Article
Automation & Control Systems
Lianlian Zhang, Fei Qiao, Junkai Wang, Xiaodong Zhai
Summary: The proposed framework in this paper uses health degree (HD) to quantitatively describe the health status of equipment. The method first removes redundant features using principal component analysis (PCA), then extracts normal observations using a support vector data description (SVDD) algorithm. Finally, an improved incremental SVDD algorithm (NISVDD) is introduced for online updating of the normal sample set to improve accuracy and computational efficiency.
IEEE TRANSACTIONS ON SYSTEMS MAN CYBERNETICS-SYSTEMS
(2021)
Article
Computer Science, Artificial Intelligence
Arin Chaudhuri, Carol Sadek, Deovrat Kakde, Haoyu Wang, Wenhao Hu, Hansi Jiang, Seunghyun Kong, Yuwei Liao, Sergiy Peredriy
Summary: Support vector data description (SVDD) is a popular anomaly detection technique that requires the use of a Gaussian kernel, with the bandwidth parameter being crucial for optimal performance. This paper introduces a new unsupervised method for selecting the Gaussian kernel bandwidth, utilizing a low-rank representation of the kernel matrix. The new technique is competitive with existing methods for low-dimensional data and excels in handling high-dimensional data.
PATTERN RECOGNITION
(2021)
Article
Engineering, Mechanical
Dayang Li, Huimin Gao, Kun Yang, Fanhao Zhou, Xinfa Shi
Summary: Oil monitoring is crucial in determining equipment's operational state by analyzing the data obtained from it. This study proposes an anomaly identification model based on LSTM and SVDD for time series wear state data collected through online monitoring system. Through the model, the wear condition of the oil can be predicted and abnormal identification of the equipment's lubricating oil can be achieved. The experimental results show high prediction accuracy and low prediction loss of the trend prediction model, indicating reliable prediction of future anomalies using the LSTM-SVDD method.
Article
Engineering, Electrical & Electronic
Chaoqi Zhang, Langfu Cui, Qingzhen Zhang, Yang Jin, Xiaoxuan Han, Yan Shi
Summary: This paper proposes an anomaly detection method for aeroengine gas path based on piecewise linear representation (PLR) and support vector data description (SVDD). Experimental results show the superiority of this method in anomaly detection for aeroengine gas path.
IEEE SENSORS JOURNAL
(2022)
Article
Engineering, Mechanical
Dandan Peng, Chenyu Liu, Wim Desmet, Konstantinos Gryllias
Summary: This study proposes a deep learning-based anomaly detection method, called deep support vector data description (deep SVDD), for wind turbine monitoring. Compared to the traditional SVDD approach, this method combines a deep convolutional neural network with the SVDD detector to automatically extract effective features. Experimental results show that the method can effectively detect the generation of ice on wind turbine blades with a successful detection rate of 91.45%.
JOURNAL OF ENGINEERING FOR GAS TURBINES AND POWER-TRANSACTIONS OF THE ASME
(2023)
Article
Engineering, Industrial
Qun Chao, Yuechen Shao, Chengliang Liu, Xiaoxue Yang
Summary: In this study, a health evaluation model for axial piston pumps is developed, which only requires normal samples for model training. The proposed method uses density weighted support vector data description (SVDD) to determine the normal baseline level of an axial piston pump and constructs a dimensionless health index to score the pump's health condition. Experimental results show that the proposed method can effectively evaluate the pump's health condition through the quantifiable health index.
RELIABILITY ENGINEERING & SYSTEM SAFETY
(2023)
Article
Computer Science, Artificial Intelligence
Wenjun Hu, Tianjie Hu, Yuzhen Wei, Jungang Lou, Shitong Wang
Summary: This paper introduces a novel support vector data description method called GL-SVDD, which uses distance metrics and probability densities to regularize the tradeoff parameter and shows encouraging performance in outlier detection.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Computer Science, Information Systems
Xinghua Li, Hengyou Zhang, Yinbin Miao, Siqi Ma, Jianfeng Ma, Ximeng Liu, Kim-Kwang Raymond Choo
Summary: This article discusses anomaly detection on the Controller Area Network (CAN) bus in the Internet of Vehicles (IoV). Existing detection schemes have limitations, so the article proposes a mechanism and two improved schemes to enhance accuracy, recall rate, and reduce computation overhead.
IEEE INTERNET OF THINGS JOURNAL
(2022)
Article
Chemistry, Multidisciplinary
Byoung-Doo Oh, Hyung Choi, Won-Jong Chin, Chan-Young Park, Yu-Seop Kim
Summary: The paper proposes an impact-echo (IE) method based on deep support vector data description (Deep SVDD) for economical void detection inside a duct. A deep SVDD model is used to classify normal and defective data obtained from IE measurements. Experimental results show that the proposed model achieves an accuracy improvement of approximately 47% compared to a supervised learning approach.
APPLIED SCIENCES-BASEL
(2023)
Article
Engineering, Multidisciplinary
Kepeng Qiu, Weihong Song, Peng Wang
Summary: Abnormal data detection for industrial processes is crucial to ensure production safety, but establishing an effective detection model is challenging. This work proposes a method based on adversarial autoencoders and support vector data description to address issues such as a small amount of data, trade-offs between sparsity and accuracy, and weak generalization ability. The proposed method enhances the feature diversity of the process data and automatically optimizes model parameters to improve the model's generalization ability. Experimental results on benchmark datasets and an industrial fermentation process demonstrate the effectiveness of the proposed method.
MEASUREMENT SCIENCE AND TECHNOLOGY
(2022)
Article
Energy & Fuels
Fei Chu, Zhenlin Lu, Shuowei Jin, Xin Liu, Ziyang Yu
Summary: In this study, a flexible dual-threshold SVDD fault warning algorithm is proposed to address the challenges related to complex network topology, accessible data, and missing fault data in the power grid. By combining wavelet packet energy features with Spearman, accurate feature extraction of multiple signal types is achieved. The proposed algorithm includes a relaxed SVDD boundary, a division of hypersphere space, and an adaptive update strategy, which enhance sensitivity to fault samples and reduce the risk of missed detection.
FRONTIERS IN ENERGY RESEARCH
(2022)
Article
Computer Science, Interdisciplinary Applications
Xinlu Zong, Zhen Chen, Lu Zhang
Summary: Abnormal event detection is achieved by extracting features using computer vision technology and applying a classification model. This paper proposes a new feature based on motion entropy to accurately describe the motion characteristics of events. The proposed model ME-DSVDD, based on motion entropy and dual support vector data description, addresses the problem of insufficient sample diversity. Experimental results demonstrate that the proposed method effectively improves the performance of the abnormal event detection model.
INTERNATIONAL JOURNAL OF MODERN PHYSICS C
(2023)
Article
Automation & Control Systems
Jing Wang, Pengyang Liu, Shan Lu, Meng Zhou, Xiaolu Chen
Summary: A decentralized fault detection and diagnosis method is proposed for effectively monitoring nonlinear plant-wide processes. It consists of two main activities: mutual information-Louvain based process decomposition and support vector data descriptions (SVDD) based fault diagnosis. The method maps the plant-wide process as an undirected graph using mechanism knowledge and process structure, and then applies mutual information and the Louvain algorithm to decompose the process into reasonable sub-blocks. Decentralized SVDD is used for fault detection in each sub-block, and a Bayesian fusion inference is used to evaluate the detection results. The proposed method is validated in the Tennessee-Eastman (TE) process.
Article
Computer Science, Artificial Intelligence
Shervin Rahimzadeh Arashloo
Summary: This study extends the support vector data description method to a general p-norm (p = 1) penalty function on slacks, enabling the formulation of a non-linear cost in the primal space. By introducing a dual norm into the objective function, the proposed method provides a controlling mechanism to adjust the intrinsic sparsity/uniformity of the problem for enhanced descriptive capability.
PATTERN RECOGNITION
(2022)
Article
Computer Science, Artificial Intelligence
Fahad Sohrab, Jenni Raitoharju, Alexandros Iosifidis, Moncef Gabbouj
Summary: The paper introduces a novel method to project data from multiple modalities into an optimized subspace for one-class classification. By iteratively transforming data from original feature spaces and utilizing information from the class of interest, the method outperforms competing methods on four out of five datasets.
PATTERN RECOGNITION
(2021)
