열화상 영상을 활용한 CNN-Transformer 네트워크의 공장 설비 이상 진단 방법

김동현,황호성,김호철 대한전자공학회 2023 전자공학회논문지 Vol.60 No.3

In this paper, we propose a deep learning algorithm optimized for diagnosing factory facility abnormalities using thermal imaging. For this purpose, the contrast of the thermal image is clearly converted with the contrast enhancement algorithm to enhance the edge information. After that, the Convolution Vision Transformer (CvT) developed using only the advantages of Convolution Neural Network (CNN) and Transformer Network is modified to suit the diagnosis of thermal image-based failure facility abnormalities. Experiments were conducted by extracting normal and abnormal images of factory facilities from the thermal image provided by AI Hub. Through this, we confirmed the excellent performance of 98.79% which is higher accuracy than CNN-based ResNet, EfficientNet, and Transformer-based Vision Transformer (ViT), SwinT (Swin Transformer), which are commonly used in the existing computer vision field. In conclusion, it was confirmed that when using the CNN and Transformer fusion network, it shows better performance than the factory facility failure diagnosis algorithm using other thermal imaging images. 본 논문에서는 열화상 영상을 이용한 공장 설비 이상 진단에 최적화된 딥러닝 알고리즘을 제안한다. 이를 위하여 대조도 향상 알고리즘으로 열화상 영상의 대조도를 명확하게 변환하여 가장자리 정보를 강화하여 준다. 그 후에 Convolution Neural Network(CNN)와 Transformer Network 각각의 장점만을 이용하여 개발된 CvT(Convolutional vision Transformer)를 열화상 영상 기반의 고장 설비 이상 진단에 적합하게 수정한 modified CvT 개발을 통하여 공장 설비의 이상을 진단한다. AI Hub에서 제공되는 열화상 영상 중에서 공장 설비의 정상 및 이상 영상들을 추출하여 실험을 진행하였으며, 이를 통하여 기존 컴퓨터 비전 분야에서 보편적으로 사용되고 있는 CNN 기반의 ResNet, EfficientNet 그리고 Transformer 기반의 ViT(Vision Transformer), SwinT(Swin Transformer)보다 높은 정확도인 98.79%의 우수한 성능을 확인하였다. 결론적으로 CNN과 Transformer 융합 네트워크를 활용하였을 때 다른 열화상 영상을 이용한 공장 설비 이상 진단 알고리즘보다 우수한 성능을 보여준다는 것을 확인하였다.

Optimization and Analysis of Leakage Reactance for a Converter Transformer of the Electric Transport System

Dawood Kamran,Kömürgöz Güven,Işik Fatih 대한전기학회 2024 Journal of Electrical Engineering & Technology Vol.19 No.1

Converter transformers are widely used in the electric transport system and it is crucial equipment for the rectifer unit of the transport’s tracking substations. Leakage reactance is a crucial criterion during the development of a converter transformer. Almost all of the analytical methods consider only the axial leakage fux density during the evaluation of the leakage reactance. Radial leakage fux density is neglected in the analytical methods. Neglecting the radial leakage fux density during the computation of the leakage reactance in the two-winding transformers and other power transformers does not signifcantly afect the results. However, in the case of the converter transformers, radial leakage fux density also needs to be fully considered. In some of the converter transformers; windings, core, insulation material, and other parts of the transformer are so complex that analytical methods are impossible or difcult to implement. Hence, any other method is needed to evaluate the diferent parameters of the transformer. The optimal selection of leakage reactance is an important parameter during the design of the converter transformer. Numerical computational methods are one of the most commonly used techniques to solve and analyse the complex models of transformers. To accurately compute the leakage reactance of the electric transport system transformer (traction transformer), a transient method is used, which considers the efect of the radial leakage fux density. A prototype converter transformer of the electric transport system has been developed to obtain the experimental results. A transient method results and prototype transformer results show excellent agreement and verify the correctness of the fnite element model. The results of the traditional analytical and magnetostatics fnite element analysis are also compared with the short-circuit experimental test.

Modelling and study on the output flow characteristics of expansion energy used hydropneumatic transformer

Yan Shi,Tiecheng Wu,Maolin Cai,Chong Liu 대한기계학회 2016 JOURNAL OF MECHANICAL SCIENCE AND TECHNOLOGY Vol.30 No.3

Hydropneumatic transformer (short for HP transformer) is used to pump pressurized hydraulic oil. Whereas, due to its insufficient usage of energy and low efficiency, a new kind of HP transformer: EEUHP transformer (Expansion energy used hydropneumatic transformer) was proposed. To illustrate the characteristics of the EEUHP transformer, a mathematical model was built. To verify the mathematical model, an experimental prototype was setup and studied. Through simulation and experimental study on the EEUHP transformer, the influence of five key parameters on the output flow of the EEUHP transformer were obtained, and some conclusions can be drawn. Firstly, the mathematical model was proved to be valid. Furthermore, the EEUHP transformer costs fewer of compressed air than the normal HP transformer when the output flow of the two kinds of transformers are almost same. Moreover, with an increase in the output pressure, the output flow decreases sharply. Finally, with an increase in the effective area of hydraulic output port, the output flow increases distinctly. This research can be referred to in the performance and design optimization of the EEUHP transformers.

CNN 모델과 Transformer 조합을 통한 토지피복 분류 정확도 개선방안 검토

심우담,이정수 한국지리정보학회 2024 한국지리정보학회지 Vol.27 No.1

본 연구는 Transformer 모듈을 기반으로 다양한 구조의 모델을 구성하고, 토지피복 분류를 수행하여 Transformer 모듈의 활용방안 검토를 목적으로 하였다. 토지피복 분류를 위한 딥러닝 모델은 CNN 구조를 가진 Unet 모델을 베이스 모델로 선정하였으며, 모델의 인코더 및 디코더 부분을 Transformer 모듈과 조합하여 총 4가지 딥러닝 모델을 구축하였다. 딥러닝 모델의 학습과정에서 일반화 성능 평가를 위해 같은 학습조건으로 10회 반복하여 학습을 진행하였다. 딥러닝 모델의 분류 정확도 평가결과, 모델의 인코더 및 디코더 구조 모두 Transformer 모듈을 활용한 D모델이 전체 정확도 평균 약 89.4%, Kappa 평균 약 73.2%로 가장 높은 정확도를 보였다. 학습 소요시간 측면에서는 CNN 기반의 모델이 가장 효율적이었으나 Transformer 기반의 모델을 활용할 경우, 분류 정확도가 Kappa 기준 평균 0.5% 개선되었다. 차후, CNN 모델과 Transformer의 결합과정에서 하이퍼파라미터 조절과 이미지 패치사이즈 조절 등 다양한 변수들을 고려하여 모델을 고도화 할 필요가 있다고 판단된다. 토지피복 분류과정에서 모든 모델이 공통적으로 발생한 문제점은 소규모 객체들의 탐지가 어려운 점이었다. 이러한 오분류 현상의 개선을 위해서는 고해상도 입력자료의 활용방안 검토와 함께 지형 정보 및 질감 정보를 포함한 다차원적 데이터 통합이 필요할 것으로 판단된다. This research aimed to construct models with various structures based on the Transformer module and to perform land cover classification, thereby examining the applicability of the Transformer module. For the classification of land cover, the Unet model, which has a CNN structure, was selected as the base model, and a total of four deep learning models were constructed by combining both the encoder and decoder parts with the Transformer module. During the training process of the deep learning models, the training was repeated 10 times under the same conditions to evaluate the generalization performance. The evaluation of the classification accuracy of the deep learning models showed that the Model D, which utilized the Transformer module in both the encoder and decoder structures, achieved the highest overall accuracy with an average of approximately 89.4% and a Kappa coefficient average of about 73.2%. In terms of training time, models based on CNN were the most efficient. however, the use of Transformer-based models resulted in an average improvement of 0.5% in classification accuracy based on the Kappa coefficient. It is considered necessary to refine the model by considering various variables such as adjusting hyperparameters and image patch sizes during the integration process with CNN models. A common issue identified in all models during the land cover classification process was the difficulty in detecting small-scale objects. To improve this misclassification phenomenon, it is deemed necessary to explore the use of high-resolution input data and integrate multidimensional data that includes terrain and texture information.

Determining the reverse fault current by the type of transformer and Distributed Generation in distribution system during the single-line to ground fault

Cho, Namhun,Yun, Sangwon,Jung, Jaesung PERGAMON 2019 RENEWABLE AND SUSTAINABLE ENERGY REVIEWS Vol.109 No.-

<P><B>Abstract</B></P> <P>Recent increases in Distributed Generation (DG) into distribution system have led to changes in the conventional protection scheme because of the reverse fault current coming from the DG side. However, it is not yet clear whether the reverse fault current is caused by the type of interconnection transformer or the type of DG. Therefore, for reliable coordination between protective devices, it is necessary to analyze the reverse fault current contribution from the DG side to fault location according to the system configuration. In this paper, the single-line to ground fault current contribution is analyzed when different combinations of the transformer and DG are interconnected in the system. For this, a methodology for calculating the fault current according to the type of transformer and DG is introduced. Using this methodology, the fault current contribution from each transformer and DG is analyzed separately to eliminate the mixed contribution from both the transformer and DG. The mixed fault current contribution from the different combinations of these devices is then analyzed. For the interconnection transformer, delta-grounded wye (D-Yg), Yg-Yg, and Yg-D transformer configurations are considered; for DG, synchronous machine-based and inverter-based DG are considered. Based on this analysis, the present paper provides effective guidelines for determining the reverse fault current according to the type of transformer and DG in the distribution system, enabling reliable protection coordination under the integration of DG.</P> <P><B>Highlights</B></P> <P> <UL> <LI> The fault analysis is introduced according to the type of transformer and DG. </LI> <LI> The sole reverse fault current contribution is determined by the type of transformer. </LI> <LI> The sole reverse fault current contribution is determined by the type of DG. </LI> <LI> The combined reverse fault current contribution is determined by transformer and DG. </LI> <LI> This paper provides the effective guideline to achieve the reliable protection coordination. </LI> </UL> </P>

Improvement of Thermal Balance in Step-down Piezo Transformers with Ring-dot shapes

김인성,정순종,김민수,송재성,Vo Viet Thang 한국물리학회 2012 THE JOURNAL OF THE KOREAN PHYSICAL SOCIETY Vol.60 No.2

The design of ring-dot-shape piezoelectric transformers for thermal radiation is investigated in this paper. The temperature distribution at the center was improved by ring-dot shape. One of the most important factors affecting piezoelectric transformers is the temperature; hence, the piezoelectric material and construction should be improved to get transformers with higher power at a lower normal temperature increase. Obviously, internal losses generate heat, which increases the temperature of piezoelectric transformers, especially at high power, and changes the characteristics of the transformers. In this work, the modeling of a multilayer step-down piezoelectric transformer with a square shape with a central hole was studied using the ATILA software before its fabrication. Firstly, the parameters of a hard piezoelectric ceramic were measured from the fabricated specimens and used in the simulation. Moreover, the effects of frequency and load resistance on the electrical properties were studied. Then, on investigations of temperature at different load resistances and of the temperature distribution were carried out. Thus, the electrical properties and the temperature of step-down piezoelectric transformers corresponding to the simulation and fabricated by using piezoelectric ceramics were measured and compared to the simulated results. The design of ring-dot-shape piezoelectric transformers for thermal radiation is investigated in this paper. The temperature distribution at the center was improved by ring-dot shape. One of the most important factors affecting piezoelectric transformers is the temperature; hence, the piezoelectric material and construction should be improved to get transformers with higher power at a lower normal temperature increase. Obviously, internal losses generate heat, which increases the temperature of piezoelectric transformers, especially at high power, and changes the characteristics of the transformers. In this work, the modeling of a multilayer step-down piezoelectric transformer with a square shape with a central hole was studied using the ATILA software before its fabrication. Firstly, the parameters of a hard piezoelectric ceramic were measured from the fabricated specimens and used in the simulation. Moreover, the effects of frequency and load resistance on the electrical properties were studied. Then, on investigations of temperature at different load resistances and of the temperature distribution were carried out. Thus, the electrical properties and the temperature of step-down piezoelectric transformers corresponding to the simulation and fabricated by using piezoelectric ceramics were measured and compared to the simulated results.

트랜스포머와 BERT로 구현한 한국어 형태소 분석기의 성능 분석

최용석,이공주 한국정보과학회 2020 정보과학회논문지 Vol.47 No.8

This paper introduces a Korean morphological analyzer using the Transformer, which is one of the most popular sequence-to-sequence deep neural models. The Transformer comprises an encoder and a decoder. The encoder compresses a raw input sentence into a fixed-size vector, while the decoder generates a morphological analysis result for the vector. We also replace the encoder with BERT, a pre-trained language representation model. An attention mechanism and a copying mechanism are integrated in the decoder. The processing units of the encoder and the decoder are eojeol-based WordPiece and morpheme-based WordPiece, respectively. Experimental results showed that the Transformer with fine-tuned BERT outperforms the randomly initialized Transformer by 2.9% in the F1 score. We also investigated the effects of the WordPiece embedding on morphological analysis when they are not fully updated in the training phases. 본 논문은 Transformer로 구현한 한국어 형태소 분석기를 다룬다. Transformer는 최근에 가장 널리 사용되는 sequence-to-sequence 모델 중 하나이다. Transformer는 인코더와 디코더로 구성되어 있는데 인코더는 원문을 고정된 크기의 벡터로 압축시키고 디코더는 이 벡터를 이용하여 형태소 분석 결과를 생성해 낸다. 본 논문에서는 또한 Transformer의 인코더를 BERT로 대체해 본다. BERT는 대용량의 학습데이터를 이용하여 미리 학습시켜 놓은 언어 표현 모델이다. 디코더에는 주의 메커니즘과 복사 메커니즘을 도입하였다. 인코더와 디코더에서의 처리 단위는 각각 어절 단위 WordPiece와 형태소 단위의 WordPiece를 사용하였다. 실험을 통해, BERT의 파라미터를 문제에 맞게 재조정했을 때의 성능이 Transformer를 임의의 값으로 초기화하여 사용했을 때에 비해 F1에서 2.9%의 성능 향상을 보임을 알 수 있었다. 또한 학습단계에서 충분히 학습되지 못한 WordPiece의 임베딩이 형태소 분석에 어떤 영향을 미치는지도 살펴보았다.

변압기의 파라미터 추출과 전기적 모델링 방법에 관한 연구

박준우(Jun-Woo Park),윤재중(Jae-Jung Yun) 한국조명·전기설비학회 2017 조명·전기설비학회논문지 Vol.31 No.5

The electric operation of a practical transformer, which is used in the various electric apparatus, is not easy to interpret, because it has some considerable parameters compared to a ideal transformer. In addition, although the commercial simulation tools provide electric models of a practical transformer for users, it is difficult to analyze intuitively the effects of the circuit performance caused by parameters of practical transformer such as a leakage inductance and mutual inductance. To solve these problems, this paper presents a method for parameter extractions and circuit modeling of a practical transformer. The equivalent model of a practical transformer by the suggested method consists of an ideal transformer model, two resistors, two leakage inductors, and a mutual inductor. The parameters for a practical transformer is extracted by open and short circuit analysis methods, and the electric model of an ideal transformer is made by solving the requirements. To verify the effectiveness of the equivalent model of a practical transformer, a test bed with two winding transformer is built. The Pspice model of a practical transformer was made of parameters extracted by a test bed, and a comparison between simulation and experiment results was given in this paper. As a result, the difference in the primary current and secondary current between simulated and experimental results is 1.2% and 1%, respectively.

Analysis of the Contactless Power Transfer System Using Modelling and Analysis of the Contactless Transformer

Myunghyo Ryu,Jonghyun Kim,Juwon Baek,Honnyong Cha 대한전기학회 2006 Journal of Electrical Engineering & Technology Vol.1 No.3

In this paper, the electrical characteristics of the contactless transformer is presented using the conventional coupled inductor theory. Compared with the conventional transformer, the contactless transformer has a large airgap, long primary wire and multi-secondary wire. As such, the contactless transformer has a large leakage inductance, small magnetizing inductance and poor coupling coefficient. Therefore, large magnetizing currents flow through the entire primary system due to small magnetizing inductance, resulting in low overall system efficiency. In high power applications, the contactless transformer is so bulky and heavy that it needs to be split by some light and small transformers. So, the contactless transformer needs several small transformer modules that are connected in series or parallel to transfer the primary power to the secondary one. This paper shows the analysis and measurement results of each contactless transformer module and comparison results between the series- and parallel-connection of the con tactless transformer. The results are verified on the simulation based on the theoretical analysis and the 30㎾ experimental prototype.

Analysis of the Contactless Power Transfer System Using Modeling and Analysis of the Contactless Transformer

Ryu Myung-Hyo,Kim Jong-Hyun,Baek Ju-Won,Cha Hon-Nyong The Korean Institute of Electrical Engineers 2006 Journal of Electrical Engineering & Technology Vol.1 No.3

In this paper, the electrical characteristics of the contactless transformer is presented using the conventional coupled inductor theory. Compared with the conventional transformer, the contactless transformer has a large airgap, long primary wire and multi-secondary wire. As such, the contactless transformer has a large leakage inductance, small magnetizing inductance and poor coupling coefficient. Therefore, large magnetizing currents flow through the entire primary system due to small magnetizing inductance, resulting in low overall system efficiency. In high power applications, the contactless transformer is so bulky and heavy that it needs to be split by some light and small transformers. So, the contactless transformer needs several small transformer modules that are connected in series or parallel to transfer the primary power to the secondary one. This paper shows the analysis and measurement results of each contactless transformer module and comparison results between the series- and parallel-connection of the contactless transformer. The results are verified on the simulation based on the theoretical analysis and the 30kW experimental prototype.