Background: Real-time ergonomic risk assessment in manufacturing environments is challenged by severe class imbalance in high-risk postures and the need for deployment-efficient models. Conventional oversampling techniques may violate biomechanical constraints, limiting their suitability for human motion data. Objectives: This study aimed to compare multiple machine learning models for real-time ergonomic risk assessment while addressing data imbalance using biomechanically appropriate learning strategies and evaluating both predictive performance and deployment efficiency. Design: Comparative study. Methods: A large-scale workplace safety dataset comprising image-based skeletal keypoints was analyzed. To mitigate class imbalance without generating biomechanically implausible samples, cost-sensitive learning and focal loss were employed instead of synthetic oversampling. Subject-wise data splitting was applied to prevent data leakage. Five model families, including Random Forest, convolutional neural networks, and a lightweight graphbased network, were evaluated using accuracy, F1-score, area under the receiver operating characteristic curve (AUC), and high-risk recall. Statistical significance was assessed using bootstrap confidence intervals and McNemar and DeLong tests. Results: The lightweight graph-based model demonstrated competitive classification performance while maintaining reduced computational complexity. Although none of the models achieved the predefined high-risk recall threshold, statistically significant performance differences were observed across model families. Conclusion: The findings suggest that biomechanically informed imbalance handling improves methodological validity in ergonomic risk assessment. While deployment feasibility appears promising, further empirical validation on edge hardware is required.

본 연구는 준거집단(취직자)들의 활동 데이터 뱅크를 생성하여 예비 취업자(고등교육기관의 체 육계열 전공자)들이 현재까지 활동했던 데이터를 데이터마이닝 기반 추천 알고리즘을 적용해 예비 취업자 들에게 가장 적합한 직업군을 추천해주는 스포츠 일자리 추천모형을 개발하고 검증하는 것이다. 따라서 평 가지표를 구성하고, 준거집단을 대상으로 인터뷰 및 조사를 통해 데이터 뱅크를 생성했다. 또한 비확률 표 본추출법 중 할당표본 추출법과 눈덩이표본 추출법을 적용해 예비 취업자 조사를 실시했으며, 총 921명의 자료를 통해 스포츠 일자리 추천모형 개발과 유사도를 통해 모형을 검증했다. 즉, 본 결과는 다음과 같다. 첫째, 준거집단과 예비 취업자의 평가지표를 구성했다. 둘째, 준거집단의 데이터 뱅크를 생성했다. 셋째, 스 포츠 전공 청년들을 위한 일자리 추천모형을 개발하고, 유사도를 통해 모형을 검증했다.

This study develops a post-value for money (post-VfM) evaluation model for public–private partnership (PPP) road projects in Korea. Following the abolition of the minimum revenue guarantee system, the demand risk was transferred to the private sector, thus necessitating an unbiased and data-driven assessment under the new adjusted build-transfer-operate (BTO-a) framework. The proposed model extends the existing ex-ante VfM analysis by incorporating actual operational data and estimating government payments for both public-sector comparator and private finance initiative alternatives on a lifecycle cost basis. Using an actual BTO project restructured as BTO-a, the simulation shows that the post-VfM ratio increases from 23.5% to 37.9%, thus confirming fiscal efficiency and balanced risk sharing. This model enables feedback between planning and operation, supports transparent policy evaluation, and provides a foundation for sustainable PPP governance in future infrastructure projects.

최근 고도화된 딥러닝 모형을 이용하여 하천 수질에 영향을 줄 수 있는 과도한 조류(algae) 발생을 예측하는 연구에 대한 관심이 지속되고 있으며, 모형의 구축에 사용되는 현장 측정 자료의 특성상 다양한 이상치를 포함할 수 있어 데이터의 이상치 관리 필요성이 높아지고 있다. 본 연구에서는 현장 자료의 이상치가 딥러닝 모형의 성능에 미치는 영향을 분석하기 위해 딥러닝 Long Short-Term Memory(LSTM) 모형을 이용하여 하천 조류 발생을 정량적으로 평가하는 지표인 클로로필-a를 예측하는 모형을 구축하였으며, 10%의 이상치를 포함한 자료와 이상치가 포함되지 않은 원본 자료로 학습된 모형의 성능을 비교하였다. 또한 딥러닝 기반 이상치 탐지 알고리즘인 Autoencoder(AE)를 이용하여 이상치를 제거한 후 모형의 성능에 미치는 영향을 비교하였다. 분석 결과 이상치를 포함하지 않은 자료로 학습된 Base 모형과 10%의 이상치를 포함한 자료로 학습된 모형의 Nash-Sutcliffe efficiency(NSE)가 각각 0.882 및 0.858로 나타나 이상치가 모형의 성능을 저하시킬 수 있음을 확인하였다. 한편 AE를 이용하여 이상치를 다양한 비율로(5–20%) 제거한 자료로 학습된 모형의 성능을 분석한 결과 NSE가 0.883–0.896으로 이상치의 제거에 따라 모형의 성능이 Base 모형과 유사한 수준으로 개선되는 것으로 나타났다. 본 연구에서는 이상치가 딥러닝 모형에 미치는 영향을 분석하고 이상치 탐지 모형의 활용에 따른 조류 발생 예측 딥러닝 모형의 성능 향상이 가능함을 확인하였다.

과도한 조류 발생은 수생태계 교란과 수질 악화를 초래하는 대표적인 환경 문제로, 효과적인 관리와 대응을 위해 정확한 예측이 필요하다. 우리나라는 사계절의 기후 특성이 뚜렷하며, 수온이 상승하는 하절기에 조류 발생이 집중되는 경향을 보인다. 이에 따라 실시간 모니터링 자료는 대부분 저농도 상태가 유지되어 데이터 불균형 문제가 발생한다. 본 연구에서는 chlorophyll-a 농도를 기준으로 하천 현장의 조류 발생 수준을 Class 1 (Chl-a ≤ 10 ㎍/L), Class 2 (10 < Chl-a ≤ 50 ㎍/L), Class 3 (Chl-a > 50 ㎍/L)와 같이 3개의 class로 구분하고, 대표적인 앙상블 머신러닝 모형인 extreme gradient boosting (XGB) 알고리즘을 이용하여 조류 발생 수준을 예측하는 분류 모형을 구축하였다. 데이터 불균형 해소를 위해 생성형 인공지능 기반 알고리즘인 conditional generative adversarial network (CGAN)과 전통적인 데이터 보강 알고리즘인 synthetic minority over-sampling technique (SMOTE), 그리고 딥러닝 기반 기법인 autoencoder (AE)를 활용한 3가지 데이터 보강 알고리즘을 활용하여 데이터의 불균형을 개선한 자료를 생성하고 이를 XGB 모형에 적용하여 성능 변화를 비교하였다. 분석 결과 macro average 기준으로 원본 데이터를 사용한 모형의 recall은 0.606이었으나 SMOTE, AE 및 CGAN의 recall은 각각 0.666, 0.682, 0.720으로 크게 개선되었고, F1 score도 데이터 불균형 해소를 통해 약 7–13%의 성능이 향상되는 등 전체적으로 데이터 불균형 해소로 모형의 성능이 향상되었으며 CGAN이 가장 우수한 성능 개선 효과를 보이는 것으로 나타냈다. 본 연구의 결과를 통해 데이터 불균형 해소를 통한 머신러닝 모형 성능 개선 가능성을 확인하였다.

Machine learning (ML) techniques have been increasingly applied to the field of structural engineering for the prediction of complex dynamic responses of safety-critical infrastructures such as nuclear power plant (NPP) structures. However, the development of ML-based prediction models requires a large amount of training data, which is computationally expensive to generate using traditional finite element method (FEM) time history analysis, especially for aging NPP structures. To address this issue, this study investigates the effectiveness of synthetic data generated using Conditional Tabular GAN (CTGAN) in training ML models for seismic response prediction of an NPP auxiliary building. To overcome the high computational cost of data generation, synthetic tabular data was generated using CTGAN and its quality was evaluated in terms of distribution similarity (Shape) and feature relationship consistency (Pair Trends) with the original FEM data. Four training datasets with varying proportions of synthetic data were constructed and used to train neural network models. The predictive accuracy of the models was assessed using a separate test set composed only of original FEM data. The results showed that models trained with up to 50% synthetic data maintained high prediction accuracy, comparable to those trained with only original data. These findings indicate that CTGAN-generated data can effectively supplement training datasets and reduce the computational burden in ML model development for seismic response prediction of NPP structures.

Republic of Korea is building a multi-layered missile defense system against North Korea’s growing ballistic missile threat. To maximize the intercept performance of a multi-layered missile defense system, it is important to develop an efficient engagement plan that considers the interceptable time/space of each interceptor system for ballistic missiles. To do so, it is necessary to predict the flight trajectory of the ballistic missile, which must be done within a short time considering the short battlefield environment and the speed of the ballistic missile. This study presents a model for rapid trajectory prediction of ballistic missiles using the kinetic characteristics of each flight phase(thrust phase, midcourse phase, and re-entry phase) of ballistic missiles, a method for estimating kinetic information from ballistic missile observation data(time and position), and a mathematical analysis of the equations of motion of ballistic missiles.

This study is a preliminary investigation into a method for updating analytical models using actual vibration measurement data to improve the reliability of the seismic performance evaluations. The research was conducted on 26 models with various parameters, aiming to develop an optimal analytical model that closely matches the natural frequencies of the actual building. By identifying the dynamic characteristics of the target building through vibration measurements taken just before the demolition of the structure, the natural frequency analysis results of the analytical models were compared to the measured data. Based on this comparison, an optimized method for adjusting the parameters of the analytical models was derived. Throughout the analysis, various parameters were adjusted, and the eigenvalue analysis results were corrected by comparing them with vibration measurements. Among the comparative analytical models, the model with the lowest error rate was selected. The results showed that, in all cases, the analytical model with a concrete compressive strength of 16 MPa (based on actual measurements), pin boundary conditions, and an idealized strip footing cross-section had the closest match to the actual building's natural frequencies, with an average error of less than 8%.

본 연구는 한국 기상대 데이터를 활용하여 콘크리트 포장의 깊이별 온도를 예측하는 ANN(Artificial Neural Network) 모델을 개발하는 것을 목표로 한다. 기존의 열평형 방정식 기반 모델은 특정 지역의 기상 데이터를 필요로 하기 때문에 일반적인 적용이 어렵다는 한계를 가지고 있다. 이에 본 연구에서는 ANN을 활용하여 기상대 데이터를 기반으로 범용적 인 온도 예측 모델을 개발하고자 한다. 이를 통해 다양한 지역 및 환경 조건에서도 적용 가능한 모델을 구축하는 것이 목적이다. 본 연구에서는 2017년 1월 1일부터 2018년 12월 31일까지의 1시간 단위 기상 및 온도 데이터를 활용하며, 0.05m, 0.15m, 0.25m, 0.35m, 0.45m 깊이별 온도 데이터를 학습 데이터로 사용한다. 입력 변수로는 기온, 풍속, 강수량, 습도, 일 조량, 일사량, 적설량, 적운량, 지면온도를 포함한다. 이러한 다양한 기상 데이터를 활용하여 신경망 모델을 학습하고, 기 존 방식보다 높은 정확도를 확보하는 것이 연구의 핵심 목표이다. 기존 ANN 구조인 O = WI + B에서 확장된 O = W(I + (WI + B)) + B 형태의 비선형 구조를 적용하여 기존 모델이 가지는 비선형 관계 반영의 한계를 극복하고자 한다. 또한, 선형 다중 은닉층 모델과 비선형 다중 은닉층 모델을 각각 개발하여 성능을 비교하고, 비선형 모델의 필요성과 일반화 능력을 평가할 예정이다. 최종적으로 두 모델의 성능을 평균 제곱 오차 및 평균 절대 오차 등과 같은 평가 지표들을 이용하여 비교 분석하고, 가장 적합한 모델을 도출하고자 한다.

The purpose of this study is to develop a timely fall detection system aimed at improving elderly care, reducing injury risks, and promoting greater independence among older adults. Falls are a leading cause of severe complications, long-term disabilities, and even mortality in the aging population, making their detection and prevention a crucial area of public health focus. This research introduces an innovative fall detection approach by leveraging Mediapipe, a state-of-the-art computer vision tool designed for human posture tracking. By analyzing the velocity of keypoints derived from human movement data, the system is able to detect abrupt changes in motion patterns, which are indicative of potential falls. To enhance the accuracy and robustness of fall detection, this system integrates an LSTM (Long Short-Term Memory) model specifically optimized for time-series data analysis. LSTM's ability to capture critical temporal shifts in movement patterns ensures the system's reliability in distinguishing falls from other types of motion. The combination of Mediapipe and LSTM provides a highly accurate and robust monitoring system with a significantly reduced false-positive rate, making it suitable for real-world elderly care environments. Experimental results demonstrated the efficacy of the proposed system, achieving an F1 score of 0.934, with a precision of 0.935 and a recall of 0.932. These findings highlight the system's capability to handle complex motion data effectively while maintaining high accuracy and reliability. The proposed method represents a technological advancement in fall detection systems, with promising potential for implementation in elderly monitoring systems. By improving safety and quality of life for older adults, this research contributes meaningfully to advancements in elderly care technology.

The purpose of this study is to produce virtual models of women aged in their 60s and to implement the virtual clothing with jackets. We referred to 3D images of standard and obese body types from the 8th Size Korea and attempted to create avatars based on their images through the various trials. Final virtual models were made to reflect the appearance of women in their 60s. For the standard body type, a 3D image with average body measurements was selected. Based on numerous trials aimed at turning her image into an avatar, the auto-converted avatar on CLO 3D was slimmer than the woman in the original image, and hence it was not suitable for the virtual model. After blending, we converted the image into an uneditable avatar for which only the joint points could be moved, thereby creating an avatar that was identical to the original image. We also selected an image of an obese woman with a “beer bottle” body shape from the 8th Size Korea. We created an avatar that resembled her shape by also converting it into an uneditable avatar for which only joint points could be moved. To use these avatars in virtual clothing, we removed masks of avatars and made faces, hair styles, and skin tones representing women in their 60s. The moderately-sized classic jackets were smooth on both virtual models and fitted satisfactorily. This study demonstrated the applicability of virtual model production of various body types or ages in special clothing studies.

In recent automated manufacturing systems, compressed air-based pneumatic cylinders have been widely used for basic perpetration including picking up and moving a target object. They are relatively categorized as small machines, but many linear or rotary cylinders play an important role in discrete manufacturing systems. Therefore, sudden operation stop or interruption due to a fault occurrence in pneumatic cylinders leads to a decrease in repair costs and production and even threatens the safety of workers. In this regard, this study proposed a fault detection technique by developing a time-variant deep learning model from multivariate sensor data analysis for estimating a current health state as four levels. In addition, it aims to establish a real-time fault detection system that allows workers to immediately identify and manage the cylinder’s status in either an actual shop floor or a remote management situation. To validate and verify the performance of the proposed system, we collected multivariate sensor signals from a rotary cylinder and it was successful in detecting the health state of the pneumatic cylinder with four severity levels. Furthermore, the optimal sensor location and signal type were analyzed through statistical inferences.

In the military, ammunition and explosives stored and managed can cause serious damage if mishandled, thus securing safety through the utilization of ammunition reliability data is necessary. In this study, exploratory data analysis of ammunition inspection records data is conducted to extract reliability information of stored ammunition and to predict the ammunition condition code, which represents the lifespan information of the ammunition. This study consists of three stages: ammunition inspection record data collection and preprocessing, exploratory data analysis, and classification of ammunition condition codes. For the classification of ammunition condition codes, five models based on boosting algorithms are employed (AdaBoost, GBM, XGBoost, LightGBM, CatBoost). The most superior model is selected based on the performance metrics of the model, including Accuracy, Precision, Recall, and F1-score. The ammunition in this study was primarily produced from the 1980s to the 1990s, with a trend of increased inspection volume in the early stages of production and around 30 years after production. Pre-issue inspections (PII) were predominantly conducted, and there was a tendency for the grade of ammunition condition codes to decrease as the storage period increased. The classification of ammunition condition codes showed that the CatBoost model exhibited the most superior performance, with an Accuracy of 93% and an F1-score of 93%. This study emphasizes the safety and reliability of ammunition and proposes a model for classifying ammunition condition codes by analyzing ammunition inspection record data. This model can serve as a tool to assist ammunition inspectors and is expected to enhance not only the safety of ammunition but also the efficiency of ammunition storage management.

빠르게 발전하는 이미지 인식 기술에도 불구하고 표 형식의 문서와 수기로 작성된 문서를 완벽하게 디지털화하기에는 아직 어려움이 따른다. 본 연구는 표 형식의 수기 문서인 선박 항해일지를 작성하는 데에 사용되는 규칙을 이용하여 보정 작업을 수행함으로 써 OCR 결과물의 정확도를 향상시키고자 한다. 이를 통해 OCR 프로그램을 통하여 추출된 항해일지 데이터의 정확성과 신뢰성을 높일 것 으로 기대된다. 본 연구는 목포해양대학교 실습선 새누리호의 2023년에 항해한 57일간의 항해일지 데이터를 대상으로 OCR 프로그램 인 식 후 발생한 오류를 보정하여 그 정확도를 개선하고자 하였다. 이 모델은 항해일지 기재 시 고려되는 몇 가지 규칙을 활용하여 오류를 식별한 후, 식별된 오류를 보정하는 방식으로 구성하였다. 모델을 활용하여 오류를 보정 후, 그 효과를 평가하고자 보정 전과 후의 데이터 를 항차별로 구분한 후, 같은 항차의 같은 변수끼리 비교하였다. 본 모델을 활용하여 실제 셀 오류율은 약 11.8% 중 약 10.6%의 오류를 식 별하였고, 123개의 오류 중 56개를 개선하였다. 본 연구는 항해일지 중 항해정보를 기입하는 Dist.Run부터 Stand Course까지의 정보만을 대 상으로 수행하였다는 한계점이 있으므로, 추후 항해정보 뿐만 아니라 기상정보 등 항해일지의 더 많은 정보를 보정하기 위한 연구를 진 행할 예정이다.