Article
Automation & Control Systems
Milad Rezamand, Mojtaba Kordestani, Marcos E. Orchard, Rupp Carriveau, David S-K Ting, Mehrdad Saif
Summary: A hybrid prognostic method using SCADA and vibration signals is introduced to predict the remaining useful life of wind turbine bearings. Experimental data validation shows higher RUL accuracy compared to the Bayesian algorithm.
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS
(2021)
Article
Engineering, Mechanical
Feng Yang, Mohamed Salahuddin Habibullah, Yan Shen
Summary: This paper proposed a generic prognostics framework with HI dynamic smoothing and multi-model ensemble realization, which enables the incorporation of different types of HI degradations. Experimental studies on real data from 8 induction motors showed that the proposed prognostic method using nonlinearly degrading HI resulted in clear performance improvements compared to linear HI degradation prediction.
MECHANICAL SYSTEMS AND SIGNAL PROCESSING
(2021)
Article
Automation & Control Systems
Mingming Yan, Liyang Xie, Isyaku Muhammad, Xiaoyu Yang, Yaoyao Liu
Summary: This paper proposes a method for bearing prognostics using degradation regression models to estimate the remaining useful life of bearings. Key steps such as detecting elbow point and smoothing are introduced, and the effectiveness of the method is validated through experiments.
Article
Computer Science, Artificial Intelligence
Xiang Li, Wei Zhang, Hui Ma, Zhong Luo, Xu Li
Summary: This paper proposes a deep learning-based RUL prediction method, which aligns the data of different entities in similar degradation levels through a cycle-consistent learning scheme to improve prediction performance. Experimental results suggest that the method offers a novel perspective on RUL estimations.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2022)
Article
Engineering, Industrial
Xiaoyan Shao, Baoping Cai, Yonghong Liu, Junyan Zhang, Zhongfei Sui, Qiang Feng
Summary: A novel hybrid model-data-driven RUL prediction method based on a fusion of Kalman filter and dynamic Bayesian network is proposed in this paper. The method improves accuracy by enhancing the performance of observation values through DBN and considering estimation error and observation error. The uncertainty distribution of degradation parameters and environmental parameters is integrated into the state estimation model. Numerical simulation of a subsea Christmas tree valves demonstrates the advantages of the proposed RUL prediction method.
RELIABILITY ENGINEERING & SYSTEM SAFETY
(2023)
Article
Engineering, Multidisciplinary
Xieyi Chen, Yi Wang, Haoran Sun, Hulin Ruan, Yi Qin, Baoping Tang
Summary: Gear is crucial for mechanical equipment, and its health directly influences the overall operation of the equipment. Therefore, accurately predicting the remaining useful life (RUL) of gearboxes is of great significance. However, current deep learning-based RUL prediction methods often overlook trend characteristics and focus on the fluctuation patterns of degradation data. To address this issue, a generalized degradation tendency tracking strategy (GDTTS) is proposed to improve the prediction performance by capturing both trend and fluctuation characteristics. Experimental results on real gearbox datasets demonstrate the effectiveness of the proposed strategy.
Article
Automation & Control Systems
Xuewen Zhang, Yan Qin, Chau Yuen, Lahiru Jayasinghe, Xiang Liu
Summary: Efforts have been made to develop an enhanced RUL framework with data self-generation for both noncyclic and cyclic degradation systems. The proposed method successfully improves RUL estimation accuracy through high-quality data generation and hierarchical data integration, achieving state-of-the-art results in prognostic tasks.
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS
(2021)
Article
Computer Science, Interdisciplinary Applications
Dmitry Zhevnenko, Mikhail Kazantsev, Ilya Makarov
Summary: The paper addresses the issue of controlling industrial devices based on sensor readings. Existing methods rely on feature extraction and prediction. The proposed method involves interacting multiple blocks with different complexities to aggregate information over time and create a shared latent space for RUL prediction. By using a new loss function and adapting hierarchical convolution, a novel TFI model achieves state-of-the-art performance on the C-MAPSS dataset.
JOURNAL OF INDUSTRIAL INFORMATION INTEGRATION
(2023)
Article
Automation & Control Systems
Wenhan Zhang, Zhenhua Wang, Tarek Raissi, Rong Su
Summary: In this paper, a novel ellipsoid-based framework is proposed for fault estimation and remaining useful life prognosis. The framework consists of three steps: actuator fault interval estimation, parameter estimation using an ellipsoid-based extended Kalman filter, and prediction of future degradation states. Numerical simulations were conducted to verify the viability and validity of the approach.
INTERNATIONAL JOURNAL OF ROBUST AND NONLINEAR CONTROL
(2023)
Article
Automation & Control Systems
Jaeyeon Jang, Chang Ouk Kim
Summary: In this article, a novel health representation learning method based on a Siamese network is proposed to prevent overfitting and enable robust Remaining Useful Life (RUL) prediction.
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS
(2022)
Article
Engineering, Industrial
Juseong Lee, Mihaela Mitici
Summary: This study proposes a framework that integrates data-driven probabilistic Remaining-Useful-Life (RUL) prognostics with predictive maintenance planning, using aircraft turbofan engines as an example. By employing this framework, the total maintenance cost can be reduced, unscheduled maintenance can be prevented, and the wasted life of engines can be limited.
RELIABILITY ENGINEERING & SYSTEM SAFETY
(2023)
Article
Computer Science, Artificial Intelligence
Anil Kumar, Chander Parkash, Hesheng Tang, Jiawei Xiang
Summary: The proposed intelligent framework seamlessly integrates degradation monitoring, defect identification, and remaining useful life (RUL) estimation for comprehensive and efficient bearing health assessment. The framework utilizes advanced techniques such as directed divergence measurement, graph convolution network, and dynamic analysis-assisted filtering to improve accuracy in anomaly detection and RUL estimation, ultimately enhancing system reliability and maintenance strategies.
ADVANCED ENGINEERING INFORMATICS
(2023)
Article
Automation & Control Systems
Linxiao Qin, Shuo Zhang, Tao Sun, Xudong Zhao
Summary: This article proposes an interpretable routine for estimating remaining useful life (RUL) using deep learning, which shows improvements in performance compared to baseline approaches and provides interpretability by detecting critical flight stages and abnormal subsystems.
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS
(2023)
Article
Engineering, Mechanical
Yang Chang, Jianxiao Zou, Shicai Fan, Chao Peng, Huajing Fang
Summary: This paper proposes a prognostic technique with the capability of uncertainty management, which consists of two phases to reduce the uncertainty and ensure the reliability of the prognostic result.
MECHANICAL SYSTEMS AND SIGNAL PROCESSING
(2022)
Article
Engineering, Mechanical
Brian Ellis, P. Stephan Heyns, Stephan Schmidt
Summary: This article introduces a hybrid approach for prognosis of mechanical components, which combines physics-based and data-driven methods and is applied to turbomachine rotor blades. Experimental results show that the hybrid approach outperforms other methods in predicting crack length and improves the accuracy and precision of remaining useful life estimation.
MECHANICAL SYSTEMS AND SIGNAL PROCESSING
(2022)
Article
Physics, Multidisciplinary
Rui-Qiang Wang, Zhen-Qiang Yin, Xiao-Hang Jin, Rong Wang, Shuang Wang, Wei Chen, Guang-Can Guo, Zheng-Fu Han
Summary: Quantum key distribution (QKD) allows secure sharing of keys between remote parties. Continuous phase randomization, commonly assumed in QKD protocols, is challenged in experiments. In this study, we propose a technique based on conjugate measurement and quantum state distinguishment to analyze the security of a QKD protocol with discrete-phase randomization. Our results show that TF-QKD with a reasonable number of discrete random phases can achieve satisfactory performance, but finite-size effects become more notable. This method is also applicable to other QKD protocols.
Article
Engineering, Electrical & Electronic
Xiaohang Jin, Xiaoying Zhang, Xu Cheng, Guoqian Jiang, Lesedi Masisi, Wei Huang
Summary: A multilevel feature extraction model combining physical information and intelligent algorithm is proposed for icing detection in wind turbines. Features characterizing the severity of icing are extracted using power characteristic curve technology and feature selection technology, and a fully connected neural network with triplet loss function is designed for extracting fusion feature. Evaluation results in two wind turbines show that fusion features extracted from the proposed model are more stable, reliable, and accurate compared to other feature groups.
IEEE SENSORS JOURNAL
(2023)
Article
Computer Science, Information Systems
Xiaohang Jin, Hao Wang, Ziqian Kong, Zhengguo Xu, Wei Qiao
Summary: An approach based on feature transfer learning and modified generative adversarial network is proposed to effectively build a condition monitoring (CM) model for wind turbines (WTs) with limited operational data. The proposed method first constructs a CM model by analyzing labeled data from WTs with the assistance of an autoencoder. Then, a generative adversarial network is trained to establish a mapping relationship between the features of different WTs. Finally, the health status of the target WT is determined by analyzing the online collected data using the proposed approach. Two case studies demonstrate the successful transfer of CM knowledge from source WT to target WT and the good performance in target WT CM.
Article
Automation & Control Systems
Jie Li, Yu Wang, Yanyang Zi, Haijun Zhang, Chen Li
Summary: Due to the lack of faulty data, intelligent networks need to learn fault knowledge from other machines. However, data from different machines may introduce individualized deviations, leading to overfitting and reduced generalization ability. To address this, this article proposes a collaborative multimachine generalization method called causal consistency network (CCN), which mines invariant causal information to achieve knowledge generalization. CCN uses causal consistency loss to depict the consistency of fault causality representations in deep latent variables, and introduces a collaborative training loss to transform individualized data into consistent representations.
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS
(2023)
Article
Computer Science, Artificial Intelligence
Jiahao Chen, Yu Wang
Summary: The trend towards greater integration and complexity of mechanical systems poses challenges to preventive maintenance plans. Traditional methods of condition-based maintenance struggle to calculate optimal maintenance thresholds. However, the deep reinforcement learning (DRL) method shows promise in solving complex control and decision-making problems, providing a new approach to maintenance planning in complex systems. Numerical results demonstrate that the DRL-based maintenance model can optimize strategies and balance component maintenance costs with system failure losses in both simple and complex multi-component systems. The proposed model outperforms other maintenance strategies, reducing the overall lifecycle cost of the system.
NEURAL COMPUTING & APPLICATIONS
(2023)
Article
Engineering, Industrial
Yu Wang, Qiufa Liu, Wenjian Lu, Yizhen Peng
Summary: The study of the remaining useful life (RUL) has gained momentum in recent years for ensuring system availability. The proposed general time-varying Wiener process (GTWP) considers the dynamic and multi-source variability of a degradation process jointly. A state-space model is constructed to depict the evolution of model parameters over time, and an approximate analytical form for the estimated RUL is derived under the concept of the first hitting time (FHT). The results from simulation cases and real-world data demonstrate the generalizability, accuracy, and faster convergence of the proposed model compared to existing homogeneous models.
RELIABILITY ENGINEERING & SYSTEM SAFETY
(2023)
Article
Engineering, Industrial
Ziqian Kong, Xiaohang Jin, Zhengguo Xu, Zian Chen
Summary: This paper proposes a contrastive learning framework for remaining useful life (RUL) prediction, aiming to improve the performance of deep learning models by utilizing unlabeled samples.
RELIABILITY ENGINEERING & SYSTEM SAFETY
(2023)
Article
Computer Science, Artificial Intelligence
Junwei Gu, Yu Wang, Tommy W. S. Chow, Mingquan Zhang, Wenjian Lu
Summary: This paper proposes a locally weighted multi-domain collaborative adaptation method (LWMDCA) to solve the problem of lack of labeled data for newly designed equipment in intelligent failure prediction. The method constructs a multi-source collaborative domain based on similarity weighted regular coefficient to provide complete diagnostic knowledge and designs a new feature extractor to focus on important failure characteristics.
KNOWLEDGE-BASED SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Jie Li, Yu Wang, Yanyang Zi, Haijun Zhang, Zhiguo Wan
Summary: This study proposes a causal disentanglement network (CDN) for knowledge generalization and continuous degradation mode diagnosis. CDN utilizes multitask instance normalization and batch normalization structures to learn task-specific knowledge and extract informative features. Additionally, a causal disentanglement loss is introduced to enhance generalization and the capturing of causal invariant fault information.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Computer Science, Hardware & Architecture
Zijun Que, Xiaohang Jin, Zhengguo Xu, Chang Hu
Summary: Remaining useful life (RUL) prediction based on machine learning requires sufficient representative data, but obtaining such data is challenging due to security and economic factors. In this study, an incremental learning approach is proposed to address this problem. It constructs a novel sequence input vector from limited condition monitoring data, proves the orthogonality of the input subspace, and constructs a projector to prevent catastrophic forgetting. An integrated gate recurrent unit model is also constructed to map the relationship between condition monitoring data and RUL. A benchmark-bearing case study demonstrates the effectiveness of this approach in updating the model with new degradation cases.
IEEE TRANSACTIONS ON RELIABILITY
(2023)
Article
Engineering, Electrical & Electronic
Yizhen Peng, Ran Bi, Yu Wang
Summary: This article proposes a switching state-space model with adaptive adjustments for degradation prognosis. It provides a unified framework for nonlinear phase-transition problems by adaptively simplifying patterns in original data. The proposed method reduces the root-mean-square error of the remaining useful lifetime prediction by at least 38% compared to existing methods, as validated using actual bearing data.
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT
(2023)
Article
Engineering, Electrical & Electronic
Ziqi Wang, Xiaohang Jin, Zhengguo Xu
Summary: This study proposes an adaptive condition monitoring method for wind turbines based on a multivariate state estimation technique and continual learning, which can accurately detect potential faults and has strong applicability.
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT
(2023)
Article
Engineering, Electrical & Electronic
Junwei Gu, Yu Wang, Guochao Wang
Summary: This article proposes a multi-instance adversarial learning domain adaptation network (MALDAN) to address the problem of SSD failure prediction. The method utilizes a multi-instance learning approach with an attention mechanism to extract features from unlabeled data. Additionally, an adversarial domain adaptation (DA) method is used to align the distributions of different SSD models for effective failure prediction. The proposed method is validated on Alibaba's dataset and achieves superior performance compared to other approaches.
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT
(2023)
Article
Engineering, Electrical & Electronic
Yuanming Zhang, Weiyue Zhou, Jiacheng Huang, Xiaohang Jin, Gang Xiao
Summary: Remaining useful life (RUL) prediction is crucial for the safety and reliability of equipment. This study introduces a temporal knowledge graph (TKG) informer network, which integrates sensor data and structure data to accurately characterize various spatiotemporal features for RUL prediction. Experimental results demonstrate that the proposed method achieves significantly higher performance compared to other approaches.
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT
(2023)
Article
Engineering, Electrical & Electronic
Yong Chang, Nana Li, Jiyuan Zhao, Yu Wang, Zhe Yang
Summary: This article proposes a new defect detection method for contact wire based on ultrasonic guided wave. The experimental results demonstrate its advantages in position information extraction and time resolution.
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT
(2022)