The buffer is a critical barrier component in an engineered barrier system, and its purpose is to prevent potential radionuclides from leaking out from a damaged canister by filling the void in the repository. No experimental parameters exist that can describe the buffer expansion phenomenon when Kyeongju bentonite, which is a buffer candidate material available in Korea, is exposed to groundwater. As conventional experiments to determine these parameters are time consuming and complicated, simple swelling pressure tests, numerical modeling, and machine learning are used in this study to obtain the parameters required to establish a numerical model that can simulate swelling. Swelling tests conducted using Kyeongju bentonite are emulated using the COMSOL Multiphysics numerical analysis tool. Relationships between the swelling phenomenon and mechanical parameters are determined via an artificial neural network. Subsequently, by inputting the swelling tests results into the network, the values for the mechanical parameters of Kyeongju bentonite are obtained. Sensitivity analysis is performed to identify the influential parameters. Results of the numerical analysis based on the identified mechanical parameters are consistent with the experimental values.

Visual inspection methods have limitations, such as reflecting the subjective opinions of workers. Moreover, additional equipment is required when inspecting the high-rise buildings because the height is limited during the inspection. Various methods have been studied to detect concrete cracks due to the disadvantage of existing visual inspection. In this study, a crack detection technology was proposed, and the technology was objectively and accurately through AI. In this study, an efficient method was proposed that automatically detects concrete cracks by using a Convolutional Neural Network(CNN) with the Orthomosaic image, modeled with the help of UAV. The concrete cracks were predicted by three different CNN models: AlexNet, ResNet50, and ResNeXt. The models were verified by accuracy, recall, and F1 Score. The ResNeXt model had the high performance among the three models. Also, this study confirmed the reliability of the model designed by applying it to the experiment.

APro, a modularized framework of the process-based total system performance assessment, has been developed by KAERI to simulate the radionuclide transport in geological disposal system considering multi-physics phenomena. However, the target problem including more than 10,000 boreholes and over 100,000 years of simulation time is computationally challenging to deal with numerical solvers provided by COMSOL Multiphysics constituting APro. To alleviate the computational burden, machine learning (ML) techniques have been studied to develop a surrogate model replacing the heavy computation part. In recent studies, attempts have been made to integrate the knowledge of physics and numerical methods into the ML model for partial differential equations (PDEs). Unlike conventional ML approaches solely relying on data-driven method, the integration can help to make the ML model more specialized for solving PDEs. The hybrid neural network (NN) solver method is one of the strategies to develop more efficient PDE solver by interleaving NN with numerical solvers like finite element method (FEM). The hybrid NN model on the premise of numerical solver is easier to train and more stable than the purely data-driven model. For example, one previous study has used the hybrid NN model as a corrector for an incomplete numerical solver for the advection-diffusion problem. In every time step of simulation, NN corrects the error of incomplete solution obtained by a relaxed numerical solver with coarse meshing. The simulation in the next time step starts from the corrected solution, so NN interacts with the numerical solver iteratively. If the corrector is successfully trained, the incomplete but fast solver with corrector can provide reliable results comparable to the original massive solver. This study adopts the hybrid concept to develop a surrogate model for the near-field region, which is the heavy computation part in the simulation of geological disposal system. Various incomplete models such as coarse meshing or emptying the borehole domain are studied to construct a hybrid NN solver. This study also covers how to embed the hybrid NN in COMSOL Multiphysics to train and use it during the simulation.

In this study, the multi-lane detection problem is expressed as a CNN-based regression problem, and the lane boundary coordinates are selected as outputs. In addition, we described lanes as fifth-order polynomials and distinguished the ego lane and the side lanes so that we could make the prediction lanes accurately. By eliminating the network branch arrangement and the lane boundary coordinate vector outside the image proposed by Chougule’s method, it was possible to eradicate meaningless data learning in CNN and increase the fast training and performance speed. And we confirmed that the average prediction error was small in the performance evaluation even though the proposed method compared with Chougule’s method under harsher conditions. In addition, even in a specific image with many errors, the predicted lanes did not deviate significantly, meaningful results were derived, and we confirmed robust performance.

The use of heat exchangers in various applications such as chemical, air conditioning systems, fuel processing, and power industries is increasing. In order to improve the performance of the heat exchanger, the problem of bonding quality of the copper tube, which is a major member, is emerging. However, since the copper tube is in the form of a pipe, it is difficult to identify internal defects with external factors. In this study, a thermal imaging camera was used to develop and verify an algorithm for detecting defects in the brazing part, and in the process, the brazing performance characteristics were analyzed according to the electrode position, and finally, a learning model was developed and performance evaluation was performed. It was confirmed that the method of supplying heat to the base material and melting the filler metal through the heat transfer effect is more effective than supplying heat input to the filler metal in the bonding process of copper tubes through high-frequency induction heating brazing. Thermal image data was used to develop a defect discrimination model, and 80% of training data and 20% of test data were selected, and a neural network-based single-layer copper tube brazing defect discrimination model was developed through k-Flod cross-validation., the prediction accuracy of 95.2% was confirmed as a result of the error matrix analysis.

인터넷은 전통적인 미디어를 대체하고 주요 뉴스 미디어 플랫폼 중 하나가 되었습니다. 인터넷 소스의 뉴스는 접근성이 좋고 편리하기 때문에 기존 뉴스 소스에 비해 빠르고 간단하게 이동할 수 있습니다. 그러나 가짜 뉴 스가 대량으로 발생하고 정치적, 상업적 이유로 온라인 커뮤니티에 퍼지면서 확인되지 않은 소식통으로부터 입수한 모든 언론 보도가 진짜인 것은 아니다. 가짜 뉴스는 이론적으로나 의도적으로 독자들을 속이거나 잘못 알릴 수 있다. 왜냐하면 사람들은 오프라인 커뮤니티에 영향을 미칠 수 있는 어떤 정보에도 쉽게 얽히게 되기 때문이다. 일부 수동 웹사이트는 정보가 사실인지 확인하도록 설계되어 있지만 온라인, 특히 웹에서 빠르게 확 산되는 정보의 양은 확장되지 않습니다. 이 문제를 해결하기 위해 자동 팩트체크 어플리케이션은 확장성과 자 동화의 요건에 대응하도록 설계되었습니다. 그러나 현재 애플리케이션 방법에는 기계 학습 분류 모델 성능을 개선하기 위해 가짜 뉴스 특징을 식별하는 포괄적인 다차원 데이터 세트가 없다. 이 문제를 해결하기 위해 본 연구논문에서는 사용자가 기사의 제목을 입력하면 데이터를 분류하는 Formb 챗봇을 제안했다. 이 연구 작업 에서 데이터 집합의 분류는 반복 신경망(RNN)과 장기 단기 기억(LSTM) 모델을 사용하여 수행되었다. 가짜 및 실제 뉴스 데이터 세트는 사전 처리되어 모델을 교육하는 데 사용됩니다. 저장된 모델은 지정된 입력 텍스트 의 신뢰성을 확인하기 위해 불일치 서버에 배포됩니다. Disconsid API는 python 파일을 chatbot으로 실행할 수 있는 액세스를 제공합니다. 분석 측면에서, 제안된 모델은 96.77%의 정확도로 CNN와 같은 기존 뉴럴 네트 워크 모델을 능가한다.

범밀도이론함수(Density Functional Theory, DFT) 기반의 제일원리전산모사는 기저상태의 DFT 에너지를 구하는데 많은 시간소요 및 전산자원을 소모하였다. 이러한 막대한 전산자원의 소모는 DFT 계산에서 고려할 원자수를 수 백개 이 하로 제한되게 되었으며, 이를 해결하기 위해서는 전자구조 계산이 아닌 원자의 환경 내에 원자간 상호작용을 정의 (Force Field, 힘장)하고 이를 통해 주어진 조성 혹은 구조에 따른 에너지를 빠르게 예측 할 수 있어야 한다. 본 논문에서 는 Behler-Parrinello가 제시한 인공신경망 모델을 활용해 인공지능 다원계 힘장을 개발하고 코발트-구리 산화물의 조성에 따른 에너지를 예측하고 안정한 구조를 탐색하는 연구를 수행하였다. 인공신경망 기술로 부터 구리-코발트 산화물에 대 해 15.7 meV/atom의 에너지 오차와 단위거리당 힘 103.6 meV/Å의 정확도를 가지는 인공신경망 포텐셜을 개발하였다. 이 방법으로 빠르고 정확하게 CuCoO 표면구조의 산소 결함률에 따른 생성에너지를 계산할 수 있었고, 에너지 컨벡스 홀을 도시 조성에 따른 안정한 구조를 예측하였다.

선박과 교각이 충돌하면 생명과 안전에 큰 위협이 될 수 있다. 따라서 선박-교각 충돌력 영향 인자를 식별하고 다양한 충돌 조 건에서의 충돌력에 대한 연구의 필요성이 있다. 본 논문에서는 선박-교각 충돌의 유한요소 모델을 설정하고, 수치 시뮬레이션을 통해 선 적상태, 운항속도, 충돌 각도의 세 가지 입력조건을 조합하여 50가지 케이스에서의 선박-교각 최대 충돌력을 계산하였다. 계산된 유한요 소해석 결과를 사용하여 신경망 추정 모델을 학습하고 최대 충돌력을 추정함으로써 빠른 시간에 최대 충돌력을 추정하는 프로세스를 제 안하였다. 신경망 예측 모델은 가장 기초적인 역전파 신경망과 시간정보를 고려할 수 있는 순환신경망인 Elman 신경망 2가지 모델을 사 용하였다. 10가지 케이스의 테스트 데이터로 시험한 결과 Elman 신경망을 사용했을 경우에 평균상대오차가 4.566%로 역전파 신경망보다 나은 최대 충돌력 추정이 가능함을 확인하였고 8가지 케이스에서 5%이하의 상대오차를 보여 주었다. 본 신경망을 이용한 최대 충돌력 추 정법은 유한요소해석을 수행하지 않아도 되므로 계산 시간이 짧아 선박 항해 중 충돌을 회피할 수 없는 경우 피해를 최소화하는 의사결 정의 기초 방법으로 사용할 수 있다.

선박으로부터 발생하는 배출가스에 대한 규제가 강화되고 이를 해결하기위한 대안으로 전기추진시스템의적용이 대형상선에서 부터 중ㆍ소형선박에 이르기까지 그 사용이 증가되고있다. 전기추진시스템의 효율 향상을 위한 방법으로 발전원의 개선, 배터리ㆍ연료전 지ㆍ태양광 등의 친환경 발전원의 시스템 연계 및 정류기, 전력변환장치, 추진전동기의 개발과 제어방식의 연구를 들 수 있다. 그 중 정류 방식에 있어 상천이변압기과 다이오드를 이용하는 방식이 널리 사용되었으나, 직류배전을 이용한 친환경발전원의 계통 연계, 가변속 발 전원의 사용, 중ㆍ소형 전기추진시스템의 적용을 통해 전력용 반도체 소자를 이용한 AFE정류기에 대한 수요가 증대되고 있다. 이러한 AFE 정류기를 제어하는 방식에 있어 기존의 비례적분제어기가 아닌 신경회로망을 이용한 방식을 본 연구에서는 제안하였다. 기존의 제 어기 데이터를 활용하여 Matlab/Simulink를 통해 학습한 신경회로망제어기를 설계하고 PSIM을 통해 설계된 정류시스템에 신경회로망 제어 기를 적용하여 부하변동에 따른 직류출력단의 파형과 역률 개선의 유효성을 확인하였다. 이는 공간이 협소한 중소형 친환경 선박의 정류 시스템으로써 적용이 가능하다.

Deep convolutional network is a deep learning approach to optimize image recognition. This study aimed to apply DCNN to the reading of mandibular cortical thinning in digital panoramic radiographs. Digital panoramic radiographs of 1,268 female dental patients (age 45.2 ± 21.1yrs) were used in the reading of the mandibular cortical bone by two maxillofacial radiologists. Among the subjects, 535 normal subject’s panoramic radiographs (age 28.6 ±7.4 yrs) and 533 those of osteoporosis pationts (age 72.1 ± 8.7 yrs) with mandibular cortical thinning were used for training DCNN. In the testing of mandibular cortical thinning, 100 panoramic radiographs of normal subjects (age 26.6 ± 4.5 yrs) and 100 mandibular cortical thinning (age 72.5 ± 7.2 yrs) were used. The sensitive area of DCNN to mandibular cortical thinning was investigated by occluding analysis. The readings of DCNN were compared by two maxillofacial radiologists. DCNN showed 97.5% accuracy, 96% sensitivity, and 99% specificity in reading mandibular cortical thinning. DCNN was sensitively responded on the cancellous and cortical bone of the mandibular inferior area. DCNN was effective in diagnosing mandibular cortical thinning.

As a new entertainment and social way, online games now have a huge and increasing user group, so it is of great significance to identify the data stream of online games. Using the excellent nonlinear fitting ability of BP neural network and the advantages of global search of genetic algorithm, the initial weights and thresholds of BP neural network are optimized, and the BP neural network model optimized by genetic algorithm is established. The muti-dimensional input information is proposed to identify online game data streams. Through the experimental simulation, it shows that the selected muti-dimensional information and the established model can be well applied to online game stream recognition.

Gravity Recovery and Climate Experiment (GRACE) gravimeter satellites observed the Earth gravity field with unprecedented accuracy since 2002. After the termination of GRACE mission, GRACE Follow-on (GFO) satellites successively observe global gravity field, but there is missing period between GRACE and GFO about one year. Many previous studies estimated terrestrial water storage (TWS) changes using hydrological models, vertical displacements from global navigation satellite system observations, altimetry, and satellite laser ranging for a continuity of GRACE and GFO data. Recently, in order to predict TWS changes, various machine learning methods are developed such as artificial neural network and multi-linear regression. Previous studies used hydrological and climate data simultaneously as input data of the learning process. Further, they excluded linear trends in input data and GRACE/GFO data because the trend components obtained from GRACE/GFO data were assumed to be the same for other periods. However, hydrological models include high uncertainties, and observational period of GRACE/GFO is not long enough to estimate reliable TWS trends. In this study, we used convolutional neural networks (CNN) method incorporating only climate data set (temperature, evaporation, and precipitation) to predict TWS variations in the missing period of GRACE/GFO. We also make CNN model learn the linear trend of GRACE/GFO data. In most river basins considered in this study, our CNN model successfully predicts seasonal and long-term variations of TWS change.

자기공명영상은 고해상도의 연부조직에 대한 영상정보를 제공하며, 뇌종양 등 연부조직 진단에 활용된다. 본 연구는 합성곱신경망 인공지능을 통해 뇌종양 자기공명영상 분류성능을 확인해 보고자 한다. 4개 종류로 구분된 3264 장의 MRI 데이터 세트(data set)를 이용하였으며, 인공지능 학습을 위해 훈련용 데이터와 시험용 데이터를 9 : 1, 훈련용 데이터의 10%를 검증용 데이터로 구분하였다. 합성곱신경망은 기본 CNN과 VGG16으로 구성하였으며, 학습 평가는 정확도와 손실율로 확인하였으며, 생성된 모델을 통해 분류성능 정확도를 확인하였다. 실험 결과 과적합은 없었으며, 분류성능은 기본 CNN과 VGG16 각각 67%와 80%의 분류성능을 보였다. 도출된 뇌종양 자기공명영상 분류 결과를 통해 자기공명영상과 인공지능 접목에 관한 기초 자료로 사용될 수 있을 것이라 사료된다.

The important thing in the field of deep learning is to find out the appropriate hyper-parameter for image classification. In this study, the main objective is to investigate the performance of various hyper-parameters in a convolutional neural network model based on the image classification problem. The dataset was obtained from the Kaggle dataset. The experiment was conducted through different hyper-parameters. For this proposal, Stochastic Gradient Descent without momentum (SGD), Adaptive Moment Estimation (Adam), Adagrad, Adamax optimizer, and the number of batch sizes (16, 32, 64, 120), and the number of epochs (50, 100, 150) were considered as hyper-parameters to determine the losses and accuracy of a model. In addition, Binary Cross-entropy Loss Function (BCLF) was used for evaluating the performance of a model. In this study, the VGG16 convolutional neural network was used for image classification. Empirical results demonstrated that a model had minimum losses obtain by Adagrad optimizer in the case of 16 batch sizes and 50 epochs. In addition, the SGD with a 32 batch sizes and 150 epochs and the Adam with a 64 batch sizes and 50 epochs had the best performance based on the loss value during the training process. Interestingly, the accuracy was higher while performing the Adagrad and Adamax optimizer with a 120 batch sizes and 150 epochs. In this study, the Adagrad optimizer with a 120 batch sizes and 150 epochs performed slightly better among those optimizers. In addition, an increasing number of epochs can improve the performance of accuracy. It can help to create a broader scope for further experiments on several datasets to perceive the suitable hyper-parameters for the convolutional neural network. Dataset: https://www.kaggle.com/c/dogs-vs-cats/data