• 한국음향학회지
• /
• 제41권2호
• /
• pp.199-209
• /
• 2022

일반적인 비정상 탐지 알고리즘은 사전 데이터를 이용하여 학습된다. 따라서 시간에 따른 정상 데이터의 특징이 변화되는 경우에 기존의 배치 학습 기반 알고리즘의 성능 저하가 불가피하다. 본 논문에서는 정상 데이터의 점진적 특징 변화를 고려할 수 있는 온라인 비정상 탐지 알고리즘을 제안한다. 제안하는 알고리즘은 단일 클래스 분류 모델에 기반하며 오프라인 및 온라인 단계의 학습 과정을 포함한다. 제안된 알고리즘의 오프라인 학습 단계에서는 사전 데이터가 잠재 공간의 중심에 근접하도록 학습하고, 이후 온라인 학습단계에서는 신규 데이터에 의한 점진적 잠재 공간의 중심을 갱신하고, 갱신된 중심을 기준으로 계속 학습을 진행한다. 공개된 수중 음향 데이터를 이용한 실험결과 제안된 온라인 비정상 탐지 알고리즘은 점진적 중심 갱신 및 학습을 위해 단지 2 % 정도의 추가 학습시간이 소요되는 것으로 확인되었다. 반면에 시변 정상데이터가 수신되는 경우에 오프라인 학습 모델과 비교하여 19.10 % 개선된 Area Under the receiver operating characteristic Curve(AUC) 성능을 보였다.

• Smart Structures and Systems
• /
• 제30권6호
• /
• pp.661-671
• /
• 2022

Stayed cables are the key components for transmitting loads in cable-stayed bridges. Therefore, it is very important to evaluate the cable force condition to ensure bridge safety. An online condition assessment and anomaly localization method is proposed for cables based on the spatiotemporal correlation of grouped cable forces. First, an anomaly sensitive feature index is obtained based on the distribution characteristics of grouped cable forces. Second, an adaptive anomaly detection method based on the k-nearest neighbor rule is used to perform dissimilarity measurements on the extracted feature index, and such a method can effectively remove the interference of environment factors and vehicle loads on online condition assessment of the grouped cable forces. Furthermore, an online anomaly isolation and localization method for stay cables is established, and the complete decomposition contributions method is used to decompose the feature matrix of the grouped cable forces and build an anomaly isolation index. Finally, case studies were carried out to validate the proposed method using an in-service cable-stayed bridge equipped with a structural health monitoring system. The results show that the proposed approach is sensitive to the abnormal distribution of grouped cable forces and is robust to the influence of interference factors. In addition, the proposed approach can also localize the cables with abnormal cable forces online, which can be successfully applied to the field monitoring of cables for cable-stayed bridges.

• 인터넷정보학회논문지
• /
• 제25권1호
• /
• pp.109-121
• /
• 2024

섬유, 자동차와 같은 여러 제조 공정에서 설비가 고장이 나 멈추게 되면 기계가 작동하지 않게 되고 이는 기업의 시간적, 금전적 손실로 이어진다. 따라서 설비의 고장이 발생하기 전, 고장을 예측하여 정비할 수 있도록 설비의 이상을 사전에 탐지하는 것이 중요하다. 대부분의 설비 고장 원인은 설비의 필수 부품인 베어링의 고장으로, 베어링의 고장을 진단하는 것은 설비예지보전 연구의 핵심이기도 하다. 본 논문에서는 베어링의 진동 신호를 분석하여 SWT-SVD 전처리 알고리즘을 제안하고 이를 시계열 이상탐지 모델 네트워크 중 하나인 어노멀리 트랜스포머에 적용하여 베어링 이상탐지 모델을 구현한다. 제조공정의 베어링 진동신호는 실시간으로 센서값들의 이력이 작성되어 노이즈가 존재하므로, 이를 줄이기 위해 본 연구에서는 정상 웨이블릿 변환(Stationary Wavelet Transform)을 사용하여 주파수 성분을 추출하고, 특이값 분해(Singular Value Decomposition) 알고리즘을 통해 유의미한 특징들을 추출하는 전처리를 진행한다. 제안하는 SWT-SVD 전처리 방법을 적용한 베어링 이상탐지 모델 실험을 위해 IEEE PHM학회에서 제공하는 PHM-2012-Challenge 데이터 세트를 활용하였으며, 실험 결과는 0.98의 정확도와 0.97의 F1-Score로 우수한 성능을 보였다. 추가로, 성능 향상을 입증하기 위해 선행 연구들과 성능 비교를 진행한다. 비교 실험을 통해 제안한 전처리 방법이 기존의 전처리보다 높은 성능을 보임을 확인하였다.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제17권6호
• /
• pp.1516-1529
• /
• 2023

In this paper, a motor gearbox fault-detection system based on a hierarchical flow-based model is proposed. The proposed system is used for the anomaly detection of a motion sound-based actuator module. The proposed flow-based model, which is a generative model, learns by directly modeling a data distribution function. As the objective function is the maximum likelihood value of the input data, the training is stable and simple to use for anomaly detection. The operation sound of a car's side-view mirror motor is converted into a Mel-spectrogram image, consisting of a folding signal and an unfolding signal, and used as training data in this experiment. The proposed system is composed of an encoder and a decoder. The data extracted from the layer of the pretrained feature extractor are used as the decoder input data in the encoder. This information is used in the decoder by performing an interlayer cross-scale convolution operation. The experimental results indicate that the context information of various dimensions extracted from the interlayer hierarchical data improves the defect detection accuracy. This paper is notable because it uses acoustic data and a normalizing flow model to detect outliers based on the features of experimental data.

• International Journal of Fuzzy Logic and Intelligent Systems
• /
• 제8권4호
• /
• pp.316-321
• /
• 2008

Advanced computer network technology enables computers to be connected in an open network environment. Despite the growing numbers of security threats to networks, most intrusion detection identifies security attacks mainly by detecting misuse using a set of rules based on past hacking patterns. This pattern matching has a high rate of false positives and can not detect new hacking patterns, which makes it vulnerable to previously unidentified attack patterns and variations in attack and increases false negatives. Intrusion detection and analysis technologies are thus required. This paper investigates the asymmetric costs of false errors to enhance the performances the detection systems. The proposed method utilizes the network model to consider the cost ratio of false errors. By comparing false positive errors with false negative errors, this scheme achieved better performance on the view point of both security and system performance objectives. The results of our empirical experiment show that the network model provides high accuracy in detection. In addition, the simulation results show that effectiveness of anomaly traffic detection is enhanced by considering the costs of false errors.

• Journal of Information Processing Systems
• /
• 제7권3호
• /
• pp.435-446
• /
• 2011

Microprocessors are becoming increasingly vulnerable to soft errors due to the current trends of semiconductor technology scaling. Traditional redundant multi-threading architectures provide perfect fault tolerance by re-executing all the computations. However, such a full re-execution technique significantly increases the verification workload on the processor resources, resulting in severe performance degradation. This paper presents a pro-active verification management approach to mitigate the verification workload to increase its performance with a minimal effect on overall reliability. An anomaly-speculation-based filter checker is proposed to guide a verification priority before the re-execution process starts. This technique is accomplished by exploiting a value similarity property, which is defined by a frequent occurrence of partially identical values. Based on the biased distribution of similarity distance measure, this paper investigates further application to exploit similar values for soft error tolerance with anomaly speculation. Extensive measurements prove that the majority of instructions produce values, which are different from the previous result value, only in a few bits. Experimental results show that the proposed scheme accelerates the processor to be 180% faster than traditional fully-fault-tolerant processor with a minimal impact on overall soft error rate.

• 대한원격탐사학회:학술대회논문집
• /
• 대한원격탐사학회 2003년도 Proceedings of ACRS 2003 ISRS
• /
• pp.1413-1414
• /
• 2003

Interannual variability in the patterns of satellitederived pigment concentrations, sea-level height anomaly, sea surface temperature anomaly, and zonal wind anomaly are observed during the 2002-2003 El Ni${\tilde{n}}$o. The largest spatial extent of the phytoplankton bloom was recovery from El Ni${\tilde{n}}$o over the equatorial Pacific. The evolution towards a warm episode (El Ni${\tilde{n}}$o) started from spring of 2002 and continued during January 2003, while equatorial Sea Surface Temperature Anomaly (SSTA) remained greater than +1$^{\circ}$C in the central equatorial Pacific. The EOS (Earth Observing System) and OSMI (Ocean Scanning Multispectral Imager) data are used for detection of dramatic changes in the patterns of pigment concentration during El Ni${\tilde{n}}$o.

• 한국정보통신학회:학술대회논문집
• /
• 한국정보통신학회 2021년도 춘계학술대회
• /
• pp.581-583
• /
• 2021

코로나19의 펜데믹 현상으로 인하여 모바일 서비스의 활용 비중이 높아지고 있다. 그러나, 대부분의 모바일 장비에 대한 이상 현상을 장비의 알람 중심으로 인지하므로, 복잡한 장애가 발생할 경우에 직관적으로 장비의 문제 판별하기 어려운 한계가 발생한다. 이를 보완하기 위해서 본 연구에서는 장비의 알람과 성능 정보를 조합하여 직관적으로 문제를 인지할 수 있도록 RRCF 알고리즘을 활용하여 Anomaly Score 생성하였으며, 과거 장애 이력을 97% 검출하는 효과를 검증하였다.

• Journal of Platform Technology
• /
• 제11권6호
• /
• pp.3-12
• /
• 2023

In today's system operation, it is difficult to detect failures and take immediate action in the case of a shortage of manpower compared to the number of equipment or failures in vulnerable time zones, which can lead to delays in failure recovery. In addition, various algorithms exist to detect abnormal symptom data, and it is important to select an appropriate algorithm for each problem. In this paper, an ensemble-based isolation forest model was used to efficiently detect multivariate point anomalies that deviated from the mean distribution in the data set generated to predict system failure and minimize service interruption. And since significant changes in memory space usage are observed together with changes in CPU usage, the problem is solved by using LSTM-Auto Encoder for a collective anomaly in which another feature exhibits an abnormal pattern according to a change in one by comparing two or more features. did In addition, evaluation indicators are set for the performance evaluation of the model presented in this study, and then AI model evaluation is performed.

• 한국멀티미디어학회논문지
• /
• 제21권3호
• /
• pp.348-357
• /
• 2018

The topic of anomaly detection is one of the emerging research themes in computer vision, computer interaction, video analysis and monitoring. Observers focus attention on behaviors that vary in the magnitude or direction of the motion and behave differently in rules of motion with other objects. In this paper, we use this information and propose a system that detects abnormal behavior by using simple features extracted by optical flow. Our system can be applied in real life. Experimental results show high performance in detecting abnormal behavior in various videos.