Nonlinear modeling of magnetic materials for circuit simulations.

Magnetic materials in the form of magnetic rings are widely used in power engineering products. In many cases, they operate in high frequency and in nonlinear conditions, e.g., as damping elements in electrical power substations equipped with Gas-Insulated Switchgear (GIS) where they provide suppression of Very Fast Transient Overvoltages (VFTOs). To model phenomena in GIS with magnetic rings it is required to have realistic time-dependent models of magnetic materials operating in a wide frequency range and nonlinear conditions. Nowadays, this has become even more relevant due to the actual trend in the industry to create digital twins of physical devices. Models composed of high-precise discrete lumped nonlinear elements are in demand in circuit simulators like SPICE. This work introduces a method based on classical algorithms that find equivalent lumped models of magnetic cores based on frequency-dependent measurements of impedance under DC-bias current. The model is specifically designed to have smooth behavior in the current domain and thanks to that to improve numerical stability in the time domain simulations.

Measuring impedance frequency characteristics of magnetic rings with DC-bias current.

This paper reports on a comprehensive study on a measurement method for the impedance characteristics in the frequency domain of magnetic rings as a function of DC-bias current up to saturation. The reported measurements in a one-wire coil arrangement with DC-bias current reproduce the operating conditions of rings used in high-power equipment and are useful for the development of circuit models used in power system simulators. The method and its key features are shown on an example of a nanocrystalline ring with large physical dimensions relevant for heavy power equipment. Low magnetic permeability of the ring material and large size of the ring were chosen as the most challenging case with respect to measurement system arrangement, allowing us to highlight the key aspects of the method. The measurement results are reported for frequency ranging from 50 Hz to 30 MHz and for the DC-bias current up to 800 A. The problems and difficulties encountered are highlighted and discussed how they were controlled and overcome.

Self-normalized density map (SNDM) for counting microbiological objects.

The statistical properties of the density map (DM) approach to counting microbiological objects on images are studied in detail. The DM is given by U[Formula: see text]-Net. Two statistical methods for deep neural networks are utilized: the bootstrap and the Monte Carlo (MC) dropout. The detailed analysis of the uncertainties for the DM predictions leads to a deeper understanding of the DM model's deficiencies. Based on our investigation, we propose a self-normalization module in the network. The improved network model, called Self-Normalized Density Map (SNDM), can correct its output density map by itself to accurately predict the total number of objects in the image. The SNDM architecture outperforms the original model. Moreover, both statistical frameworks-bootstrap and MC dropout-have consistent statistical results for SNDM, which were not observed in the original model. The SNDM efficiency is comparable with the detector-base models, such as Faster and Cascade R-CNN detectors.

Generation of microbial colonies dataset with deep learning style transfer.

We introduce an effective strategy to generate an annotated synthetic dataset of microbiological images of Petri dishes that can be used to train deep learning models in a fully supervised fashion. The developed generator employs traditional computer vision algorithms together with a neural style transfer method for data augmentation. We show that the method is able to synthesize a dataset of realistic looking images that can be used to train a neural network model capable of localising, segmenting, and classifying five different microbial species. Our method requires significantly fewer resources to obtain a useful dataset than collecting and labeling a whole large set of real images with annotations. We show that starting with only 100 real images, we can generate data to train a detector that achieves comparable results (detection mAP [Formula: see text], and counting MAE [Formula: see text]) to the same detector but trained on a real, several dozen times bigger dataset (mAP [Formula: see text], MAE [Formula: see text]), containing over 7 k images. We prove the usefulness of the method in microbe detection and segmentation, but we expect that it is general and flexible and can also be applicable in other domains of science and industry to detect various objects.