradual malfunction. Online condition monitoring allows the realization of predictive maintenance concepts, a thorough analysis of occurring events guards against grid sector breakdown by an appropriate and selective tripping of protective devices.

Low end extra low voltage direct current grids require selective fault protection designed for the specific application and system voltage. System identification and machine learning methods are helpful to identify, to localize and to classify occurring fault events. A category of non-recursive large-signal methods in the time domain for system identification and for refined fault detection and analysis is introduced.

Fault detection in DC grids requires new concepts for the design of protection devices. A promising concept for predictive circuit protection is model based learning, e.g. Kalman estimation. It requires detailed system models of electrical circuits. An established modeling technique in circuit theory is the wave digital principle which preserves circuit passivity and stability. This contribution combines both ideas and derives a Kalman filter in the wave digital domain.

Specific system protection is needed for direct current microgrids in the low voltage range. Model-based analysis of transient events applying system identification and machine learning procedures provides additional knowledge on system topology and components. Using this information, refined model-based fault detection and discrimination methodologies can be set up in addition to the basic functionality of conventional protective devices. Consequential damage and grid sector breakdown can be avoided by fast and selective tripping of suitable switching devices.

Measurements of the input impedance of electrical transmission lines indicate a frequency dependency of the line parameters. This behavior can be simulated by adjustments to a suitable transfer function model for frequency independent parameters. To this end, functional transformations in time and space set up a transfer function model as a decomposition into individual modes. Adjusting the parameters for each single mode to measured input impedances faithfully describes the frequency dependent behavior. The presented approach does not only model the input impedance, it is also suitable for time domain simulations.

Various fault scenarios have been analyzed by running a number of differently combined 380 VDC microgrid tests. These tests represented a common grid topology with low system impedance at grid resonance points within the single- or lower double-digit kHz range. At serial arc faults, self-excited resonant modes of the arc plasma column have been observed. They lead to an increased arc column stability compared to non-resonant arcs with colored noise behavior. These characteristics require a special focus on pattern recognition methods for arc fault sensors along with extended suitability tests for mechanical and hybrid switchgear concerning the altered stability of switching arcs. The use of small-signal models for system components such as source and load converters as well as for arcs with regard to large-signal DC operating points and converter control modes is helpful in order to describe the reaction of the system in the event of a malfunction. This is essential for the development of suitable protective components and algorithms.

A transfer function description is derived for a general class of linear distributed parameter systems dependent on time and one spatial variable. Suitable functional transformations are the Laplace transformation for the time variable and the Sturm- Liouville transformation for the space variable. A practical problem is the determination of the eigenfunctions of the Sturm- Liouville transformation since these depend on the type and the parameters of the boundary conditions. This contribution shows that the design of a transfer function model can be separated from the correct treatment of the boundary conditions. The presented approach exhibits strong parallels to state feedback techniques from control theory. Examples for an electrical transmission line demonstrate how terminations with arbitrary complex impedances can be considered without redesigning the transmission line model.

380 V DC distribution grids have reached maturity for the power supply of data centers and central offices over recent years. New developments in this field are tending towards integrating more distributed energy resources like photovoltaics and wind turbines. Also, high-capacity battery storage systems based on lithium-ion cells are on the rise to increase self-reliance and reduce operating cost. With every grid component being connected to the 380 V DC supply bus via a DC/DC or AC/DC converter, the dynamic system behavior will be entirely dominated by the control loops and operational limits of the power electronic components. In consequence, a re-evaluation of the fault current propagation for various fault types is necessary to dimension safety devices correctly. This paper describes a modelling approach using linearized converter models to analyze the system behavior. The presented models are verified with a laboratory test grid. Finally, guidelines to properly select safety elements and setting up self-protecting mechanisms for power converters in next generation 380V DC distribution grids are outlined.

Modern DC distribution grids and DC systems of renewable energy production require more elaborated safety concepts than AC applications. Due to the stability of DC arcs for system voltages above 20V, a short fault detection time is essential for circuit protection devices.In many applications, the implementation of overcurrent and overvoltage protection is not sufficient. Reliable and fast fault detection, including low power faults and gradual malfunction in an early stage, will increase system safety by preventing further severe damage.Signal processing functions and predictive methods based on self-learning system models can detect fault signatures before dangerous situations occur. Protection devices with implemented model-based machine learning methods are advantageous compared to conventional mechanical or electronic circuit breakers. For this purpose, analytical and numerical models of the system components and the lines have to be used, including models of possible fault scenarios. For an implementation of intelligent real-time methods in local safety devices with restricted computing power and memory, system modeling based on the wave digital principle is beneficial for the setup of accurate digital representations.

DC grids of system voltages above 20V demand different safety concepts compared to conventional AC grids. More elaborate protection devices have to be developed to detect not only high-power, but also low-power faults. The discrimination between faults and load variations can be supported by model-based machine learning methods. For this purpose, this contribution developes time-invariant and linearized system-models with chains of two-ports including possible faults. To consider faults in transmission lines occuring after steady-state system conditions, the initial distribution of voltages and currents is modeled by spatially concentrated equivalent sources. This approach leads to an analytic frequency domain solution without spatial discretization. Measurements on a two-core cable compare favourably with this closed form model.

The analysis of the quantization error in fixed-point arithmetic is usually based on simplifying assumptions. The quantization error is modelled as a random variable which is independent of the quantized variable. This contribution investigates the wordlength reduction of a digital multiplier in greater detail. The power spectrum of the quantization is expressed by the power spectrum of the multiplier input. The analytical results agree with measurements of the quantization error. The presented error model is shown to be superior to the simplified one for wordlengths in the range of eight bit.

Compared to conventional AC grids, DC microgrids demand different switchgear and safety concepts. Electric arcs across mechanical contacts of circuit breakers during switching operations and also arc faults within the installation are more stable in the case of continuous current. Model based methods are important tools not only for selecting the appropriate method of converter control, but also for analyzing the boundary conditions for faults and to develop reliable arc fault detection devices.

Especially in electrical networks with distributed sources and a large variety of possible loads it makes sense to combine models of the components in a modular conception in order to analyze the vulnerability to arc faults and to develop reliable arc fault detection referring to the specific system characteristics. Therefore it is appropriate to analyze the small-signal behavior of the sources, the lines and the loads at a great variety of operating points and to consider the inherent characteristics in the time and frequency domain (e.g. switching frequencies of converters), if normal operation is to be distinguished from operation with arc fault and its typical broadband noise. This model-based approach allows a system-adapted design of pre-processing analogue filters in LF- or VLF-sensors and precise feature selection in pattern recognition algorithms for the purpose of arc fault detection and classification. After a generalized description the model-based methods are specified for the application field of photovoltaics and for lithium-ion-batteries.

While the speed of operation of conventional circuit breakers for equipment (CBE) is suitable for the majority of modern day equipment applications, ultra-sensitive solidstate circuitry typically found in telecom power distribution applications will often require faster switching protection under overcurrent conditions than electromechanical CBEs can achieve.

In photovoltaic systems a large amount of electrical connectors has been used in the combination of serial and parallel structures of the modules. Due to the high DC voltages and the aging of the systems, long-lasting arc faults can occur which may cause serious fires. As an initial step to develop sensor-devices for detecting arc faults in photovoltaic systems, a test set-up consisting of several modules, a solar inverter, and a unit for creating artificial arc faults was installed. The analysis of the measured signals in time and frequency domain showed the following: Parallel arcing involves significant changes in the current at the primary side of the converter and is therefore easily detectable. Series arc faults, however, can usually not be detected by a low frequency analysis of current and voltage signals due to the specific characteristic curve of the photovoltaic modules, the control-concept of the converter for maximum power tracking, and possibly changing solar irradiance. Sensors and methods considering the higher spectral components of the arc effects are necessary.