In this study, a new model using artificial neural networks is proposed to improve the thickness error between the plates, which occurs when the rolling conditions change a lot during the thick rolling. The model was developed by using Python, and the input values are the change in the finish rolling temperature between the plates, the change in target tensile strength, the change in target thickness, and the change in rolling force. The new model is 31.76% better than the existing model based on the standard deviation value of the thickness error. This result is expected to reduce quality costs when applied to online models at actual production sites in the future.

This study deals with the application of an artificial neural network (ANN) model to predict power consumption for utilizing seawater source heat pumps of recirculating aquaculture system. An integrated dynamic simulation model was constructed using the TRNSYS program to obtain input and output data for the ANN model to predict the power consumption of the recirculating aquaculture system with a heat pump system. Data obtained from the TRNSYS program were analyzed using linear regression, and converted into optimal data necessary for the ANN model through normalization. To optimize the ANN-based power consumption prediction model, the hyper parameters of ANN were determined using the Bayesian optimization. ANN simulation results showed that ANN models with optimized hyper parameters exhibited acceptably high predictive accuracy conforming to ASHRAE standards.

Engineering design involves making numerous decisions as the design process. These decisions can be broadly categorized into selection decisions and compromise decisions. The outcomes of these decisions heavily depend on the designer's intentions, highlighting the need to systematically and accurately incorporate the designer's intentions. The Analytic Hierarchy Process (AHP) is a design technique that systematically reflects the designer's intentions by hierarchically analyzing and evaluating ambiguous decision problems. Therefore, in this study, effective optimal structure designs that maximally reflect the designer's intentions were confirmed by introducing AHP (Analytic Hierarchy Process) and Neural Network into the foundational decision-making process of engineering design.

Tomato is one of the major widely cultivated crops around the world. The leaf area is directly related to the total amount of photosynthesis, which affects the yield and quality of the fruit. Traditional methods of measuring the leaf area are time-consuming and can cause damage to the leaves. To address these problems, various studies are being conducted for measuring the leaf area. In this study, we introduced a model to estimate the leaf area using images of tomatoes. Using images captured by a camera, we measured the leaf length and width and used linear regression analysis to derive the leaf area estimation formula. Furthermore, we used a Neural Network (NN) for additional analysis to compare the accuracy of the models. Initially, to verify the reliability of the image data, we conducted a correlation analysis between the actual measurement data and the image data, which showed a high positive correlation. The leaf area estimation model presented 23 estimation formulas. We used regression analysis to estimate the coefficients of each model and also used employed an artificial neural network analysis to derive high R-squared (R2) values and low Root Mean Square Error (RMSE) values. Among the estimation formulas, the ninth model showed the highest reliability with an R-squared value of 0.863. We conducted a verification experiment to confirm the accuracy of the selected model, and the R-squared value was 0.925. This study confirmed the reliability of data measured from images and the reliability of the leaf area estimation model using image data. These methods are expected to be an important tool in agriculture, using imaging equipment for measuring and monitoring the crop growth.

이 연구는 위성사진을 활용하여 건설지점의 기대풍속을 예측하기 위한 인공신경망 방법론을 제안한다. 제안된 방법은 기존 의 엔지니어의 판단을 대체하여, Auto-Encoder를 사용해 지형적 특성을 정량화하고, 이를 바탕으로 대상지점과 유사한 지역의 관측소 풍속 데이터를 선형 조합해 기대 풍속을 예측한다. 또한, 머신러닝과 인공신경망을 활용한 종단간 풍속 예측 모델을 제안하고, 성능을 비교 분석하였다. 그 결과 관측소의 풍속 데이터의 선형 조합보다는 종단간 모델을 구성하는 방법이 더 높은 정확도를 보였으며, 특히 Graph Neural Network (GNN)이 Multi-Layer Perceptron (MLP)에 비해 상당히 우수한 예측 성능을 나타내었다.

PURPOSES : This study aims to determine whether machine learning techniques based on the results of chemical analysis experiments can be rationally applied to evaluate the aging of various asphalt binders used throughout the country. METHODS : We conducted chemical experiments such as FT-IR, H-NMR, C- NMR, and GPC for the three-stage aging levels of eight types of asphalt binders used in the country and utilized two artificial neural network models to determine valid chemical experimentation and conditions for the use of neural modeling through predictions. RESULTS : The M-prop model, which combined the findings from each neural network model into a single artificial neural network model, demonstrated superior predictive performance compared with the M-base model, which assessed aging using two cluster layers. In addition, the minimum amount of data required to achieve 100% predictive accuracy for the target asphalt binders, regardless of the artificial neural network model, was 18, and the amount of training data decreased to less than 50%. CONCLUSIONS : The predictive accuracy of the aging of asphalt binders was significantly enhanced when GPC data was used, indicating that GPC should be prioritized in evaluating the aging of asphalt binders.

본 논문에서는 3차원 엮임 재료의 재료 물성치들을 효율적으로 분석하고 추후 최적설계 연구에 활용하기 위해서 파라메트릭 배치 해석 워크플로우를 제시하였다. 3차원 엮임 재료를 구성하는 와이어들 사이의 간격을 설계 매개변수로 하는 파라메트릭 모델에 대해 서 임의의 변수 조합을 가지는 2,500개의 수치 모델을 생성하였으며, 상용 프로그램인 매트랩과 앤시스의 여러 모듈을 사용하여 체적 탄성계수, 열전도도, 유체투과율과 같은 다양한 재료 물성치들을 배치 해석을 통해서 자동으로 얻어질 수 있도록 구성하였다. 이와 같 이 얻어진 대용량의 재료 물성치 데이터베이스를 활용해서 회귀 분석을 수행하였으며, 그 결과 설계 변수들과 재료 물성치 사이의 경 향성과 수치 해석 결과의 정확도를 검증하였다. 또한 확보된 데이터베이스를 통해서 3차원 엮임 재료의 물성치를 예측할 수 있는 인 공 신경망을 구성하고 학습시켰으며, 그 결과 임의의 설계 매개변수 값들을 가지는 엮임 재료 모델에 대해서 구조 및 유체해석 과정 없 이도 높은 정확도로 재료 물성치들을 추정할 수 있음을 확인하였다.

새만금 내에서는 종종 식물플랑크톤이 증식하기에 알맞은 환경조건이 생성되며 일시에 식물플랑크톤 대증식이 발생하면서 조 류 관리기준을 초과하는 사례가 발생하고 있다. 이를 대비하기 위하여 과학적 예측기법을 토대로, 식물플랑크톤의 종별로 가장 효과적이 고 효율적인 녹조발생 억제 방안을 제안하기 위하여 식물플랑크톤 대증식 가능성을 예측하고, 제어할 수 있는 모델을 개발하였다. 즉, 하 천에서 유입하는 영양염(DIN, PO4-P)을 정책적으로 조절하고, 갑문운영을 통해 호 내 염분을 제어하는 것이다. 먼저 관측치로부터 인공신 경망 알고리즘을 이용해 식물플랑크톤 대증식 가능성을 예측 결과, 모델의 Kappa 수는 0.7889 ~ 1.0000의 범위로, good ~ excellent 수준이었 다. 다음으로 Garson 알고리즘을 이용하여 종별로 설명변수의 중요도를 평가하였고, 또한 DIN 및 염분 값의 변화에 따른 식물플랑크톤 대 량 증식 확률을 예측하였다. 그 결과, 각 종별로 식물플랑크톤의 대증식을 억제할 수 있는 DIN과 염분 농도를 정량적으로 예측할 수 있었 다. 따라서, 향후 새만금과 같은 거대한 인공 호수에서 식물플랑크톤의 대증식을 억제하기 위한 효율적이고 효과적인 대응방안을 마련할 수 있도록 녹조제어모델을 활용할 수 있을 것으로 판단된다.

PURPOSES : This paper proposes an artificial neural network (ANN)-based real-time traffic signal time design model using real-time field data available at intersections equipped with smart intersections. The proposed model generates suitable traffic signal timings for the next cycle, which are assumed to be near the optimal values based on a set of counted directional real-time traffic volumes. METHODS : A training dataset of optimal traffic signal timing data was prepared through the CORSIM Optimal Signal Timing program developed for this study to find the best signal timings, minimizing intersection control delays estimated with CORSIM and a heuristic searching method. The proposed traffic signal timing design model was developed using a training dataset and an ANN learning process. To determine the difference between the traditional pre-time model primarily used in practice and the proposed model, a comparison test was conducted with historical data obtained for a month at a specific intersection in Uiwang, Korea. RESULTS : The test results revealed that the proposed method could reduce control delays for most of the day compared to the existing methods, excluding the peak hour periods when control delays were similar. This is because existing methods focus only on peak times in practice. CONCLUSIONS : The results indicate that the proposed method enhances the performance of traffic signal systems because it rapidly provides alternatives for all-day cycle periods. This would also reduce the management cost (repeated field data collection) required to increase the performance to that level. A robust traffic-signal timing design model (e.g., ANN) is required to handle various combinations of directional demands.

In this study, based on the saturation magnetic flux density experimental values (Bs) of 622 Fe-based bulk metallic glasses (BMGs), regression models were applied to predict Bs using artificial neural networks (ANN), and prediction performance was evaluated. Model performance evaluation was investigated by using the F1 score together with the coefficient of determination (R2 score), which is mainly used in regression models. The coefficient of determination can be used as a performance indicator, since it shows the predicted results of the saturation magnetic flux density of full material datasets in a balanced way. However, the BMG alloy contains iron and requires a high saturation magnetic flux density to have excellent applicability as a soft magnetic material, and in this study F1 score was used as a performance indicator to better predict Bs above the threshold value of Bs (1.4 T). After obtaining two ANN models optimized for the R2 and F1 score conditions, respectively, their prediction performance was compared for the test data. As a case study to evaluate the prediction performance, new Fe-based BMG datasets that were not included in the training and test datasets were predicted using the two ANN models. The results showed that the model with an excellent F1 score achieved a more accurate prediction for a material with a high saturation magnetic flux density.

Lightweight face recognition models, as one of the most popular and long-standing topics in the field of computer vision, has achieved vigorous development and has been widely used in many real-world applications due to fewer number of parameters, lower floating-point operations, and smaller model size. However, few surveys reviewed lightweight models and reimplemented these lightweight models by using the same calculating resource and training dataset. In this survey article, we present a comprehensive review about the recent research advances on the end-to-end efficient lightweight face recognition models and reimplement several of the most popular models. To start with, we introduce the overview of face recognition with lightweight models. Then, based on the construction of models, we categorize the lightweight models into: (1) artificially designing lightweight FR models, (2) pruned models to face recognition, (3) efficient automatic neural network architecture design based on neural architecture searching, (4) Knowledge distillation and (5) low-rank decomposition. As an example, we also introduce the SqueezeFaceNet and EfficientFaceNet by pruning SqueezeNet and EfficientNet. Additionally, we reimplement and present a detailed performance comparison of different lightweight models on the nine different test benchmarks. At last, the challenges and future works are provided. There are three main contributions in our survey: firstly, the categorized lightweight models can be conveniently identified so that we can explore new lightweight models for face recognition; secondly, the comprehensive performance comparisons are carried out so that ones can choose models when a state-of-the-art end-to-end face recognition system is deployed on mobile devices; thirdly, the challenges and future trends are stated to inspire our future works.

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.

This ammonia prediction study was performed using the time-series artificial neural network model, Long-short term memory (LSTM), after long-term monitoring of ammonia and environmental factors (ventilation rate (V), temperature (T), humidity (RH)) from a slurry finishing pig farm on mechanical ventilation system. The difference with the actual ammonia concentration was compared through prediction of the last three days of the entire breeding period. As a result of the analysis, the model which had a low correlation (ammonia concentration and humidity) was confirmed to have less error values than the models that did not. In addition, the combination of two or more input values [V, RH] and [T, V, RH] showed the lowest error value. In this study, the sustainability period of the model trained by multivariate input values was analyzed for about two days. In addition, [T, V, RH] showed the highest predictive performance with regard to the actual time of the occurrence of peak concentration compared to other models . These results can be useful in providing highly reliable information to livestock farmers regarding the management of concentrations through artificial neural network-based prediction models.

본 논문에서는 전산구조 해석 데이터를 기반으로 인공 신경망을 활용하여 헬리데크 구조물에 대한 손상 추정 기법을 제안하였다. 헬리데크를 구성하는 트러스와 서포트 부재들에 대해서 절점을 공유하는 부재들을 70개의 모델로 그룹화 하였으며, 최대 3가지 부재 그룹에 무작위로 손상을 부여하여 총 37,400개의 손상 시나리오를 생성하였다. 이들 각각에 대해서 구조 해석 프로그램을 통해 모드 해석을 수행하였으며, 전체 손상 시나리오를 사용 목적에 따라 학습, 유효성 검사, 그리고 검증 시나리오로 분리하였다. 헬리데크의 손상 및 비손상 상태의 동적 응답 특성에 대한 패턴 인식을 위해 PyTorch 프로그램을 활용하여 3개의 은닉층을 가지는 인공 신경망을 구성하였으며, 이에 대해서 다양한 손상 시나리오를 반복 학습함으로써 손실 함수를 최소로 하는 인공 신경망을 도출하였다. 최종적으로 총 400개의 검증 시나리오에 대해서 인공 신경망이 추정한 손상률과 실제 부여된 손상률을 비교하였으며, 그 결과 본 연구를 통해 얻어진 인공 신경망이 손상 부재의 위치와 손상 정도를 매우 높은 정확도로 예측하는 것을 확인하였다.

본 연구에서는 인공신경망을 이용해 건물 구조물의 가속도계 설치 위치 및 개수를 선정하는 방법을 제안한다. 인공신경망의 입력 층에는 층에 설치되는 가속도계로부터 얻는 가속도이력데이터가 입력되며, 출력층에는 구조물을 정의하는 각 층의 질량과 강성 값을 출력하도록 신경망을 구성한다. 가속도계의 설치 위치 및 개수를 선정하기 위해 여러 설치 시나리오를 가정하고 훈련을 통해 인공신경망을 구한다. 훈련에 사용되지 않은 예제를 이용해 예측 성능을 비교하였다. 센서 개수 및 위치에 따른 예측 성능을 비교하여 설치위치 및 개수를 선정한다. 6층과 10층 예제 적용을 통해 제안하는 방법을 검증하였다.