Article
Engineering, Electrical & Electronic
Yao Ma, Hongbo Shi, Shuai Tan, Yang Tao, Bing Song
Summary: In this article, a consistency regularization autoencoder (CRAE) framework based on encoder-decoder network is proposed to address the problem caused by limited labeled samples. The experiments show that the proposed method is effective for process fault diagnosis when labeled samples are limited.
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT
(2022)
Article
Chemistry, Analytical
Xiaochu Tang, Jiawei Yan, Yuan Li
Summary: In this paper, a new multi-layer conditional variational auto-encoder is proposed, which controls the output data generation direction by injecting label information into the latent subspace and strengthens the correlation between input and output. Two real industrial process cases are compared to demonstrate the superiority and effectiveness of the proposed method.
Article
Computer Science, Artificial Intelligence
Wei Song, Yuxuan Zhang, Soon Cheol Park
Summary: In this paper, an energy and label constrained deep auto-encoder (ELDAE) is proposed to improve feature extraction ability for classification. Experimental results demonstrate that ELDAE outperforms six state-of-the-art algorithms in terms of classification accuracy.
EXPERT SYSTEMS WITH APPLICATIONS
(2021)
Article
Computer Science, Artificial Intelligence
Zisheng Wang, Jianping Xuan, Tielin Shi
Summary: Vibration signals can be used for fault diagnosis, but traditional supervised learning is not practical for this task. In this study, a semi-supervised deep reinforcement learning method is proposed to improve the accuracy of fault diagnosis. Experimental results show that this method performs better than other intelligent methods, even with a small number of labeled samples, in compound fault diagnosis.
NEURAL COMPUTING & APPLICATIONS
(2022)
Article
Chemistry, Analytical
Keshav Thapa, Yousung Seo, Sung-Hyun Yang, Kyong Kim
Summary: The study focuses on classifying human activities and inferring human behavior using modern sensing technology. However, the issue of domain adaptation for inertial sensing-based human activity recognition is still challenging. The requirement of labeled training data for adapting classifiers to new individuals or devices is a significant barrier. We propose a semi-supervised HAR method that improves reconstruction and generation without changes to a pre-trained classifier, achieving competitive improvement in handling new and unlabeled activity.
Article
Computer Science, Artificial Intelligence
Kenta Yamamoto, Koji Inoue, Tatsuya Kawahara
Summary: Character of spoken dialogue systems is crucial for system impression and user rapport. A character expression model was proposed, which controlled spoken dialogue behaviors to express extroversion, emotional instability, and politeness. The main challenge was the expensive and time-consuming collection of labeled pair data, solved by using semi-supervised learning with a variational auto-encoder. Results showed that the proposed model accurately expressed characters compared to a supervised baseline. Furthermore, implementing the model in an autonomous robot system and conducting subjective experiments confirmed its effectiveness in improving user impression in formal dialogue scenarios.
COMPUTER SPEECH AND LANGUAGE
(2023)
Article
Computer Science, Artificial Intelligence
Yuyan Zhang, Liang Gao, Xiaoyu Wen, Haoqi Wang
Summary: This study proposes a novel EOCA method to improve fault diagnosis accuracy by learning robust features from noisy input. Sixteen different OCAs with complementary activation functions and a new ensemble strategy based on expert models are developed to produce stable diagnosis results. Experiment results show that EOCA outperforms several advanced ensemble methods under strong noise.
EXPERT SYSTEMS WITH APPLICATIONS
(2022)
Article
Computer Science, Artificial Intelligence
Xiangli Yang, Zixing Song, Irwin King, Zenglin Xu
Summary: This paper provides a comprehensive survey on deep semi-supervised learning methods, including model design and unsupervised loss functions. It categorizes existing methods into different types and reviews 60 representative methods with a detailed comparison. The paper also discusses the shortcomings of existing methods and proposes heuristic solutions.
IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING
(2023)
Article
Computer Science, Information Systems
Yanfang Fu, Yu Ji, Gong Meng, Wei Chen, Xiaojun Bai
Summary: This study addresses the challenges of limited fault samples, noise interference, and low accuracy in existing fault diagnosis methods for three-phase inverters under real acquisition conditions. By using Wavelet Packet Decomposition (WPD) denoising and a Conditional Variational Auto-Encoder (CVAE) for sample enhancement, the number of samples is increased. The resulting dataset is then used to train an improved deep residual network (SE-ResNet18) fault diagnosis model with a channel attention mechanism, achieving higher fault diagnosis accuracy.
Article
Engineering, Industrial
Jianbo Yu, Xing Liu
Summary: The paper introduces a new DNN model, 1DRCAE, which utilizes unsupervised learning to extract representative features from complex industrial processes, yielding good results. This technique effectively combines convolutional kernels, auto-encoders, and residual learning blocks to provide a new approach for fault detection and feature learning.
INTERNATIONAL JOURNAL OF PRODUCTION RESEARCH
(2022)
Article
Computer Science, Artificial Intelligence
Gabriel Turinici
Summary: The quality of generative models heavily relies on the choice of a good probability distance. To address shortcomings in popular metrics, a class of distances with built-in convexity has been introduced, showing fast implementation and being applied in an adapted Variational Auto-Encoder called Radon-Sobolev VAE, which produces high quality results on standard generative datasets.
Article
Engineering, Electrical & Electronic
Yukun Liu, Daming Shi
Summary: This study aims to distinguish samples using a latent prototype that represents class information in latent space. By introducing the semi-supervised variational prototyping encoder (SS-VPE) method, the sensitivity of the latent prototype to outliers is addressed. Experimental results demonstrate that the SS-VPE method outperforms existing methods in terms of clustering and classification ability in few-shot learning tasks.
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT
(2023)
Article
Computer Science, Artificial Intelligence
Sinan Li, Tianfu Li, Chuang Sun, Xuefeng Chen, Ruqiang Yan
Summary: Proposed an interpretable wavelet packet kernel-constrained convolutional network (WPConvNet) for noise-robust fault diagnosis, which combines the feature extraction ability of wavelet bases and the learning ability of convolutional kernels. The proposed architecture outperforms other diagnosis models in terms of interpretability and noise robustness.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Engineering, Mechanical
Wentao Mao, Wushi Feng, Yamin Liu, Di Zhang, Xihui Liang
Summary: In recent years, deep learning techniques have shown promising prospects in bearing fault diagnosis, but introducing discriminant information about different fault types into the model remains a challenge. A new deep auto-encoder method is proposed to address this issue, utilizing a new loss function with structural discriminant information and a gradient descent method for optimization, leading to improved diagnostic accuracy and stability.
MECHANICAL SYSTEMS AND SIGNAL PROCESSING
(2021)
Article
Engineering, Electrical & Electronic
Shen Zhang, Fei Ye, Bingnan Wang, Thomas G. Habetler
Summary: In this paper, a semi-supervised learning scheme using VAE-based deep generative models for bearing fault diagnosis is proposed, which can effectively leverage a dataset with only a small subset of labeled data samples. This approach outperformed some mainstream supervised and semi-supervised benchmarks with the same percentage of labeled data samples, and can mitigate label inaccuracy when identifying naturally-evolved bearing defects.
IEEE SENSORS JOURNAL
(2021)
Article
Automation & Control Systems
Kai Zhang, Baoping Tang, Lei Deng, Xiaoxia Yu, Jing Wei
Summary: This study aims to obtain the fault location of wind turbines based on a complex network, and improves the fault location method based on heterogeneous nodes complex networks, which is verified by real cases.
ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE
(2021)
Article
Engineering, Multidisciplinary
Kai Zhang, Baoping Tang, Lei Deng, Xiaoli Liu
Summary: This paper proposes a method for fault diagnosis of wind turbine gearboxes by highlighting the essential frequency bands of wavelet coefficients and the fault features of convolution channels. By performing wavelet packet transformation on the raw signal and improving the ResNet, the method effectively enhances the performance of fault diagnosis.
Article
Automation & Control Systems
Honghai Huang, Baoping Tang, Jun Luo, Huayan Pu, Kai Zhang
Summary: This article proposes a new method for gearbox fault diagnosis using multisensor fusion. The residual gated dynamic sparse network is used to improve feature learning and fusion ability. Experimental results and engineering application show that this method is more effective than others under noise interference.
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS
(2022)
Article
Chemistry, Analytical
Xi Xu, Yi Qin, Dejun Xi, Ruotong Ming, Jie Xia
Summary: This paper proposes a novel multi-scale transformer network for improving the segmentation accuracy of marine animals. By designing a dimensionality reduction CNN module and a multi-scale transformer module, rich contextual information from different scales and subregions can be captured. Experimental results demonstrate that the proposed method outperforms existing approaches in marine animal dataset and ISIC 2018 dataset, indicating its important application value for segmenting underwater images.
Article
Engineering, Multidisciplinary
Xuwei Lai, Kai Zhang, Qing Zheng, Zhixuan Li, Guofu Ding, Kun Ding
Summary: Real-time and accurate monitoring of tool wear status is critical for optimizing product quality and production costs. Deep learning has shown potential in tool wear monitoring due to its powerful feature extraction and nonlinear mapping capabilities. However, the lack of interpretability of deep learning models hinders their practical application in machining. In this study, a frequency-spatial hybrid attention network (FSHAN-SPD) driven by tool structure and process parameters is proposed to address this issue. The effectiveness and flexibility of FSHAN-SPD are demonstrated using the PHM2010 dataset and a milling experiment, with interpretable results in both frequency and spatial domains.
Article
Engineering, Multidisciplinary
Kai Zhang, Yantao Liu, Yisheng Zou, Kun Ding, Yongzhi Liu, Qing Zheng, Guofu Ding
Summary: This research proposes an RUL (remaining useful life) prognosis approach that uses a degradation trend feature generation variational autoencoder. By capturing the elements of time distribution for run-to-failure data, this approach resolves the issue of insufficient data. A network with a variational autoencoder and tendency block is used to generate high-quality time series data correlation features. Cross-validation experiments on a bearing dataset confirm the feasibility of the proposed approach, reducing the prediction error by 22.309%.
MEASUREMENT SCIENCE AND TECHNOLOGY
(2023)
Article
Engineering, Manufacturing
Yisheng Zou, Kun Ding, Keming Shi, Xuwei Lai, Kai Zhang, Guofu Ding, Guohao Qin
Summary: The imbalance of samples with different wear states in processing is due to the nonlinear fluctuation of end mill wear rate. Current approaches for end mill wear condition identification have limitations, particularly in the dominance of majority class samples in forming the classification decision boundary. It is challenging to identify the wear state of minority classes due to the close proximity and difficulty to separate minority class samples from the majority class samples. In this study, a deep feature-weighted convolutional neural network (DFWCNN)-based approach is proposed to overcome these limitations and improve identification accuracy, especially for accelerated wear state samples in the case of sample imbalance.
JOURNAL OF MANUFACTURING PROCESSES
(2023)
Article
Engineering, Chemical
Wei Hao, Zhixuan Li, Guohao Qin, Kun Ding, Xuwei Lai, Kai Zhang
Summary: This study proposed a novelty RUL prediction model for rolling bearings based on a bi-channel hierarchical vision transformer to reduce the impact of the above problems on prediction accuracy improvement.
Article
Engineering, Electrical & Electronic
Kai Zhang, Zhixuan Li, Qing Zheng, Guofu Ding, Baoping Tang, Minghang Zhao
Summary: This study presents a bidirectional guidance method that enables deep networks to diagnose faults robustly with the ability to label noise tolerance. The proposed method uses discrete wavelet packet transform (DWPT) to preprocess vibration signals and a bidirectional loss (BL) function is used to improve the robustness of the model in the case of noisy labels. A fast cosine decay (FCD) strategy of the learning rate is also designed to further boost the recognition performance of the model under severe noisy labels. The experimental results demonstrate that the proposed method outperforms other cutting-edge methods in fault diagnosis with severely noisy labels.
IEEE SENSORS JOURNAL
(2023)
Article
Acoustics
Kai Zhang, Kun Ding, Qing Zheng, Yisheng Zou, Guofu Ding
Summary: Transfer across bearings results in a larger domain shift compared to transfer across working conditions. Direct transfer may decrease the diagnostic accuracy of the target bearing due to differences in structural parameters, measurement environments, and working conditions. A novel fault diagnosis method based on pseudo-label transitive domain adaptation networks (PLTDANs) is proposed to address this issue, which includes empirical selection criteria for selecting appropriate intermediate domains and the cross-domain pseudo-label constraint (CDPLC) to improve diagnostic accuracy.
JOURNAL OF VIBRATION AND CONTROL
(2023)
Article
Engineering, Civil
Lei Wang, Kai Zhang, Qing Zheng, Guofu Ding, Weihua Zhang, Dejun Chen, Bin Liu
Summary: In this paper, a metro vehicle anomaly detection method based on undercarriage images and adversarial memory enhancement is proposed, which can accurately detect anomalies.
PROCEEDINGS OF THE INSTITUTION OF MECHANICAL ENGINEERS PART F-JOURNAL OF RAIL AND RAPID TRANSIT
(2023)
Article
Engineering, Industrial
Tao Zuo, Kai Zhang, Qing Zheng, Xianxin Li, Zhixuan Li, Guofu Ding, Minghang Zhao
Summary: This research proposes a hybrid attention-based multiwavelet coefficient fusion method for evaluating the remaining useful life (RUL) of bearings. It creates a two-dimensional map by organizing the decomposed individual frequency bands using multiple wavelets. A hybrid attention-based ConvLSTM network is designed to weight wavelet coefficient channels adaptively. Tests on the PHM2012 rolling bearing dataset validate the superiority of the proposed method in performance index, while also resolving the wavelet basis function matching issue for periodic transient waveforms.
RELIABILITY ENGINEERING & SYSTEM SAFETY
(2023)
Article
Chemistry, Analytical
Yongzhi Liu, Yisheng Zou, Kai Zhang, Pavlos Lazaridis
Summary: This paper proposes a prediction method for the remaining useful life (RUL) of bearings based on feature evaluation and deep transfer learning. By designing methods for feature evaluation and selection, as well as a domain adversarial transfer model, the limitations of poor analytical ability of features and the influence of changing working conditions are effectively overcome.
Article
Engineering, Electrical & Electronic
Zhixuan Li, Kai Zhang, Yongzhi Liu, Yisheng Zou, Guofu Ding
Summary: This study proposes a method for RUL transfer prediction of rolling bearings based on a common benchmark, which uses an attention mechanism autoencoder to extract the benchmark and incorporates a dynamic benchmark constraint to ensure accuracy. Experimental results demonstrate that this method effectively improves prediction accuracy.
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT
(2022)
Article
Engineering, Electrical & Electronic
Kai Zhang, Baoping Tang, Lei Deng, Xiaoxia Yu
Summary: The study presents a fault detection frame based on subspace reconstruction-based robust kernel principal component analysis (SR-RKPCA) model for wind turbines SCADA data. By utilizing RKPCA method, permutation entropy, and combined index, the stability, accuracy, and non-linear feature extraction capability of the wind turbine fault detection model are enhanced.
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT
(2021)
Article
Engineering, Mechanical
Xuanen Kan, Yanjun Lu, Fan Zhang, Weipeng Hu
Summary: A blade disk system is crucial for the energy conversion efficiency of turbomachinery, but differences between blades can result in localized vibration. This study develops an approximate symplectic method to simulate vibration localization in a mistuned bladed disk system and reveals the influences of initial positive pressure, contact angle, and surface roughness on the strength of vibration localization.
MECHANICAL SYSTEMS AND SIGNAL PROCESSING
(2024)
Article
Engineering, Mechanical
Zimeng Liu, Cheng Chang, Haodong Hu, Hui Ma, Kaigang Yuan, Xin Li, Xiaojian Zhao, Zhike Peng
Summary: Considering the calculation efficiency and accuracy of meshing characteristics of gear pair with tooth root crack fault, a parametric model of cracked spur gear is established by simplifying the crack propagation path. The LTCA method is used to calculate the time-varying meshing stiffness and transmission error, and the results are verified by finite element method. The study also proposes a crack area share index to measure the degree of crack fault and determines the application range of simplified crack propagation path.
MECHANICAL SYSTEMS AND SIGNAL PROCESSING
(2024)
Article
Engineering, Mechanical
Rongjian Sun, Conggan Ma, Nic Zhang, Chuyo Kaku, Yu Zhang, Qirui Hou
Summary: This paper proposes a novel forward calculation method (FCM) for calculating anisotropic material parameters (AMPs) of the motor stator assembly, considering structural discontinuities and composite material properties. The method is based on multi-scale theory and decouples the multi-scale equations to describe the equivalence and equivalence preconditions of AMPs of two scale models. The effectiveness of this method is verified by modal experiments.
MECHANICAL SYSTEMS AND SIGNAL PROCESSING
(2024)
Article
Engineering, Mechanical
Hao Zhang, Jiangcen Ke
Summary: This research introduces an intelligent scheduling system framework to optimize the ship lock schedule of the Three Gorges Hub. By analyzing navigational rules, operational characteristics, and existing problems, a mixed-integer nonlinear programming model is formulated with multiple objectives and constraints, and a hybrid intelligent algorithm is constructed for optimization.
MECHANICAL SYSTEMS AND SIGNAL PROCESSING
(2024)
Article
Engineering, Mechanical
Jingjing He, Xizhong Wu, Xuefei Guan
Summary: A sensitivity and reliability enhanced ultrasonic method has been developed in this study to monitor and predict stress loss in pre-stressed multi-layer structures. The method leverages the potential breathing effect of porous cushion materials in the structures to increase the sensitivity of the signal feature to stress loss. Experimental investigations show that the proposed method offers improved accuracy, reliability, and sensitivity to stress change.
MECHANICAL SYSTEMS AND SIGNAL PROCESSING
(2024)
Article
Engineering, Mechanical
Benyamin Hosseiny, Jalal Amini, Hossein Aghababaei
Summary: This paper presents a method for monitoring sub-second or sub-minute displacements using GBSAR signals, which employs spectral estimation to achieve multi-dimensional target detection. It improves the processing of MIMO radar data and enables high-resolution fast displacement monitoring from GBSAR signals.
MECHANICAL SYSTEMS AND SIGNAL PROCESSING
(2024)
Article
Engineering, Mechanical
Xianze Li, Hao Su, Ling Xiang, Qingtao Yao, Aijun Hu
Summary: This paper proposes a novel method for bearing fault identification, which can accurately identify faults with few samples under complex working conditions. The method is based on a Transformer meta-learning model, and the final result is determined by the weighted voting of multiple models.
MECHANICAL SYSTEMS AND SIGNAL PROCESSING
(2024)
Article
Engineering, Mechanical
Xiaomeng Li, Yi Wang, Guangyao Zhang, Baoping Tang, Yi Qin
Summary: Inspired by chaos fractal theory and slowly varying damage dynamics theory, this paper proposes a new health monitoring indicator for vibration signals of rotating machinery, which can effectively monitor the mechanical condition under both cyclo-stationary and variable operating conditions.
MECHANICAL SYSTEMS AND SIGNAL PROCESSING
(2024)
Article
Engineering, Mechanical
Hao Wang, Songye Zhu
Summary: This paper extends the latching mechanism to vibration control to improve energy dissipation efficiency. An innovative semi-active latched mass damper (LMD) is proposed, and different latching control strategies are tested and evaluated. The latching control can optimize the phase lag between control force and structural response, and provide an innovative solution to improve damper effectiveness and develop adaptive semi-active dampers.
MECHANICAL SYSTEMS AND SIGNAL PROCESSING
(2024)
Article
Engineering, Mechanical
Menghao Ping, Xinyu Jia, Costas Papadimitriou, Xu Han, Chao Jiang, Wang-Ji Yan
Summary: Identification of non-Gaussian processes is a challenging task in engineering problems. This article presents an improved orthogonal series expansion method to convert the identification of non-Gaussian processes into a finite number of non-Gaussian coefficients. The uncertainty of these coefficients is quantified using polynomial chaos expansion. The proposed method is applicable to both stationary and nonstationary non-Gaussian processes and has been validated through simulated data and real-world applications.
MECHANICAL SYSTEMS AND SIGNAL PROCESSING
(2024)
Article
Engineering, Mechanical
Lei Li, Wei Yang, Dongfa Li, Jianxin Han, Wenming Zhang
Summary: The frequency locking phenomenon induced by modal coupling can effectively overcome the dependence of peak frequency on driving strength in nonlinear resonant systems and improve the stability of peak frequency. This study proposes the double frequencies locking phenomenon in a three degrees of freedom (3-DOF) magnetic coupled resonant system driven by piezoelectricity. Experimental and theoretical investigations confirm the occurrence of first frequency locking and the subsequent switching to second frequency locking with the increase of driving force. Furthermore, a mass sensing scheme for double analytes is proposed based on the double frequencies locking phenomenon.
MECHANICAL SYSTEMS AND SIGNAL PROCESSING
(2024)
Article
Engineering, Mechanical
Kai Ma, Jingtao Du, Yang Liu, Ximing Chen
Summary: This study explores the feasibility of using nonlinear energy sinks (NES) as replacements for traditional linear tuned mass dampers (TMD) in practical engineering applications, specifically in diesel engine crankshafts. The results show that NES provides better vibration attenuation for the crankshaft compared to TMD under different operating conditions.
MECHANICAL SYSTEMS AND SIGNAL PROCESSING
(2024)
Article
Engineering, Mechanical
Wentao Xu, Li Cheng, Shuaihao Lei, Lei Yu, Weixuan Jiao
Summary: In this study, a high-precision hydraulic mechanical stand and a vertical mixed-flow pumping station device were used to conduct research on cavitation signals of mixed-flow pumps. By analyzing the water pressure pulsation signal, it was found that the power spectrum density method is more sensitive and capable of extracting characteristics compared to traditional time-frequency domain analysis. This has significant implications for the identification and prevention of cavitation in mixed-flow pump machinery.
MECHANICAL SYSTEMS AND SIGNAL PROCESSING
(2024)
Article
Engineering, Mechanical
Xiaodong Chen, Kang Tai, Huifeng Tan, Zhimin Xie
Summary: This paper addresses the issue of parasitic motion in microgripper jaws and its impact on clamping accuracy, and proposes a symmetrically stressed parallelogram mechanism as a solution. Through mechanical modeling and experimental validation, the effectiveness of this method is demonstrated.
MECHANICAL SYSTEMS AND SIGNAL PROCESSING
(2024)
Article
Engineering, Mechanical
Zhifeng Shi, Gang Zhang, Jing Liu, Xinbin Li, Yajun Xu, Changfeng Yan
Summary: This study provides useful guidance for early bearing fault detection and diagnosis by investigating the effects of crack inclination and propagation direction on the vibration characteristics of bearings.
MECHANICAL SYSTEMS AND SIGNAL PROCESSING
(2024)