• Geomechanics and Engineering
• /
• 제31권3호
• /
• pp.265-279
• /
• 2022

Tunnel convergence prediction is essential for the safe construction and design of tunnels. This study proposes five machine learning models of deep neural network (DNN), K-nearest neighbors (KNN), Gaussian process regression (GPR), support vector regression (SVR), and decision trees (DT) to predict the convergence phenomenon during or shortly after the excavation of tunnels. In this respect, a database including 650 datasets (440 for training, 110 for validation, and 100 for test) was gathered from the previously constructed tunnels. In the database, 12 effective parameters on the tunnel convergence and a target of tunnel wall convergence were considered. Both 5-fold and hold-out cross validation methods were used to analyze the predicted outcomes in the ML models. Finally, the DNN method was proposed as the most robust model. Also, to assess each parameter's contribution to the prediction problem, the backward selection method was used. The results showed that the highest and lowest impact parameters for tunnel convergence are tunnel depth and tunnel width, respectively.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제17권9호
• /
• pp.2483-2504
• /
• 2023

Most existing low-light enhancement algorithms either use a large number of training parameters or lack generalization to real-world scenarios. This paper presents a novel lightweight and robust pixel-wise polynomial approximation-based deep network for low-light image enhancement. For mapping the low-light image to the enhanced image, pixel-wise higher-order polynomials are employed. A deep convolution network is used to estimate the coefficients of these higher-order polynomials. The proposed network uses multiple branches to estimate pixel values based on different receptive fields. With a smaller receptive field, the first branch enhanced local features, the second and third branches focused on medium-level features, and the last branch enhanced global features. The low-light image is downsampled by the factor of 2b-1 (b is the branch number) and fed as input to each branch. After combining the outputs of each branch, the final enhanced image is obtained. A comprehensive evaluation of our proposed network on six publicly available no-reference test datasets shows that it outperforms state-of-the-art methods on both quantitative and qualitative measures.

• Earthquakes and Structures
• /
• 제25권6호
• /
• pp.469-479
• /
• 2023

A probabilistic seismic damage analysis is an essential procedure to identify seismically vulnerable structures, prioritize the seismic retrofit, and ultimately minimize the overall seismic risk. To assess the seismic risk of multiple structures within a region, a large number of nonlinear time-history structural analyses must be conducted and studied. As a result, each assessment requires high computing resources. To overcome this limitation, we explore a deep learning-based metamodel to enable the prediction of the mean and the standard deviation of the seismic damage distribution of track-on steel-plate girder railway bridges in Korea considering the geometric variation. For machine learning training, nonlinear dynamic time-history analyses are performed to generate 800 high-fidelity datasets on the seismic response. Through intensive trial and error, the study is concentrated on developing an optimal machine learning architecture with the pre-identified variables of the physical configuration of the bridge. Additionally, the prediction performance of the proposed method is compared with a previous, well-defined, response surface model. Finally, the statistical testing results indicate that the overall performance of the deep-learning model is improved compared to the response surface model, as its errors are reduced by as much as 61%. In conclusion, the model proposed in this study can be effectively deployed for the seismic fragility and risk assessment of a region with a large number of structures.

• 대한임베디드공학회논문지
• /
• 제18권6호
• /
• pp.277-283
• /
• 2023

Vital signals provide essential information regarding the health status of individuals, thereby contributing to health management and medical research. Present monitoring methods, such as ECGs (Electrocardiograms) and smartwatches, demand proximity and fixed postures, which limit their applicability. To address this, Non-contact vital signal measurement methods, such as CW (Continuous-Wave) radar, have emerged as a solution. However, unwanted signal components and a stepwise processing approach lead to errors and limitations in heart rate detection. To overcome these issues, this study introduces an integrated neural network approach that combines noise removal, demodulation, and dominant-frequency detection into a unified process. The neural network employed for signal processing in this research adopts a MLP (Multi-Layer Perceptron) architecture, which analyzes the in-phase and quadrature signals collected within a specified time window, using two distinct input layers. The training of the neural network utilizes CW radar signals and reference heart rates obtained from the ECG. In the experimental evaluation, networks trained on different datasets were compared, and their performance was assessed based on loss and frequency accuracy. The proposed methodology exhibits substantial potential for achieving precise vital signals through non-contact measurements, effectively mitigating the limitations of existing methodologies.

• 한국공간구조학회논문집
• /
• 제23권4호
• /
• pp.81-88
• /
• 2023

Numerous factors contribute to the deterioration of reinforced concrete structures. Elevated temperatures significantly alter the composition of the concrete ingredients, consequently diminishing the concrete's strength properties. With the escalation of global CO2 levels, the carbonation of concrete structures has emerged as a critical challenge, substantially affecting concrete durability research. Assessing and predicting concrete degradation due to thermal effects and carbonation are crucial yet intricate tasks. To address this, multiple prediction models for concrete carbonation and compressive strength under thermal impact have been developed. This study employs seven machine learning algorithms-specifically, multiple linear regression, decision trees, random forest, support vector machines, k-nearest neighbors, artificial neural networks, and extreme gradient boosting algorithms-to formulate predictive models for concrete carbonation and thermal impact. Two distinct datasets, derived from reported experimental studies, were utilized for training these predictive models. Performance evaluation relied on metrics like root mean square error, mean square error, mean absolute error, and coefficient of determination. The optimization of hyperparameters was achieved through k-fold cross-validation and grid search techniques. The analytical outcomes demonstrate that neural networks and extreme gradient boosting algorithms outshine the remaining five machine learning approaches, showcasing outstanding predictive performance for concrete carbonation and thermal effect modeling.

• Computers and Concrete
• /
• 제33권2호
• /
• pp.137-145
• /
• 2024

The technique of experimentally determining concrete's compressive strength for a given mix design is time-consuming and difficult. The goal of the current work is to propose a best working predictive model based on different machine learning algorithms such as Gradient Boosting Machine (GBM), Stacked Ensemble (SE), Distributed Random Forest (DRF), Extremely Randomized Trees (XRT), Generalized Linear Model (GLM), and Deep Learning (DL) that can forecast the compressive strength of ternary geopolymer concrete mix without carrying out any experimental procedure. A geopolymer mix uses supplementary cementitious materials obtained as industrial by-products instead of cement. The input variables used for assessing the best machine learning algorithm not only include individual ingredient quantities, but molarity of the alkali activator and age of testing as well. Myriad statistical parameters used to measure the effectiveness of the models in forecasting the compressive strength of ternary geopolymer concrete mix, it has been found that GBM performs better than all other algorithms. A sensitivity analysis carried out towards the end of the study suggests that GBM model predicts results close to the experimental conditions with an accuracy between 95.6 % to 98.2 % for testing and training datasets.

• 한국해양공학회지
• /
• 제38권3호
• /
• pp.103-114
• /
• 2024

Remote sensing wave observation data are crucial when analyzing ocean waves, the main external force of coastal disasters. Nevertheless, it has limitations in accuracy when used in low-wind environments. Therefore, this study collected the raw data from MIROS Wave and Current Radar (MWR) and wave radar at the Ieodo Ocean Research Station (IORS) and applied the optimal filter by combining filters provided by MIROS software. The data were validated by a comparison with South Jeju ocean buoy data. The results showed it maintained accuracy for significant wave height, but errors were observed in significant wave periods and extreme waves. Hence, this study used an artificial neural network (ANN) to improve these errors. The ANN was generalized by separating the data into training and test datasets through stratified sampling, and the optimal model structure was derived by adjusting the hyperparameters. The application of ANN effectively improved the accuracy in significant wave periods and high wave conditions. Consequently, this study reproduced past wave data by enhancing the reliability of the MWR, contributing to understanding wave generation and propagation in storm conditions, and improving the accuracy of wave prediction. On the other hand, errors persisted under high wave conditions because of wave shadow effects, necessitating more data collection and future research.

• 한국군사과학기술학회지
• /
• 제27권2호
• /
• pp.147-155
• /
• 2024

Based on the recently developed deep learning technology, many studies have been conducted on deep learning networks that simultaneously detect and classify targets of interest in synthetic aperture radar(SAR) images. Although numerous research results have been derived mainly with the open SAR ship datasets, there is a lack of work carried out on the deep learning network aimed at detecting and classifying SAR ground targets and trained with the synthetic dataset generated from electromagnetic scattering simulations. In this respect, this paper presents the deep learning network trained with the synthetic dataset and applies it to detecting and classifying real SAR ground targets. With experiment results, this paper also analyzes the network performance according to the composition ratio between the real measured data and the synthetic data involved in network training. Finally, the summary and limitations are discussed to give information on the future research direction.

• ETRI Journal
• /
• 제46권1호
• /
• pp.127-136
• /
• 2024

The Korean language has written (formal) and spoken (phonetic) forms that differ in their application, which can lead to confusion, especially when dealing with numbers and embedded Western words and phrases. This fact makes it difficult to automate Korean speech recognition models due to the need for a complete transcription training dataset. Because such datasets are frequently constructed using broadcast audio and their accompanying transcriptions, they do not follow a discrete rule-based matching pattern. Furthermore, these mismatches are exacerbated over time due to changing tacit policies. To mitigate this problem, we introduce a data-driven Korean spoken-to-written transcription conversion technique that enhances the automatic conversion of numbers and Western phrases to improve automatic translation model performance.

• 지능정보연구
• /
• 제28권2호
• /
• pp.191-206
• /
• 2022

대량의 말뭉치를 비지도 방식으로 학습하여 자연어 지식을 획득할 수 있는 사전학습 언어모델(Pre-trained Language Model)은 최근 자연어 처리 모델 개발에 있어 매우 일반적인 요소이다. 하지만, 여타 기계학습 방식의 성격과 동일하게 사전학습 언어모델 또한 학습 단계에 사용된 자연어 말뭉치의 특성으로부터 영향을 받으며, 이후 사전학습 언어모델이 실제 활용되는 응용단계 태스크(Downstream task)가 적용되는 도메인에 따라 최종 모델 성능에서 큰 차이를 보인다. 이와 같은 이유로, 법률, 의료 등 다양한 분야에서 사전학습 언어모델을 최적화된 방식으로 활용하기 위해 각 도메인에 특화된 사전학습 언어모델을 학습시킬 수 있는 방법론에 관한 연구가 매우 중요한 방향으로 대두되고 있다. 본 연구에서는 금융(Finance) 도메인에서 다양한 자연어 처리 기반 서비스 개발에 활용될 수 있는 금융 특화 사전학습 언어모델의 학습 과정 및 그 응용 방식에 대해 논한다. 금융 도메인 지식을 보유한 언어모델의 사전학습을 위해 경제 뉴스, 금융 상품 설명서 등으로 구성된 금융 특화 말뭉치가 사용되었으며, 학습된 언어 모델의 금융 지식을 정량적으로 평가하기 위해 토픽 분류, 감성 분류, 질의 응답의 세 종류 자연어 처리 데이터셋에서의 모델 성능을 측정하였다. 금융 도메인 말뭉치를 기반으로 사전 학습된 KB-BERT는 KoELECTRA, KLUE-RoBERTa 등 State-of-the-art 한국어 사전학습 언어 모델과 비교하여 일반적인 언어 지식을 요구하는 범용 벤치마크 데이터셋에서 견줄 만한 성능을 보였으며, 문제 해결에 있어 금융 관련 지식을 요구하는 금융 특화 데이터셋에서는 비교대상 모델을 뛰어넘는 성능을 보였다.