Article
Engineering, Electrical & Electronic
Jie Dong, Yaqi Wang, Kaixiang Peng
Summary: This article proposes a novel fault detection framework to improve the performance of fault detection in the presence of nonstationary and stationary variables. The framework combines slow feature analysis and K-nearest neighbor based on Mahalanobis distance, and considers fault detection rate and false alarm rate by setting dynamic control limits.
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT
(2022)
Article
Computer Science, Artificial Intelligence
Junwei Hu, Yong Zhang, Weigang Li, Xiujuan Zheng, Zhiqiang Tian
Summary: This study aims to address the impact of noise interference on model accuracy and interpretability in industrial fault diagnosis. It proposes an explicable temporal feature network (ETFN) based on Deep SHAP values, which combines adaptive features and empirical features to enhance the model's robustness, and interprets the model diagnosis using Deep SHAP. The proposed ETFN achieves a good balance between stability, accuracy, and interpretation on real-world datasets.
COGNITIVE COMPUTATION
(2023)
Article
Automation & Control Systems
Han Zhou, Hongpeng Yin, Dandan Zhao, Li Cai
Summary: Incremental learning-based fault diagnosis is effective for continuous learning from industrial data. However, in nonstationary processes, the distribution of new data often shifts away from historical data, which affects diagnosis performance. This article focuses on the conditional drift phenomenon, where the conditional distribution of industrial data changes over time. A target mapping strategy and an incremental diagnosis model with adaptation ability are proposed to address this problem. Experimental results on two industrial processes demonstrate the effectiveness of the proposed method.
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS
(2023)
Article
Automation & Control Systems
Min Wang, Donghua Zhou, Maoyin Chen
Summary: In this paper, a model called recursive hybrid variable monitoring (RHVM) is proposed to address the issue of process monitoring with hybrid variables and nonstationarity. RHVM utilizes a recursive strategy to suppress nonstationary trends and reveal fault information, and it has the ability to update itself with arriving samples.
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS
(2022)
Article
Computer Science, Artificial Intelligence
Pengyu Song, Chunhui Zhao, Biao Huang, Jinliang Ding
Summary: Industrial processes often exhibit both temporal and spatial dependencies due to dynamic changes and inter-variable coupling. However, existing methods struggle to effectively separate and represent these dependencies, leading to inaccurate fault detection and isolation. This study proposes a framework that utilizes a double-level separation method and an information aliasing loss function to explicitly represent and isolate temporal and spatial characteristics. By monitoring explicit statistics obtained from the separation modules, anomalies affecting different dependencies can be identified and located. Furthermore, a customized isolation strategy is introduced to accurately characterize and isolate anomalies in temporal and spatial characteristics. The proposed framework is validated through numerical examples and real-world processes.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Automation & Control Systems
Dongdong Liu, Lingli Cui, Weidong Cheng
Summary: Rotating machinery fault diagnosis under nonstationary conditions commonly relies on manual analysis of vibration signals' frequency spectrums or time-frequency representations. However, expert experience heavily determines the results of these methods, and identifying the frequency content becomes difficult due to the intricate interference frequency components caused by complex modulation characteristics and operation conditions. This article investigates a novel intelligent fault diagnosis method for rolling bearings under nonstationary conditions. A flexible generalized demodulation method is proposed, which overcomes the effects of operation conditions on demodulation spectrums. Based on this, a fault feature extraction method is further proposed to capture useful fault information. Experimental results show that the proposed method can automatically recognize health conditions that cannot be manually identified by demodulation spectrums due to interference frequency components and it is more adaptive to new operation conditions because of its definite physical meaning.
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS
(2023)
Article
Computer Science, Information Systems
Yonghui Xu, Shengjie Sun, Huiguo Zhang, Chang'an Yi, Yuan Miao, Dong Yang, Xiaonan Meng, Yi Hu, Ke Wang, Huaqing Min, Hengjie Song, Chuanyan Miao
Summary: This article presents a Robustly Time-aware Graph Embedding (RTGE) method that incorporates temporal smoothness, aimed at improving the performance of knowledge graph embedding by considering temporal information. The proposed method integrates a measure of temporal smoothness in the learning process and provides a task-oriented negative sampling strategy associated with temporally aware information to enhance adaptability and achieve superior performance in various tasks. Experimental results demonstrate the effectiveness of RTGE in entity/relationship/temporal scoping prediction tasks.
ACM TRANSACTIONS ON KNOWLEDGE DISCOVERY FROM DATA
(2022)
Article
Automation & Control Systems
Jian Huang, Xiaoyang Sun, Xu Yang, Yuri A. W. Shardt
Summary: In this paper, an active nonstationary variables selection-based just-in-time co-integration analysis and slow feature analysis monitoring approach is proposed to explore the real-time variations in dynamic processes. The method selects active nonstationary variables by analyzing the time-varying stationarity of online data and updates the offline model using a just-in-time strategy. Co-integration analysis and slow feature analysis are used to extract long-run equilibrium relationships and slowly varying features, and a comprehensive statistic is generated by Bayesian inference to monitor the operation status. Two case studies on benchmark processes demonstrate the advantages and feasibility of the proposed method.
JOURNAL OF PROCESS CONTROL
(2022)
Article
Automation & Control Systems
Chao Jiang, Yusheng Lu, Weimin Zhong, Biao Huang, Dayu Tan, Wenjiang Song, Feng Qian
Summary: Inferential modeling plays a significant role in estimating quality-related process variables in modern manufacturing. This article proposes a new nonlinear extension of probabilistic slow feature analysis (PSFA) under the deep learning framework to enhance dynamic feature extraction and improve prediction accuracy by incorporating variational inference and Monte Carlo inference. The proposed model considers the relevance of inputs with outputs to enhance prediction performance. The model is validated through an industrial hydrocracking process and achieves a significant reduction in root mean squared error compared to PSFA.
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS
(2023)
Article
Automation & Control Systems
Jingxin Zhang, Donghua Zhou, Maoyin Chen, Xia Hong
Summary: A novel continual learning-based probabilistic slow feature analysis algorithm is proposed for monitoring multimode nonstationary processes. It utilizes elastic weight consolidation to handle sequential modes and introduces a regularization term to prevent new data from interfering with learned knowledge.
IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING
(2022)
Article
Engineering, Electrical & Electronic
Daye Li, Jie Dong, Kaixiang Peng
Summary: In this work, a novel adaptive fault detection framework is proposed to solve the nonstationary characteristics caused by external disturbances and the inconsistent data distribution problems. The framework combines stationary features with nonstationary variables extracted by short-time Fourier analysis and employs isolation-based anomaly detection as monitoring metrics. It also introduces model update factor and update strategy to calculate the differences between online normal data and training data, and proposes an adaptive fault detection framework based on short-time Fourier transform-SFA.
IEEE SENSORS JOURNAL
(2023)
Article
Engineering, Electrical & Electronic
Yang Yang, Sha Wei, Tianqi Li, Hui Liu, Jun He
Summary: This article proposes a novel fault feature extraction method to extract the weak fault features of the planet bearing cage. The method utilizes the general parameterized time-frequency transform to accurately extract instantaneous rotational speed information from the planet bearing vibration signal for resampling. The fault characteristic components of the planet bearing cage can be extracted from the square envelop spectrum. The proposed method has been verified in experiments of a planetary transmission system test rig of an armored vehicle.
IEEE SENSORS JOURNAL
(2023)
Article
Computer Science, Information Systems
Yaming Wang, Dawei Xu, Wenqing Huang, Xiaoping Ye, Mingfeng Jiang
Summary: In this paper, an innovative neural network architecture is proposed to address dense Non-Rigid Structure from Motion (NRSFM) challenges. By embedding the 2D sequence image into a low-dimensional space and employing multiple self-attention layers to extract inter-frame features, the network achieves exceptional performance in this task.
Article
Engineering, Chemical
Feng Yu, Jianchang Liu, Dongming Liu, Honghai Wang
Summary: A new feature learning method, SCAE, addresses the issue of CAE's inability to ensure that extracted features are related to fault types, by pretraining the network to learn internal spatial information and fault information. The fault-relevant features obtained by SCAE can clearly distinguish between different fault types, providing more appropriate predefined parameters for fine-tuning to enhance classification performance.
JOURNAL OF THE TAIWAN INSTITUTE OF CHEMICAL ENGINEERS
(2022)
Article
Automation & Control Systems
Gang Chen, Mei Liu, Zhaodan Kong
Summary: The article discusses a method of converting industrial Internet of Things data into actionable intelligence through semantic fault diagnosis, and proposes an algorithm to solve the issue of combinatorial explosion. This algorithm combines ideas from agenda-based searching and imitation learning to train a policy that searches formulas in a strategic order.
IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS
(2021)