Journal of Applied Research in Electrical Engineering
Volume:2 Issue: 1, Winter and Spring 2023

The predictive performance of a neural network depends on the one hand on the difficulty of a problem, defined by the number of classes and complexity of the visual domain, and on the other hand on the capacity of the model, determined by the number of parameters and its structure. By applying layer saturation and logistic regression probes, we confirm that these factors influence the inference process in an antagonistic manner. This analysis allows the detection of over- and under-parameterization of convolutional neural networks. We show that the observed effects are independent of previously reported pathological patterns, like the “tail pattern”. In addition, we study the emergence of saturation patterns during training, showing that saturation patterns emerge early in the optimization process. This allows for quick detection of problems and potentially decreased cycle time during experiments. We also demonstrate that the emergence of tail patterns is independent of the capacity of the networks. Finally, we show that information processing within a tail of unproductive layers is different, depending on the topology of the neural network architecture.

The stochastic computing (SC) method is a low-cost alternative to conventional binary computing that processes digital data in the form of pseudo-random bit-streams in which bit-flip errors have a trivial effect on the signal final value because of the highly redundant encoding format of this method. As a result, this computational method is used for fault-tolerant digital applications. In this paper, stochastic computing has been chosen to implement 2-dimensional discrete wavelet transform (2-D DWT) as a case study. The performance of the circuit is analyzed through two different faulty experiments. The results show that stochastic 2-D DWT outperforms binary implementation. Although SC provides inherent fault tolerance, we have proposed four structures based on dual modular redundancy to improve SC reliability. Improving the reliability of the stochastic circuits with the least area overhead is considered the main objective in these structures. The proposed methods are applied to improve the reliability of stochastic wavelet transform circuits. Experimental results show that all proposed structures improve the reliability of stochastic circuits, especially in extremely noisy conditions where fault tolerance of SC is reduced.

Increasing the penetration level of distributed generation (DG) units in radial power distribution systems can increase the short-circuit level in these networks, which can, in turn, have destructive effects such as exceeding the tolerable current of the equipment and disrupting the protective coordination in the network. The active superconducting fault current limiter (ASFCL) is a new device that can limit fault current using voltage series compensation. This paper discusses the modeling of ASFCL and control strategies including fault detection and converter performance in normal and fault modes. Initially, its performance in limiting the fault current is investigated by simulating a sample three-phase system with ASFCL. In the next step, three operating modes including normal mode, upstream fault mode, and downstream fault mode are proposed to achieve an adaptive FCL that solves these problems in grid-connected microgrids. The simulation results confirm the proper performance of the ASFCL modes in both fault current limiting and protective coordination of overcurrent relays in the network.

This paper presents a solution to the Optimal Power Flow (OPF) problem combined economic dispatch with valve-point effect and Emission Index (EI) in electrical power networks using the physics-inspired optimization method, which is the Gravitational Search Algorithm (GSA). Our main goal is to minimize the objective function necessary for the best balance between energy production and its consumption which is presented in a nonlinear function, taking into account equality and inequality constraints. The objective is to minimize the total cost of active generations, the active power losses, and the emission index. The GSA method has been examined and tested on the standard IEEE 30-bus test system with various objective functions. The simulation results of the used methods have been compared and validated with those reported in the recent literature. The results are promising and show the effectiveness and robustness of the used method. It should be mentioned that from the base case, the cost generation, the active power losses, and the emission index are significantly reduced to 823 ($/h), 6.038 (MW), and 0.227 (ton/h), which are considered 5.85%, 61.61%, and 44.63%, respectively.

This paper is an extension of our previous research on presenting a novel Gaussian Mixture-based (MOG2) Video Coding for CCTVs. The aim of this paper is to optimize the MOG2 algorithm used for foreground-background separation in video streaming. In fact, our previous study showed that traditional video encoding with the help of MOG2 has a negative effect on visual quality. Therefore, this study is our main motivation for improving visual quality by combining the previously proposed algorithm and color optimization method to achieve better visual quality. In this regard, we introduce Artificial Intelligence (AI) video encoding using Color Clustering (CC), which is used before the MOG2 process to optimize color and make a less noisy mask. The results of our experiments show that with this method the visual quality is significantly increased, while the latency remains almost the same. Consequently, instead of using morphological transformation which has been used in our past study, CC achieves better results such that PSNR and SSIM values have been shown to rise by approximately 1dB and 1 unit respectively.

The increasing risk of cardiovascular diseases, stress, high blood pressure, obesity, sleep disorders, and depression causes electrocardiogram (ECG) monitors to be used for diagnosing health. The main objective of this research is to enhance the quality of the ECG signal using wavelet transform and adaptive filters. This research has been made as descriptive-analytic and the method is used in the signal processing stages to calculate the ECG modulation spectrum, the spectral-modulation filtering scheme, and the ECG database from the standard algorithm and performance criteria. The results of the simulation indicate that the conversion of Sym4 and the adaptive filter with the size of 0.0005 and the length of the filter of 25 signals to the noise will be greatly improved to reveal the main features of the ECG signal.

In fuel cell systems, voltage balancing is an important consideration. The utilization of a modular construction based on a three-level boost converter was able to balance DC voltage. This paper investigates the effect of parameter variations, such as inductors and capacitors, on the converter's steady-state controllable areas. The plot of the inductor current and the voltages of the output capacitors are illustrated for different scenarios. The system simulation results were performed using MATLAB / Simulink software.

This work represents a new method for robustness analysis of the model reference adaptive controller (MRAC) in the presence of input saturation. Saturation is one of the nonlinear factors affecting the stability of control systems which must be considered in controller design and stability analysis experiments. Various methods are presented for the stability and robustness analysis of adaptive control systems, and employment of describing function (DF) can be attractive and practical, due to the appropriate effectiveness of DF in estimating limit cycles and also the application of quasi-linearization theory. In this work, the stability analysis and a limit cycle estimation of a saturated system in the frequency domain are performed. The controller parameters are adjusted in a way that the system achieves its stable limit cycle in the presence of the initial conditions for the states. Moreover, the efficiency of the proposed method for second-order systems is reported in the presence of symmetric saturation and uncertainty model in Rohrs’s counterexample as the unmodeled dynamics. The results demonstrate the proposed method provides a proper analysis of system stability during the changes in the control parameters and the saturation amplitude.

Nowadays, planar spiral coils are widely used in different applications. Mutual inductance of two adjacent coils, is one of the critical operating principles in near-field wireless power and data transmission systems, significantly impacting their performance. Hence, in this study, the mutual inductance between two similar concentric planar spiral coils is investigated. The effect of main parameters, including the track width, w, and the space between two consecutive turns, s, with a fixed inner and outer diameter of the coils are investigated. The Taguchi method using the L16 array in Minitab environment is used to optimize design parameters. The samples of applied Taguchi, are modeled and simulated via ANSYS Maxwell. The results show that the mutual inductance increases by reducing the two investigated parameters. Based on the Taguchi analysis, it is revealed that the effect of the response for both of the investigated parameters is very close. By applying the main effect analysis the obtained results are verified. This interesting result is important in the design of planar spiral coils while we have fabrication limitations in a real sensor design realization.

Partial Discharge (PD) measurement is one of the best solutions for condition assessment of Gas Insulated Switchgears (GISs). For having Condition-based maintenance of GIS, online PD monitoring is of great importance. For this aim, Ultra High Frequency (UHF) PD sensors should be installed inside the GIS during the installation. However, in most installed GISs in industries, the internal UHF PD sensors are not installed. In this paper, a new method for online defect type recognition according to external UHF PD sensors and based on the time-frequency representation of PD signal is proposed. In this case, four artificial defect types named protrusion on the main conductor, protrusion on the enclosure, free moving metal particle, and metal particle on spacer are implanted inside the 132 kV L-Shaped structure of one phase in enclosure GIS. The signal energy at each level of the decomposed signal by Discrete Wavelet Transform (DWT) is applied for features of each defect type. The trends of signal energy variations at each frequency range of signal are applied for discriminating between each defect type. The Deep Feed Forward Network (DFFN) classifier is applied for PD pattern recognition. The results show the benefits and simplicity of the proposed method for PD signal classification, independent from the position of the PD sensor, especially in the case of online PD monitoring of GIS.

A new active gate drive for Silicon carbide (SiC) metal–oxide–semiconductor field-effect transistor (MOSFET) is proposed in this paper. The SiC MOSFET as an attractive replacement for insulated gate bipolar transistor (IGBT) has been regarded in many high power density converters. The proposed driver is based on a feedforward control method. This simple analog gate driver (GD) improves switching transient with minimum undesirable effect on the efficiency. This paper involves the entire switching condition (turn on/off), and the GD is applied to the SiC base technology of MOSFET. To evaluate the performance of the proposed GD, it will be compared with a conventional gate driver. The presented GD is validated by experimental tests. All the evaluations are carried out in a hard switching condition and at high-frequency operation.

Approximate computing is considered a promising way to design high-performance and low-power arithmetic units recently. This paper proposes an energy-efficient logarithmic multiplier for error-tolerant applications. The proposed multiplier uses a novel technique to calculate the powers of two products to reduce critical path complexity. Also, a correction term is provided to improve the multiplier accuracy. Additionally, the use of approximate adders in our design is investigated, and optimal truncation length is obtained through simulations. We evaluated our work both in accuracy and hardware criteria. Experiments on a 16-bit proposed multiplier with approximate adder show that power-delay product (PDP) is significantly reduced by 34.05% compared to the best logarithmic multipliers available in the literature, while the mean relative error distance (MRED) is also decreased by 21.1%. The results of embedding our multiplier in the dequantization step of the JPEG standard show that the image quality is improved in comparison with other logarithmic multipliers. In addition, a subtle drop in image quality compared to utilizing exact multipliers proves the viability of our design.