GAN-based Data Augmentation with Vehicle Color Change for Training Vehicle Detection CNN

AYUB AROONA 충북대학교 2023 국내석사

최근 객체 감지를 위한 CNN(Convolutional Neural Networks)이 크게 개선되었지만 정확도를 인간 수준으로 더욱 향상시키기 위해서는 많은 양의 주석이 달린 데이터가 필요합니다. 이러한 주석이 달린 데이터는 부족합니다. 개체 레이블의 수동 주석은 시간과 비용이 많이 드는 프로세스입니다. 연구자들은 훈련 데이터의 양을 늘리기 위해 전통적인 데이터 증대 기술을 사용합니다. 최근에는 다양한 학습 데이터를 생성하는 데이터를 보강하기 위해 생성 모델이 사용되고 있습니다. 이는 대상 모델의 정확도를 높입니다. 본 논문은 GAN 네트워크를 훈련시키고 생성된 이미지의 품질에 최소한의 손상을 주면서 관심 대상 도메인의 증강 데이터를 생성하는 방법을 제시합니다. 다양한 색상의 차량 이미지로 GAN을 훈련하는 방법을 시연합니다. 그런 다음 주어진 차량 데이터 세트의 차량 색상을 지정된 색상 세트로 변경할 수 있습니다. 실험 결과 제안한 방법이 원본 훈련 데이터가 제한된 응용 프로그램에서 CNN의 탐지 성능을 향상시키는 데 도움이 되는 것으로 나타났습니다. While recent Convolutional Neural Networks (CNNs) for object detection have been substantially improved, they require a large amount of annotated data to further improve their accuracy to the level of human. Such annotated data is scarce. Manual annotation of object labels is a time consuming and expensive process. Researchers use traditional data augmentation techniques to increase the amount of training data. Recently, generative models are being employed to augment data which produces diverse training data. This leads to an increase in the accuracy of the target model. This paper presents a method to train a GAN network and generate augmented data of target domain of interest with the least compromise in the quality of generated images. We demonstrate a method to train a GAN with images of vehicles in different colors. Then it can change the color of vehicles of any given vehicle dataset to a set of specified colors. The experimental results showed that the proposed method helped enhance the detection performance of CNN for applications where original training data is limited.

Study of edge localized modes using mm-wave interferometer array and electron cyclotron emission imaging diagnostics

AYUB, MUHAMMAD KHAWAR POSTECH 2017 국내박사

The study of edge localized modes (ELMs) is one of the most important and interesting subject in the field of fusion plasma research. As the ELM bursts involve rapid loss of considerable particles and heat flux from the edge of plasma, control and physical understanding of ELM dynamics are essential for the steady-state and safe operation of future magnetic fusion devices such as international thermonuclear experimental reactor (ITER). The recently visualized ELM dynamics are found to be very complicated due to its non-linear and non-axis symmetric nature. It is speculated that the linear theory based on peeling-ballooning (PB) model may not be applicable and non-ideal effects such as damping, diffusion and nonlinear saturation come into play during the evolution of an ELM. Therefore, detailed study of the non-ideal effects and accurate measurement of ELM properties such as radial size and perturbation amplitude are essential for the complete understanding of underlying physics of ELM dynamics. The electron cyclotron emission imaging (ECEI) system on the KSTAR, an advanced 2D imaging diagnostics based on the principle of conventional ECE radiometry can visualize the ELM dynamics, using large aperture optics and microwave diagnostic technique, but it has certain limitations. In particular, perturbations with very small radial size (< 2 cm) are difficult to resolve using the ECEI system due to the fundamental limitation in the radial resolution (~ 2 cm at the plasma edge of a typical plasma discharge with toroidal magnetic field ~ 2 T) imposed by the relativistic broadening of ECE. However, at the edge of tokamak plasmas the exact interpretation of ECE signals has intrinsic complexity due to rapid change in the optical thickness from thick to thin. For the optically marginal plasma region (which separates optically thick and optically thin plasma while the optical thickness ~ < 1), e.g., the steep gradient region where the ELMs usually appear during a typical H-mode plasma, ECE signal not only involves the electron temperature fluctuations but also has a significant contribution from electron density fluctuations making difficult the exact estimation of the amplitude. Therefore, the accurate measurement of the radial size and density fluctuation amplitude of the ELMs as well as the interpretation of ECE signals from the optically marginal region of plasma at the edge is very important and need especial attention. A novel technique to estimate the range of radial size and density fluctuation amplitude of ELMs in the KSTAR tokamak plasma is presented. A microwave imaging reflectometry (MIR) system is used as a multi-channel mm-wave interferometer array (MIA) to measure the density fluctuations associated with ELMs, while ECEI system is used as a reference diagnostics to confirm the MIA observation. Two dimensional full-wave (FWR2D) simulations integrated with optics simulation are performed to investigate the Gaussian beam propagation and reflection through the plasma as well as the MIA optical components and obtain the interferometric phase undulations of individual channels at the detector plane due to ELM perturbation. The simulation results show that the amplitude of the phase undulation depends linearly on both radial size and density perturbation amplitude of ELM. For a typical discharge with ELMs, it is estimated that the ELM structure observed by the MIR system has density perturbation amplitude in the range ~ 7 % to 14 % while radial size in the range ~ 1 to 3 cm. The interpretation of ECE signal from the optically marginal region of plasma is also presented. For this purpose, the density and temperature fluctuations associated with the ELMs observed in the KSTAR tokamak are estimated by assuming the ELM filamentary structure as a flux tube bulged out like ballooning mode instability. Synthetic ECE signals from the rotating ELM are calculated based on the measured electron temperature profile and an assumed electron density profile constrained by the measured line averaged density, yielding ~ 0.02 relative fluctuation level in agreement with the experimental observations. The measured ECE signal is nearly in phase with the density modulation associated with the rotating ELM. This implies that the ECE signal corresponding to the ELM filaments has a significant contribution from the density fluctuations. This study can be helpful for the complete understanding of underlying ELMs physics.

Single inverter based novel stator winding scheme for brushless operation of wound-field synchronous machine with improved starting torque

Ayub, Muhammad 한양대학교 대학원 2020 국내박사

Efficient, torque-dense, and cost-effective electrical machines are preferred to be used in most of the applications of electrical-mechanical energy conversion processes. Permanent magnet synchronous machine (PMSM) is one of these machines meant to be highly efficient and with high torque density. However, on the factor of cost-effectiveness it has significantly changed over the past few decades. Price fluctuation in the high energy magnetic material used for manufacturing the permanent magnet has let to alternatives such as wound-field synchronous machine (WFSM) and PM-assisted synchronous reluctance machine (PMa-SynRM). Both WFSM and PMa-SynRM are cost-effective with comparable performance to a PMSM. However, WFSM is generally preferred due to its permanent magnet (PM) less rotor structure which uses field winding for field flux control through direct current (DC) source through slip rings/brushes. WFSMs have more degrees of freedom for machine control (Id, Iq, and If), no PM losses, improved safety through direct field control during inverter fault conditions, and no de-magnetization problem. However, conventional WFSMs have an inherent problem associated with the rotor brushes and slip rings. In the long run, the brushes usually wear out from continuous operation of the machine, creating a maintenance problem. In this regard, several brushless excitation topologies have been investigated which are proposed in the literature. Brushless operation can be carried out by means of a direct wireless power transfer to the field winding of the machine, by additional exciter mounted on the machine system or by transferring harmonic power from the stator of the machine. In this thesis, a novel stator winding scheme is proposed to generate an additional sub-harmonic magneto-motive force (SH-MMF) component for brushless operation of a WFSM. The existing sub-harmonic generation schemes for brushless operation use a dual inverter, or a single inverter with asymmetrical winding and low slot fill factor in half of the stator rendering the machine with low power density. To counter this problem, the proposed scheme is designed to work with single inverter with symmetrical stator winding distribution and the same fill factor for all stator slots. The generated additional SH-MMF component in the proposed scheme is induced in the harmonic winding wound on the machine rotor. A rotating bridge rectifier mounted on the rotor periphery connects the rotor harmonic winding with the rotor field winding, and rectified DC is supplied to the rotor field winding. With the proposed topology, different machine structures were used for variable speed analysis with and without PM-assisted rotor structure. To reduce the usage of PM volume, consequent-pole rotors of different PM shapes were investigated. Additionally, a fault-tolerant scheme was also analyzed to investigate the working of the machine in case of total loss of excitation in the brushless operation of the machine. 2-D finite-element analysis (FEA) was performed on all the aforementioned machine structures to analyze the proposed idea. The FEA results were validated by experiments based on a 1-kW prototype. The results were also compared with the conventional BL-WFSM topologies. It is obtained that the proposed machine topology exhibits greater efficiency, improved starting torque, high average torque, and improved torque density compared to the previous sub-harmonic excitation schemes available in literature.

Source Characterization and Vulnerability Assessment of Nitrate Contamination in Groundwater

Ayub, Riyana North Carolina State University ProQuest Dissertat 2022 해외박사(DDOD)

AYUB, RIYANA. Source Characterization and Vulnerability Assessment of Nitrate Contamination in Groundwater. (Under the supervision of Dr. G. Mahinthakumar)Nitrate contamination in groundwater and surface water from agricultural areas has been a major concern in the United States and around the world for several decades. This study investigates various aspects of nitrate contamination in groundwater and surface water from agricultural practices. This dissertation consists of three major parts. First, non-point source characterization of nitrate contamination into groundwater resources under geological uncertainty was carried out. A Bayesian framework using the Markov Chain Monte Carlo approach (MCMC) was developed to estimate the posterior distribution of non-point sources where uncertainties are due to geological formation (i.e., heterogeneous hydraulic conductivity field). This modeling approach was demonstrated for a hypothetical test case based on the surficial aquifer system in Duplin County, North Carolina. Second, model-based regression approaches for source apportionment were developed using the concept of influence zones to account for transport pathways. The methods are demonstrated using pseudo observations of point-level groundwater nitrate concentrations extracted from a previously published Bayesian Maximum Entropy (BME) model for Sampson and Duplin Counties, North Carolina (NC). Vulnerability assessment of groundwater resources due to nitrate leaching from potential sources of contamination is examined through sensitivity-based scenario analysis which is meaningful for planning and implementation of appropriate mitigation measures. Finally, to account for transport pathways in the integrated groundwater- surface water system, a fully integrated model using SWAT and MODFLOW is developed to simulate the nitrate transport in aquifer using MODFLOW’s reactive transport model, RT3D. This integrated model was used to predict nitrate contamination in groundwater and surface water systems under different climatic and anthropogenic stresses. Climate related analyses were conducted for future periods till the end of this century. A Global Climate Model (GCM), CCSM4, with RCP4.5 emission scenario and downscaled by Multivariate Adaptive Constructed Analogs (MACA) was used for this purpose. From the climate model, an increase of 2.5 to 3 ̊C in annual average temperature is projected; whereas a minimal increase in total precipitation is anticipated over the time. For the anthropogenic stresses, point and non-point source loading from animal farming operations are considered. Model simulations indicate that due to the combined climatic and anthropogenic effects, the basin will experience a decrease of 27% in annual average groundwater nitrate concentration and an increase of 4.5% in-stream nitrate loading for the future period. These projections may aid in planning future agricultural activities under changing climate and increased food demands.

전기적 배터리 모델과 확장된 칼만 필터를 이용한 SOC 예측

Novie Ayub Windarko 충북대학교 2011 국내박사

For several decades, energy storages have been entering into wide area of applications such as mobile electronic appliances, transportation, renewable energy, etc. The type of energy storage for each of those applications may be different and it can be selected with considering the mobility of application, the amount of saved energy, life cycle, the level of response speed, etc. Among those energy storage systems, batteries have advantages of fast response speed, high ramp rates, easily sited, modular, and good energy efficiency. The operation of battery is supported by the Battery Management Storage (BMS). BMS has several tasks to support the battery operation. For example, it can prevent battery from entering into overcharging or overdischarging condition and give the information of residual energy to users. Mainly, BMS needs the State-of-Charge (SOC) estimator to do the tasks. Several researchers have proposed several battery models to support the work of SOC estimator. They are including electrochemical model, mathematical model and electrical model. Furthermore, the electrical engineers can understand easily and can learn intuitively the electrical model. Therefore, this thesis focuses on an electrical model of battery. The electrical model in this thesis includes the resistive and the RC battery models. The resistive model includes OCV and internal resistance modeling. To obtain parameters for the resistive model, the battery should be tested in continuous operation. Note that several papers have reported that OCV may have hysteresis characteristics. Therefore, Takacs model is used to represent the hysteresis characteristics of OCV. To verify the effectiveness of the Takacs model, the battery used in the tests is NiMH types which is well known has strong hysteresis characteristics. Furthermore, the test results show that internal resistance has non-linear characteristics. The results of resistive model can be applied for SOC estimation in continuous operation of battery by simple look-up table. The RC battery model is proposed to represent transient response of battery. This model is useful for intermittent operation of battery which battery operates dynamically during charging, discharging and rest alternately. Furthermore, the RC model is adopted for SOC estimation using extended Kalman Filter (EKF). EKF works to estimate OCV then converts it to SOC. However, EKF cannot estimate well. To avoid this condition, single varying parameter for EKF is proposed. Therefore, the parameter of bulk capacitance is used as the dominant varying parameter among the other parameters to follow SOC value. By using varying parameter, the error of estimation can be reduced. EKF is also tested over the entire SOC region and various current magnitudes. Finally, the results show that the error of estimation is maximum 5%SOC.

Simulation Model of the Automatic Planting Depth Control System for Rice Transplanter Using Proportional Valve

Md Abu Ayub Siddique 충남대학교 대학원 2019 국내석사

이앙기는 벼 재배에 필수적이다. 한국은 벼 이식 작업이 거의 99.9% 기계화가 되어있다. 최근 몇 년간, 전세계적으로 이앙기 기술은 진보되었다. 그러나 국내의 이앙기의 수출은 전년도 대비 감소하였다. 이에, 이앙기에 비례 제어 밸브를 적용하여 세계 시장에서 경쟁력을 갖춰야한다. 따라서, 본 연구의 목적은 비례 제어 밸브를 이용한 이앙기의 자동 식부 깊이 제어 시스템의 시뮬레이션 모델을 개발하는 것이다. 본 연구에서, PID 제어기는 이앙기의 유압 시스템을 제어하기 위해 연구되었다. 두가지의 PID 제어 알고리즘을 개발하였으며 Ziegler-Nichols (Z-N) 방법을 사용하여 PID 계수를 결정했다. 시뮬레이션은 ISO 표준을 만족하는 유압 오일 3가지에 대해 수행하였다. 두개의 컨트롤러의 검증 테스트는 동일한 유압 오일에 대한 비례 제어 밸브 테스트 벤치를 수행하였다. 작동유의 점도는 기준 온도에서 계산되었다. K_p: 5.24, K_i:4.39, K_d:1.34, K_p= 200, T_i=2.52, T_d=0.63으로 감쇠비 0.8에서 압력 제어가 가능하다. 위치 제어기의 시뮬레이션 결과는 비례 제어 밸브의 압력이 작동유 점도의 감소율에 반비례로 나타났다. 이는 위치 컨트롤러가 점도 영향을 제어 할 수 없음을 나타냈다. 압력 제어기의 경우 1A 및 2A 공급 전류에 대한 비례 제어 밸브의 최대 압력은 시뮬레이션 및 실험 모두 동일하며, 각각 약 15.40 bar 및 17.87 bar로 계산되었다. 시뮬레이션과 실험 모두에 대한 안정 시간과 정상 상태 오차의 비교도 동일하게 나타났다. 따라서, 정착 시간과 정상 상태 오차는 0.43 초 및 0 % 였다. 실험 결과는 시뮬레이션 결과보다 높았다. 그러나 실험 결과의 오버 슈트는 경계 조건에 만족하여 약 21.80 %를 차지하는데 이는 25 % 미만이다. 결론적으로 검증 테스트에서는 압력 제어기가 유압 시스템의 유압 오일의 점성 영향을 보상 할 수 있다고 판단된다. 이 컨트롤러는 유압 액츄에이터의 움직임을 제어 할 수 있다. 액츄에이터의 움직임은 비례 제어 밸브에서 공급 된 압력에 의존한다. 따라서 모종이 기울어져 이앙기의 식부의 정밀도를 향상시키는 이앙기의 자동 식부 깊이 제어 시스템에 있어서 압력 제어기가 가능할 수 있다고 할 수 있다. 궁극적으로, 이 컨트롤러는 이앙기의 종합 성능을 향상시킬 수 있을 것으로 판단된다. Rice transplanter is an inevitable technology for rice farming. In Korea, rice transplanter is almost 99.9% mechanized. In recent year, rice transplanter is spreading out overseas with sophisticated facilities. However, the export of domestic rice transplanter has decreased compared to previous years due to the backdated technologies. This is why the mechanized rice transplanter should modify with advanced technology like a proportional valve to hold the global share market of the rice transplanter. Therefore, the goal of this study is to develop the simulation model of the automatic planting depth control system of a rice transplanter using proportional valve. In this study, the PID control law was also studied to control the hydraulic system of rice transplanter. Actually, two PID control algorithm was developed: i) to control the displacement of actuator ii) to compensate for the viscous effect on the hydraulic system of a rice transplanter. Ziegler-Nichols (Z-N) methods were used to determine the PID coefficients. The simulation was conducted for three ISO standard hydraulic oils. Validation test for both control laws was conducted by performing proportional valve test bench for the same hydraulic oils. The viscosity of the hydraulic oils was calculated at its reference temperature. The lowest overshoot and settling time were found at K_p: 5.24, K_i:4.39, and K_d:1.34 for the position control without considering viscosity, and K_p= 200 T_i=2.52 and T_d=0.63 for position control considering viscosity at the damping ratio 0.8. The simulation results of the position control without considering viscosity show that the pressure of the proportional valve is proportional to the decreasing rate of hydraulic oils viscosity. This result indicates that the position control without considering viscosity cannot able to control the viscosity effects. In case of position control considering viscosity, results were found that the maximum pressures of proportional valve for 1 A and 2A supplied current was the same for both simulation and experiment, calculated around 15.40 bar and 17.87 bar, respectively. The comparison of the settling time and steady-state error for both simulation and experiment were also the same. The settling time and steady-state error were 0.43 s and 0%, accordingly. Only, the overshoot of the experimental results was found higher than that of the simulation results. However, the overshoot of the experimental results was satisfied with the boundary condition, accounting for approximately 21.80% which is less than 25%. In conclusion, the validation test stated that the position control considering viscosity could be able to compensate for the viscous effects of hydraulic oils on the hydraulic system. This control law might also be able to control the movement of the hydraulic actuator. Because the movement of the actuator depends on the pressure supplied from the proportional valve. Therefore, it could be said that the position control considering viscosity could be feasible for the automatic planting depth control system of a rice transplanter that improve the rice transplanter planting accuracy with tilting of the planted seedlings. Ultimately, this control law could also be able to increase the comprehensive performance of the rice transplanter.

Synthesis of 2-Dimensional Ti3C2Tx MXene modified Sn/SnOx nanoparticlesfor Strontium Removal from wastewater : 수열 합성법에 기반하여 폐수 내 스트론튬 제거를 위한 Sn/SnOx 나노입자로 개질된 2차원 Ti3C2Tx Mxene의 합성

Ayub, Ammara 경북대학교 대학원 2023 국내석사

새로운 Sn/SnOx@Ti3C2Tx 흡착제는 정전기적 인력을 사용하여 액상 환원에 의한 Ti3C2Tx 나노시트 층에 Sn/SnOx 입자 생성을 통해 제조되었다. 특성 분석 결과는 Sn/SnOx 입자가 Ti3C2Tx 표면에 균일하게 고정되었고 샌드위치형 구조의 나노복합체가 형성되었음을 보여주었다. Sn/SnOx 입자의 증착은 Ti3C2Tx 나노시트 사이의 공간을 증가시켜 Sr2+-이온 제거를 위한 활성부위를 증가시켰다. 또한, 내부 입자는 Ti3C2Tx 매트릭스의 구조적 안정성을 강화하는 기둥 역할을 했으며 Sr2+ 이온이 제거되는 동안 나노층의 restacking 이 발생하는 것을 억제하였다. 합성 이후, Sn/SnOx@Ti3C2Tx의 흡착 성능을 자세히 평가하였다. 실험결과를 통해, 94.36%의 Sr2+ 이온이 90분 이내에 제거되었고 넓은 pH 범위(2-10)에서 pseudo-second 동역학적 모델을 따랐음을 확인 할 수 있었다. 또한, Langmuir 등온선은 흡착 데이터에 가장 적합했으며 Sr2+ 이온에 대한 Sn/SnOx@Ti3C2Tx의 최대 흡착능은 298K에서 71.04 mg/g이었다. 따라서, Sn/SnOx@Ti3C2Tx 나노복합체는 수중에 있는 Sr2+ 이온을 효과적으로 제거할 수 있는 흡착제임을 확인하였다.

Optimization and Modifications of Thermodynamic Models for the Biomass Gasification Process to Predict the Syngas Composition

Hafiz Muhammad Uzair Ayub 동국대학교 2021 국내박사

Global energy demand is mostly fulfilled by the burning of fossil fuels and their derivatives like coal, oil, and natural gas. These conventional energy sources have dominated all the other sources such as wind, solar, and biomass energy. Due to rapid industrialization and use of fossil fuels have hosted several environmental issues and pollutants like greenhouse gas emissions. Modern research paved that sustainable energy sources like wind, solar, and biomass have the potential to help meet the environmental and energy demand problems of the world. Renewable biomass conversion is being utilized as the most promising clean energy source. Among all biomass conversion technologies like combustion, pyrolysis, and gasification, etc., biomass gasification is the most reliable thermochemical conversion technology which converts the biomass into gaseous fuel like H2, CO, CH4. This producer gas can be used for heat, power, and liquid fuel generation via various synthesis technologies. The main objective of this study is to develop the different thermochemical equilibrium biomass gasification models for the downdraft gasifiers. There are two types of equilibrium models: (a) stoichiometric equilibrium models, (b) Non-stoichiometric equilibrium models. Stoichiometric equilibrium models are based on the chemical reactions involved inside the gasifier. A set of chemical reactions is selected and their equilibrium constants are calculated to predict the syngas composition and behavior of other parameters like gasification agent, temperature, and moisture contents in the available biomass sample. While non-stoichiometric equilibrium models are independent of the reaction mechanism involved in the gasification process. These models are based on the components of the product gas and the equilibrium conditions are achieved by the minimization of the Gibbs free energy to predict the syngas gas composition. These models are easy to develop in modern computers and their applications. Finally, to improve efficiency, these developed models are modified and optimized with the correction factors to get the best suitable validations against the published experimental studies. These correction factors are calculated based on the large data set of experimental results to enhance the efficiency and to reduce the errors of the predicted results. The results of the developed modified models are also compared and validated with the already published experimental and modelling studies. In stoichiometric models, the overall RMSE is reduced from 2.59 and 2.77 to 2.30 and 2.68 respectively and in non-stoichiometric models, the overall RMSE is reduced from 2.25 and 1.86 to 1.49 and 1.23 respectively. For the future perspective, this study can be very helpful in designing the new gasification system to integrate the modern combine heat and power generation systems and synthesis of renewable liquid fuels through liquefaction technologies.