Fault detection in electromechanical systems plays a significant role in product quality and manufacturing efficiency during the transition to smart manufacturing. Because collecting a sufficient number of datasets under faulty conditions of the system is challenging in practical industrial sites, unsupervised fault detection methods are mainly used. Although fault datasets accumulate during machine operation, it is not straightforward to utilize the information it contains for fault detection after the deep learning model has been trained in an unsupervised manner. However, the information in fault datasets is expected to significantly contribute to fault detection. In this regard, this study aims to validate the effectiveness of the transition from unsupervised to supervised learning as fault datasets gradually accumulate through continuous machine operation. We also focus on experimentally analyzing how changes in the learning paradigm of the deep learning model and the output representation affect fault detection performance. The results demonstrate that, with a small number of fault datasets, a supervised model with continuous outputs as a regression problem showed better fault detection performance than the original model with one-hot encoded outputs (as a classification problem).

Reinforcement learning (RL) is successfully applied to various engineering fields. RL is generally used for structural control cases to develop the control algorithms. On the other hand, a machine learning (ML) is adopted in various research to make automated structural design model for reinforced concrete (RC) beam members. In this case, ML models are developed to produce results that are as similar to those of training data as possible. The ML model developed in this way is difficult to produce better results than the training data. However, in reinforcement learning, an agent learns to make decisions by interacting with an environment. Therefore, the RL agent can find better design solution than the training data. In the structural design process (environment), the action of RL agent represent design variables of RC beam. Because the number of design variables of RC beam section is many, multi-agent DQN (Deep Q-Network) was used in this study to effectively find the optimal design solution. Among various versions of DQN, Double Q-Learning (DDQN) that not only improves accuracy in estimating the action-values but also improves the policy learned was used in this study. American Concrete Institute (318) was selected as the design codes for optimal structural design of RC beam and it was used to train the RL model without any hand-labeled dataset. Six agents of DDQN provides actions for beam with, beam depth, bottom rebar size, number of bottom rebar, top rebar size, and shear stirrup size, respectively. Six agents of DDQN were trained for 5,000 episodes and the performance of the multi-agent of DDQN was evaluated with 100 test design cases that is not used for training. Based on this study, it can be seen that the multi-agent RL algorithm can provide successfully structural design results of doubly reinforced beam.

본 연구는 다시마 양식을 위한 통합 자동화 시스템을 개발하고, 이를 통해 생산성, 비용 효율성, 환경적 지속 가능성을 모두 개선하는 데 중점을 두고 있다. 기존의 노동 집약적 수확 방식과 넓은 공간을 필요로 하는 수평 건조 방식은 비효율적이며, 환경적 부작 용을 초래했다. 이에 본 연구는 자동화된 수확 시스템, 해상-육상 연계 운송 시스템, 그리고 수직 건조 시스템을 통합적으로 개발하여 양 식업의 생산성을 극대화하고 자원 사용을 최적화하였다. 자동화된 수확 시스템은 작업 속도를 약 35% 향상시켰으며, 작업의 일관성을 유지하여 품질 오차율을 2% 이하로 줄이는 성과를 보였다. 해상-육상 연계 운송 시스템은 모듈형 컨테이너를 활용하여 운송 중 손상률 을 기존 15%에서 5%로 감소시켰고, 운송 시간을 평균 6시간에서 4시간으로 단축하였다. 또한, 수직 건조시스템은 고밀도 적재와 자연 대류 방식을 도입하여 건조 시간을 기존 48시간에서 28시간으로 40% 단축하였으며, 에너지 소비를 25% 감소시켰다. 이러한 시스템은 데이터 기반으로 설계 및 검증되었으며, 통합적으로 양식업의 경제성 향상과 환경적 부담 감소를 동시에 실현하였다. 본 연구의 결과는 다른 해조류 양식에도 적용 가능하며 지속 가능한 해양 자원 관리에 기여할 것으로 기대된다.

본 논문에서는 2자유도 매니퓰레이터(manipulator)가 탑재된 지상형 이동 로봇을 활용한 균열 지도 작성 기법을 소개한다. 로봇의 앞·측면에 각각 스테레오 비전 센서를 설치하였으며, 앞면에 설치된 센서의 포인트 클라우드 정보를 이용하여 로봇의 위치를 인식하 고 지도를 작성하며, 측면에 설치된 센서의 영상 정보를 바탕으로 벽면 내 균열을 검출한다. 이때, 두 센서의 좌표계를 좌표 변환식을 통해 통일하여 정합 및 검출된 균열 정보를 생성된 지도에 실시간으로 표기하고, 손상의 정보가 기록 및 관리될 수 있도록 하였다. 2자 유도 움직임이 가능한 매니퓰레이터 장치를 이동로봇에 탑재하고 사각지역의 제한 없이 로봇의 진행 방향을 벗어난 균열을 촬영할 수 있도록 하였다. 촬영된 영상 내 딥러닝 기법을 적용하여 균열을 검출하고, 해당 균열이 촬영된 영상 내 일부만 존재한다고 판단하 는 경우 매니퓰레이터를 동작하여 남은 균열의 위치를 추정 및 추가 촬영, 스티칭할 수 있도록 하였다. 본 시스템의 성능 확인을 위하 여 실내 환경에서 실험을 진행하였으며, IoU기반 검출율 0.6 이상의 정확도로 실시간 균열 정보를 구축된 지도 위에 작성하는 것을 확 인하였다.

This study evaluated the safety impact of automated traffic enforcement cameras targeting tailgating behavior at signalized intersections by comparing traffic conditions shortly after installation and one year later. The Kukkiwon intersection in Gangnam-gu, Seoul, South Korea was selected as the study site. Individual vehicle speeds, accelerations, and subsequent distances were extracted from video data using YOLOv8 and ByteTrack, which are advanced deep learning-based object detection and tracking algorithms. Surrogate safety measures (SSM), such as time to collision (TTC), modified time to collision (MTTC), and proportion of stopping distance (PSD), were calculated to assess changes in traffic safety. Every SSM indicated an improvement one year after the installation of enforcement cameras, suggesting a reduction in collision risks. In particular, the PSD indicator showed a notable improvement, reflecting a better maintenance of safe following distances. These results highlight the effectiveness of automated enforcement in improving intersection safety and suggest its scalability to other intersections with similar tail-gating issues. Future research should explore the long-term and multisite effects using diverse intersection types and behavioral indicators.

Korea has many test beds where various mobility services are provided by automated vehicles. The test beds are operated in their operational design domain (ODD). However, disengagement frequently occurs, even in the ODDs of automated vehicles. In particular, human drivers have to take control of the automated vehicles at SAE Level 3 whenever the vehicles cannot drive by themselves because of an emergency or unknown factors. This study analyzed the driving safety of right turning at signalized intersections where automated vehicles face selfdriving issues because of potential conflicts with other vehicles, crossing pedestrians, and geometric factors. To conduct this analysis, we categorized right-turning intersections into two types with right-turning lanes and channelization islands and divided them into three sections, with a total of six sections. Subsequently, the six sections were compared with each other by disengagements of the automated vehicles as the key index to investigate their self-driving safety. Their significant differences indicate that ODD-related variables must be considered when designing and updating target test beds for automated vehicles.

Background: The growing need for objective and accurate evaluation in Taekwondo poomsae competitions has highlighted the limitations of subjective human judgment. Objectives: This study aims to develop an automated scoring framework using camera-based pose estimation and advanced neural networks to improve the consistency and accuracy of poomsae evaluation. Design: Comparative analysis of neural network architectures on a large-scale dataset of poomsae movements. Methods: A dataset of 902,306 labeled frames, captured from 48 participants performing 62 distinct movements using synchronized multi-view cameras, was analyzed. Five neural networks (HNN, 1D CNN, GCN, MLP, SANN) were implemented and evaluated using accuracy, precision, recall, and F1-score. Results: The HNN demonstrated superior performance with an F1-score of 0.78 in classifying Taekwondo poomsae postures. The 1D CNN followed with an F1-score of 0.76, while GCN, MLP, and SANN achieved F1-scores of 0.74, 0.70, and 0.66, respectively. The HNN's hierarchical feature extraction approach proved effective in capturing the complex spatial and temporal patterns inherent in poomsae movements. Conclusion: Hierarchical Neural Networks outperform other architectures in poomsae classification, establishing a foundation for objective and scalable scoring systems in competitive settings.

Automated structural design methods for reinforced concrete (RC) beam members have been widely studied with various techniques to date. Recently, artificial intelligence has been actively applied to various engineering fields. In this study, machine learning (ML) is adopted to make automated structural design model for RC beam members. Among various machine learning methods, a supervised learning was selected. When a supervised learning is applied to development of ML-based prediction model, datasets for training and test are required. Therefore, the datasets for rectangular and t-shaped RC beams was constructed by commercial structural design software of MIDAS. Five supervised learning algorithms, such as Decision Tree (DT), Random Forest (RF), K-Nearest Neighbor (KNN), Artificial Neural Networks (ANN), eXtreme Gradient Boosting (XGBoost) were used to develop the automated structural design model. Design moment (Mu), design shear force (Vu), beam length, uniform load (wu) were used for inputs of structural design model. Width and height of the designed section, diameter of top and bottom bars, number of top and bottom bars, diameter of stirrup bar were selected for outputs of structural design model. Performance evaluation of the developed structural design models was conducted using metrics sush as root mean square error (RMSE), mean square error (MSE), mean absolute error (MAE), and coefficient of determination (R2). This study presented that random forest provides the best structural design results for both rectangular and t-shaped RC beams.

In general, optimized pavement thickness design abilities and reliable construction procedures have been considered being crucial element for expressway in South Korea till millenium. However, after 2005, a proper management efforts on existing expressway became recognized as an important factor after 2,005. One of good example is rising attention of HPMS(Highway Pavement Management System). In HPMS, the crucial component is: surveying the existing expressway surface condition with reasonable, reliable and efficient procedure. Becasue of this reason, application of various advanced and sophisticated technologies on HPMS area were considered since 2010. During this time, many advanced technologies such as AI(Artificial Intelligence) techniques and corresponding physical equipment were considered to be applied. Through application of AI technologies in HPMS business area, two major outcomes can be achieved: first one is an automated pavement surface monitoring work system for maximized efficiency and second thing is saving current HPMS management budget through faster and more reasonable surveying results. In this paper, the feasibility of AI technology on actual pavement surface monitoring and analysis procedure was considered. As a result, AI based pavement surface monitoring and analysis approach succesfully provided reasonable results compared to the conventional human effort analysis approach. This findings provide a promising signal that more AI based technologies can successfully applied in HPMS business area in the next future. Morevoer, the achievement of automated HPMS can also be possible in the near future.

Background: Automated classification systems using Artificial Intelligence (AI) and Machine Learning (ML) can enhance accuracy and efficiency in diagnosing pet skin diseases within veterinary medicine. Objectives: This study created a system that classifies pet skin diseases by evaluating multiple ML models to determine which method is most effective. Design: Comparative experimental study. Methods: Pet skin disease images were obtained from AIHub. Models, including Multi-Layer Perceptron (MLP), Boosted Stacking Ensemble (BSE), H2O AutoML, Random Forest, and Tree-based Pipeline Optimization Tool (TPOT), were trained and their accuracy assessed. Results: The TPOT achieved the highest accuracy (94.50 percent), due to automated pipeline optimization and ensemble learning. H2O AutoML also performed well at 94.25 percent, illustrating the effectiveness of automated selection for intricate imaging tasks. Other models scored lower. Conclusion: These findings highlight the potential of AI-driven solutions for faster and more precise pet skin disease diagnoses. Future investigations should incorporate broader disease varieties, multimodal data, and clinical validations to solidify the practicality of these approaches in veterinary medicine.

PURPOSES : For autonomous vehicles, abnormal situations, such as sudden changes in driving speed and sudden stops, may occur when they leave the operational design domain. This may adversely affect the overall traffic flow by affecting not only autonomous vehicles but also the driving environment of manual vehicles. Therefore, to minimize the traffic problems and adverse effects that may occur in mixed traffic situations involving manual and autonomous vehicles, an autonomous vehicle driving support system based on traffic operation optimization is required. The main purpose of this study was to build a big-data-classification system by specifying data classification to support the self-driving of Lv.4 autonomous vehicles and matching it with spatio-temporal data. METHODS : The research methodology is explained through a review of related literature, and a traffic management index and big-dataclassification system were built. After collecting and mapping the ITS history traffic information data of an actual Living Lab city, the data were classified using the traffic management indexing method. An AI-based model was used to automatically classify traffic management indices for real-time driving support of Lv.4 autonomous vehicles. RESULTS : By evaluating the AI-based model performance using the test data from the Living Lab city, it was confirmed that the data indexing accuracy was more than 98% for the KNN, Random Forest, LightGBM, and CatBoost algorithms, but not for Logistics Regression. The data were severely unbalanced, and it was necessary to classify very low probability nonconformities; therefore, precision is also important. All four algorithms showed similarly good performances in terms of accuracy. CONCLUSIONS : This paper presents a method for efficient data classification by developing a traffic management index to easily fuse and analyze traffic data collected from various institutions and big data collected from autonomous vehicles. Additionally, EdgeRSU is presented to support the driving of Lv.4 autonomous vehicles in mixed autonomous and manual vehicles traffic situations. Finally, a database was established by classifying data automatically indexed through AI-based models to quickly collect and use data in real-time in large quantities.

교통안전시설물의 관리는 도로교통의 안전과 직결되는 문제이다. 운전자는 신호등, 표지판, 노면표시 등을 통해 운전에 필요한 정보 를 얻는다. 노후된 표지판과 노면표시는 운전자에게 잘못된 정보를 제공할 수 있으므로 주기적인 시설물의 관리가 필요하다. 본 연구 는 딥 러닝 기술을 활용해 운전자 시각의 영상 자료에서 교통안전표지를 자동으로 탐지하고자 하며, 교통안전표지의 공통된 색상 특 징을 기반으로 클래스를 그룹으로 묶어 데이터셋을 구축하는 방법을 제안한다. 객체탐지의 성능지표로 널리 활용되는 mAP를 사용해 클래스 묶음 여부에 따른 탐지 성능을 비교한 결과, 색상 기반 클래스 묶음을 적용한 모델의 탐지 성능이 비교군에 비해 약 36% 상승 함을 확인하였다.