Article
Computer Science, Artificial Intelligence
Xiaowei Zhou, Ivor W. Tsang, Jie Yin
Summary: Deep Neural Networks have achieved great success in classification tasks, but they are vulnerable to adversarial attacks. Adversarial training is an effective strategy to improve the robustness of DNN models, but existing methods fail to generalize well to standard test data. To achieve a better trade-off between standard accuracy and adversarial robustness, a novel adversarial training framework called LADDER is proposed, which generates high-quality adversarial examples through perturbations on latent features.
Article
Engineering, Electrical & Electronic
Jingde Li, Bojian Chen, Changqing Shen, Dong Wang, Juanjuan Shi, Xingxing Jiang
Summary: Recently, unsupervised domain adaptation (UDA) has been widely used in fault diagnosis. Many UDA methods reduce domain differences through alignment tasks, enabling cross-domain diagnosis of the target domain. However, the alignment and classification tasks are independent in these methods, resulting in contaminated semantic features and insufficient prediction certainty for target domain samples. In this study, a new probability guided domain adversarial network (PG-DAN) is proposed to address this issue. The proposed method utilizes classification ability as a measure of domain differences and introduces a gradient supervision module to guide alignment tasks. Experimental results demonstrate that the proposed method achieves better diagnostic performance.
IEEE SENSORS JOURNAL
(2023)
Article
Computer Science, Artificial Intelligence
Xiaoyu Liang, Yaguan Qian, Jianchang Huang, Xiang Ling, Bin Wang, Chunming Wu, Wassim Swaileh
Summary: This paper proposes a novel adversarial training scheme called Moderate-Margin Adversarial Training (MMAT), which achieves a better trade-off between robustness and natural accuracy by generating finer-grained adversarial examples and designing a hybrid loss.
PATTERN RECOGNITION LETTERS
(2023)
Article
Automation & Control Systems
Wentao Mao, Ling Ding, Yamin Liu, Sajad Saraygord Afshari, Xihui Liang
Summary: This paper proposes a novel online early fault detection method based on deep transfer learning, which enhances the feature sensitivity to early faults and the robustness of the detection model by introducing priori degradation information and designing a deep domain adaptation neural network. Comparative experiments demonstrate the effectiveness of the proposed method in reducing false alarms and accurately detecting fault locations.
Article
Automation & Control Systems
Liang Chen, Qi Li, Changqing Shen, Jun Zhu, Dong Wang, Min Xia
Summary: In this article, a generic domain-regressive framework called ADIG is proposed for fault diagnosis, which leverages adversarial learning to extract domain-invariant knowledge and generalize knowledge from the source domain to diagnose unseen but related target domain signals. Customized strategies of feature normalization and adaptive weight are proposed to improve diagnosis performance.
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS
(2022)
Article
Computer Science, Artificial Intelligence
Shaowei Liu, Hongkai Jiang, Yanfeng Wang, Ke Zhu, Chaoqiang Liu
Summary: This paper proposes an unsupervised domain adaptation approach called Deep Feature Alignment Adaptation Network (DFAAN) to improve the domain adaptability of fault diagnosis. By aligning the latent distributions of two domains guided by a Gaussian prior, a common latent space is created to promote feature alignment. A novel discriminative reconstruction distance based on the autoencoder mechanism is introduced to narrow the discrepancy of the feature distribution. The results of diagnostic experiments show the effectiveness and versatility of the proposed approach.
ADVANCED ENGINEERING INFORMATICS
(2022)
Article
Automation & Control Systems
Qin Wang, Gabriel Michau, Olga Fink
Summary: Domain adaptation aims to improve model performance by transferring knowledge from the source domain to the target domain, but the vulnerability of domain adversarial methods to incomplete target label spaces has been demonstrated. To address this issue, a two-stage unilateral alignment approach is proposed, utilizing source domain interclass relationships to unilaterally align the target domain, which has been shown to be effective in experiments on different tasks.
IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS
(2021)
Article
Computer Science, Artificial Intelligence
Zhenghong Wu, Hongkai Jiang, Shaowei Liu, Yunpeng Liu, Wangfeng Yang
Summary: The application of transfer learning to identify rolling bearing fault has attracted much attention. However, most existing studies focus on single-source domains or multi-source domains constructed from different working conditions of the same machine. This study proposes a conditional distribution-guided adversarial transfer learning network with multi-source domains (CDGATLN) for fault diagnosis of bearings installed on different machines. The network transfers knowledge from multiple source domains to a single target domain and aligns conditional distributions to promote knowledge transfer. Experimental results demonstrate the effectiveness and superiority of CDGATLN.
ADVANCED ENGINEERING INFORMATICS
(2023)
Article
Automation & Control Systems
Chao Zhao, Weiming Shen
Summary: This article proposes an adversarial mutual information-guided single domain generalization network for machinery fault diagnosis, which learns domain-invariant representations to address domain shift problems. A domain generation module is designed to generate fake target domains with significant distribution discrepancies, and an iterative min-max game of mutual information is implemented to learn generalized features for resisting unknown domain shift. Extensive diagnosis experiments on two mechanical rigs validated the effectiveness of the proposed method.
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS
(2023)
Article
Engineering, Electrical & Electronic
Yao Liu, Bojian Chen, Dong Wang, Lin Kong, Juanjuan Shi, Changqing Shen
Summary: In this study, a lifelong learning method based on generative feature replay (LLMGFR) is proposed to tackle the problem of bearing diagnosis with incremental fault types. A feature distillation method is introduced to prevent forgetting in the feature extractor, and the generator is trained to produce features from old tasks. The generated features are combined with real features from the current task to effectively solve the imbalance problem and catastrophic forgetting of the classifier. LLMGFR can learn constantly and adaptively in dynamic environments with incremental fault types based on incremental fault diagnosis cases.
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT
(2023)
Article
Engineering, Industrial
Pengfei Chen, Rongzhen Zhao, Tianjing He, Kongyuan Wei, Jianhui Yuan
Summary: A novel approach named joint attention adversarial domain adaptation (JAADA) is proposed to overcome the challenge of transferring feature representations. The method manually divides the extracted features into multiple regions and uses MMD to reduce the distribution discrepancy among the segmented features. Different weights obtained by the attention mechanism and MMD values are assigned to different regions, and local and global attention are fused into one unified adversarial domain adaptation framework. Comprehensive experiments show that the proposed method achieves superior convergence and improves accuracy by 1.9%, 3.0%, 2.1%, and 3.5% compared to state-of-the-art methods, respectively.
RELIABILITY ENGINEERING & SYSTEM SAFETY
(2023)
Article
Computer Science, Information Systems
Xu Chen, Chuancai Liu, Yue Zhao, Zhiyang Jia, Ge Jin
Summary: This study proposes a novel multi-task adversarial training approach to enhance the adversarial robustness of Bayesian neural networks (BNNs). The approach generates diverse and stronger adversarial examples for training and achieves significant improvements in adversarial robustness compared to state-of-the-art defense methods.
INFORMATION SCIENCES
(2022)
Article
Engineering, Electrical & Electronic
Jingde Li, Changqing Shen, Lin Kong, Dong Wang, Min Xia, Zhongkui Zhu
Summary: In this study, a adversarial domain generalization network (ADGN) based on class boundary feature detection is proposed, which can diagnose faults in unknown operating environments and only uses one fully labeled domain for training.
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT
(2022)
Article
Neurosciences
Sen Lu, Abhronil Sengupta
Summary: Neuromorphic computing algorithms based on Spiking Neural Networks (SNNs) are a disruptive technology in machine learning research. This study aims to develop a training framework based on neuroevolution to optimize the unique properties of SNNs, and shows its effectiveness in image classification and adversarial attack scenarios.
FRONTIERS IN NEUROSCIENCE
(2022)
Article
Engineering, Multidisciplinary
Meixia Jia, Jinrui Wang, Zongzhen Zhang, Baokun Han, Zhaoting Shi, Lei Guo, Weitao Zhao
Summary: This study proposes a novel method called DGDAN, which addresses the issue of incomplete similarity in transfer learning through domain-adversarial network and MMD-guided domain adaptation, further improving the classification accuracy and robustness of fault diagnosis.
MEASUREMENT SCIENCE AND TECHNOLOGY
(2022)
Article
Computer Science, Artificial Intelligence
Fangyu Chen, Yongchang Wei, Hongchang Ji, Gangyan Xu
Summary: This paper introduces a dual-layer network analytical framework for evaluating standard systems in construction safety management and validates its effectiveness through a case study. The research findings suggest that key standards often encompass a wider array of risks, providing suggestions for revising construction standards.
ADVANCED ENGINEERING INFORMATICS
(2024)
Article
Computer Science, Artificial Intelligence
Minghao Li, Qiubing Ren, Mingchao Li, Ting Kong, Heng Li, Huijing Tian, Shiyuan Liu
Summary: This study proposes a method using digital twin technology to construct a collision early warning system for marine piling. The system utilizes a five-dimensional model and four independently maintainable development modules to maximize its effectiveness. The pile positioning algorithm and collision early warning algorithm are capable of providing warnings for complex pile groups.
ADVANCED ENGINEERING INFORMATICS
(2024)
Article
Computer Science, Artificial Intelligence
Seokhyun Ryu, Sungjoo Lee
Summary: This study proposes the use of patent information to develop a robust technology tree and applies it to the furniture manufacturing process. Through methods such as clustering analysis, semantic analysis, and association-rule mining, technological attributes and their relationships are extracted and analyzed. This approach provides meaningful information to improve the understanding of a target technology and supports research and development planning.
ADVANCED ENGINEERING INFORMATICS
(2024)
Article
Computer Science, Artificial Intelligence
Shuai Ma, Kechen Song, Menghui Niu, Hongkun Tian, Yanyan Wang, Yunhui Yan
Summary: This paper proposes a feature-based domain disentanglement and randomization (FDDR) framework to improve the generalization of deep models in unseen datasets. The framework successfully addresses the appearance difference issue between training and test images by decomposing the defect image into domain-invariant structural features and domain-specific style features. It also utilizes randomly generated samples for training to further expand the training sample.
ADVANCED ENGINEERING INFORMATICS
(2024)
Article
Computer Science, Artificial Intelligence
Fang Xu, Tianyu Zhou, Hengxu You, Jing Du
Summary: This study explores the impact of AR-based egocentric perspectives on indoor wayfinding performance. The results reveal that participants using the egocentric perspective demonstrate improved efficiency, reduced cognitive load, and enhanced spatial awareness in indoor navigation tasks.
ADVANCED ENGINEERING INFORMATICS
(2024)
Review
Computer Science, Artificial Intelligence
Yujie Lu, Shuo Wang, Sensen Fan, Jiahui Lu, Peixian Li, Pingbo Tang
Summary: Image-based 3D reconstruction plays a crucial role in civil engineering by bridging the gap between physical objects and as-built models. This study provides a comprehensive summary of the field over the past decade, highlighting its interdisciplinary nature and integration of various technologies such as photogrammetry, 3D point cloud analysis, semantic segmentation, and deep learning. The proposed 3D reconstruction knowledge framework outlines the essential elements, use phases, and reconstruction scales, and identifies eight future research directions. This review is valuable for scholars interested in the current state and future trends of image-based 3D reconstruction in civil engineering, particularly in relation to deep learning methods.
ADVANCED ENGINEERING INFORMATICS
(2024)
Article
Computer Science, Artificial Intelligence
Hang Zhang, Wenhu Wang, Shusheng Zhang, Yajun Zhang, Jingtao Zhou, Zhen Wang, Bo Huang, Rui Huang
Summary: This paper presents a novel framework for segmenting intersecting machining features using deep reinforcement learning. The framework enhances the effectiveness of intersecting machining feature segmentation by leveraging the robust feature representation, decision-making, and automatic learning capabilities of deep reinforcement learning. Experimental results demonstrate that the proposed approach successfully addresses some existing challenges faced by several state-of-the-art methods in intersecting machining feature segmentation.
ADVANCED ENGINEERING INFORMATICS
(2024)
Article
Computer Science, Artificial Intelligence
Chao Zhao, Weiming Shen
Summary: This paper proposes a semantic-discriminative augmentation-driven network for imbalanced domain generalization fault diagnosis, which enhances the model's generalization capabilities through synthesizing reliable samples and optimizing representations.
ADVANCED ENGINEERING INFORMATICS
(2024)
Article
Computer Science, Artificial Intelligence
Ching-Chih Chang, Teng-Wen Chang, Hsin-Yi Huang, Shih-Ting Tsai
Summary: Ideation is the process of generating ideas through exploring visual and semantic stimuli for creative problem-solving. This process often requires changes in user goals and insights. Using pre-designed content and semantic-visual concepts for ideation can introduce uncertainty. An adaptive workflow is proposed in this study that involves extracting and summarizing semantic-visual features, using clusters of adapted information for multi-label classification, and constructing a design exploration model with visualization and exploration.
ADVANCED ENGINEERING INFORMATICS
(2024)
Article
Computer Science, Artificial Intelligence
Zhen Wang, Shusheng Zhang, Hang Zhang, Yajun Zhang, Jiachen Liang, Rui Huang, Bo Huang
Summary: This research proposes a novel approach for machining feature process planning using graph convolutional neural networks. By representing part information with attribute graphs and constructing a learning model, the proposed method achieves higher accuracy and resolves current limitations in machining feature process planning.
ADVANCED ENGINEERING INFORMATICS
(2024)
Article
Computer Science, Artificial Intelligence
Hong-Wei Xu, Wei Qin, Jin-Hua Hu, Yan-Ning Sun, You -Long Lv, Jie Zhang
Summary: Wafer fabrication is a complex manufacturing system, where understanding the correlation between parameters is crucial for identifying the cause of wafer defects. This study proposes a Copula network deconvolution-based framework for separating direct correlations, which involves constructing a complex network correlation diagram and designing a nonlinear correlation metric model. The proposed method enables explainable fault detection by identifying direct correlations.
ADVANCED ENGINEERING INFORMATICS
(2024)
Article
Computer Science, Artificial Intelligence
Yida Hong, Wenqiang Li, Chuanxiao Li, Hai Xiang, Sitong Ling
Summary: An adaptive push method based on feature transfer is proposed to address sparsity and cold start issues in product intelligent design. By constructing a collaborative filtering algorithm model and transforming the rating model, the method successfully alleviates data sparsity and cold start problems.
ADVANCED ENGINEERING INFORMATICS
(2024)
Article
Computer Science, Artificial Intelligence
Hairui Fang, Jialin An, Bo Sun, Dongsheng Chen, Jingyu Bai, Han Liu, Jiawei Xiang, Wenjie Bai, Dong Wang, Siyuan Fan, Chuanfei Hu, Fir Dunkin, Yingjie Wu
Summary: This work proposes a model for real-time fault diagnosis and distance localization on edge computing devices, achieving lightweight design and high accuracy in complex environments. It also demonstrates a high frame rate on edge computing devices, providing a novel solution for industrial practice.
ADVANCED ENGINEERING INFORMATICS
(2024)
Article
Computer Science, Artificial Intelligence
Yujun Jiao, Xukai Zhai, Luyajing Peng, Junkai Liu, Yang Liang, Zhishuai Yin
Summary: This paper proposes a digital twin-based motion forecasting framework that predicts the future trajectories of workers on construction sites, accurately predicting workers' motions in potential risk scenarios.
ADVANCED ENGINEERING INFORMATICS
(2024)
Article
Computer Science, Artificial Intelligence
Ling-Zhe Zhang, Xiang-Dong Huang, Yan-Kai Wang, Jia-Lin Qiao, Shao-Xu Song, Jian-Min Wang
Summary: Time-series DBMSs based on the LSM-tree have been widely applied in various scenarios. The characteristics of time-series data workload pose challenges to efficient queries. To address issues like query latency and inaccurate range, we propose a novel compaction algorithm called Time-Tiered Compaction.
ADVANCED ENGINEERING INFORMATICS
(2024)
