Jianghong Zhou
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Research Interests
Jianghong Zhou has a total of 48 co-authors. Here are the top co-authors by number of shared publications:
- Yi Qin
Chongqing University
17 shared publications - Quan Qian
University of Electronic Science and Technology of China
6 shared publications - Dingliang Chen
State Key Laboratory of Mechanical Transmission for Advanced Equipment, Chongqing University, Chongqing, China
4 shared publications - Jiahong Yang
State Key Laboratory of Mechanical Transmission for Advanced Equipment, Chongqing University, Chongqing, China
4 shared publications - Jun Luo
School of Advanced Interdisciplinary Sciences, University of Chinese Academy of Sciences, Beijing, China
3 shared publications Bingchang Hou
State Key Laboratory of Mechanical Transmission for Advanced Equipment, Chongqing University, Chongqing, Chin
2 shared publications
Latest Publications
- Top Cited Papers
- 1 October 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE/ASME Transactions on Mechatronics
- Vol. 31 (5) , 5066-5076
Knowledge Library Network for Non-Exemplar Incremental Remaining Useful Life Prediction
Aiming at the difference of various incremental remaining useful life (RUL) prediction tasks, a novel nonexemplar incremental learning (IL) method is proposed, named knowledge library network (KLN). First, a multilevel focal self-attention (MLFSA) is developed to model the multiscale temporal dependencies and extract the key features. On this basis, a dual domain fusion temporal convolutional network (DDFTCN) is designed to mine the degradation information from both time-domain and frequency-domain data. Next, a task similarity judgment strategy specifically tailored for IL scenarios is proposed to assess the discrepancies between different tasks. Based on the task differences, a network augmentation rule is proposed for implementing the feature extraction network augmentation or task branch network augmentation. The feature extraction network augmentation is employed for the tasks with significant differences, while the task branch network augmentation is utilized for the tasks with small differences. Especially for the task branch network augmentation, a retention and absorption distillation strategy is proposed to enhance the model's capability to retain previous knowledge while adapting to new task. The network augmentation network rule is integrated into DDFTCN to form the KLN. Finally, the experimental results on the life-cycle bearing and gear datasets demonstrate that DDFTCN achieve an average accuracy of 93.6% across both datasets, surpassing the existing prediction methods. The experimental results on the incremental RUL prediction tasks validate the superiority of KLN over the typical and state-of-the-art IL methods.
- 1 October 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE/ASME Transactions on Mechatronics
- Vol. 31 (5) , 5066-5076
Top Cited PapersKnowledge Library Network for Non-Exemplar Incremental Remaining Useful Life Prediction
Aiming at the difference of various incremental remaining useful life (RUL) prediction tasks, a novel nonexemplar incremental learning (IL) method is proposed, named knowledge library network (KLN). First, a multilevel focal self-attention (MLFSA) is developed to model the multiscale temporal dependencies and extract the key features. On this basis, a dual domain fusion temporal convolutional network (DDFTCN) is designed to mine the degradation information from both time-domain and frequency-domain data. Next, a task similarity judgment strategy specifically tailored for IL scenarios is proposed to assess the discrepancies between different tasks. Based on the task differences, a network augmentation rule is proposed for implementing the feature extraction network augmentation or task branch network augmentation. The feature extraction network augmentation is employed for the tasks with significant differences, while the task branch network augmentation is utilized for the tasks with small differences. Especially for the task branch network augmentation, a retention and absorption distillation strategy is proposed to enhance the model's capability to retain previous knowledge while adapting to new task. The network augmentation network rule is integrated into DDFTCN to form the KLN. Finally, the experimental results on the life-cycle bearing and gear datasets demonstrate that DDFTCN achieve an average accuracy of 93.6% across both datasets, surpassing the existing prediction methods. The experimental results on the incremental RUL prediction tasks validate the superiority of KLN over the typical and state-of-the-art IL methods.
- 27 September 2026
- Published by Elsevier in Engineering Applications of Artificial Intelligence
- Vol. 184
Prior distribution-guided disentanglement autoencoder for abnormal detection under variable operating conditions
- 27 September 2026
- Published by Elsevier in Engineering Applications of Artificial Intelligence
- Vol. 184
Prior distribution-guided disentanglement autoencoder for abnormal detection under variable operating conditions
- Top Cited Papers
- 5 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Transportation Electrification
- Vol. PP, 1
Source free progressive subspace calibration for transportation equipment diagnosis under time varying conditions
Plenty of source-free domain adaptation (SFDA)-based diagnosis approaches have been explored to tackle distribution shift and data privacy protection. Nevertheless, these approaches do not consider the uncertainty and large distribution characteristic of target predictions in time-varying conditions. Hence, their effectiveness is only suitable for scenarios with stable data distribution. Therefore, a new SFDA paradigm named source-free progressive subspace calibration (SFPSC) is developed to enhance model adaptation in complex dynamic data structures. In SFPSC, the CAS-Softmax is first constructed to suppress the overfitting of the source model, improving the mismatch of target samples. A new progressive pseudo-label selection strategy is designed for the training phase of model adaptation to gradually leverage easy-to-learn samples and guide the self-correction of hard-to-learn ones. Moreover, the scale robust subspace calibration is proposed to explicitly reduce the cross-domain distribution discrepancy. Finally, Extensive experiments on a laboratory gearbox and an actual equipment bearing validate the effectiveness and advantage of the proposed SFPSC.
- 5 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Transportation Electrification
- Vol. PP, 1
Top Cited PapersSource free progressive subspace calibration for transportation equipment diagnosis under time varying conditions
Plenty of source-free domain adaptation (SFDA)-based diagnosis approaches have been explored to tackle distribution shift and data privacy protection. Nevertheless, these approaches do not consider the uncertainty and large distribution characteristic of target predictions in time-varying conditions. Hence, their effectiveness is only suitable for scenarios with stable data distribution. Therefore, a new SFDA paradigm named source-free progressive subspace calibration (SFPSC) is developed to enhance model adaptation in complex dynamic data structures. In SFPSC, the CAS-Softmax is first constructed to suppress the overfitting of the source model, improving the mismatch of target samples. A new progressive pseudo-label selection strategy is designed for the training phase of model adaptation to gradually leverage easy-to-learn samples and guide the self-correction of hard-to-learn ones. Moreover, the scale robust subspace calibration is proposed to explicitly reduce the cross-domain distribution discrepancy. Finally, Extensive experiments on a laboratory gearbox and an actual equipment bearing validate the effectiveness and advantage of the proposed SFPSC.
- Top Cited Papers
- 22 January 2026
- Published by Elsevier in Reliability Engineering & System Safety
- Vol. 271
Dimension-mismatched adversarial network: a new feature distribution adaptation method for rolling bearing RUL prediction
- 22 January 2026
- Published by Elsevier in Reliability Engineering & System Safety
- Vol. 271
Top Cited PapersDimension-mismatched adversarial network: a new feature distribution adaptation method for rolling bearing RUL prediction
- Top Cited Papers
- 1 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 74 (00189456) , 1-19
Remaining Useful Life Prediction Methodologies With Health Indicator Dependence for Rotating Machinery: A Comprehensive Review
Facilitated by artificial intelligence technologies such as deep learning, predictive maintenance (PdM) has emerged as the most cost-effective maintenance approach for significantly enhancing the safety and availability of industrial equipment. Remaining useful life (RUL) prediction of rotating machinery is one of the core tasks of PdM, which can assist in maintenance decision-making and optimize asset allocation. Among them, the prognostic approach with health indicator (HI) as an intermediate bridge has attracted increasing attention due to its advantages in combining health condition monitoring and RUL prediction capabilities. Unfortunately, there has not been a comprehensive review to systematically analyze and summarize the recent advances in HI-dependent prognostic approaches. To fill this gap, this paper comprehensively reviews the RUL prediction methods with HI dependence for rotating machinery to help researchers quickly outline the current research status and state-of-the-art methods. Firstly, several run-to-failure datasets used to validate the performance of prognostic algorithms are briefly introduced. Secondly, three types of HI-based prognosis methods for rotating machinery, including similarity-based prognostic methods, model and data-based statistical methods, and intelligent time series prediction methods, are analyzed comprehensively. Finally, the current main challenges and prospects are presented to conclude this paper. It is anticipated that the target audience of this paper will be practitioners and researchers involved in the field of machinery PdM, providing them with the necessary basic knowledge and technical overview.
- 1 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 74 (00189456) , 1-19
Top Cited PapersRemaining Useful Life Prediction Methodologies With Health Indicator Dependence for Rotating Machinery: A Comprehensive Review
Facilitated by artificial intelligence technologies such as deep learning, predictive maintenance (PdM) has emerged as the most cost-effective maintenance approach for significantly enhancing the safety and availability of industrial equipment. Remaining useful life (RUL) prediction of rotating machinery is one of the core tasks of PdM, which can assist in maintenance decision-making and optimize asset allocation. Among them, the prognostic approach with health indicator (HI) as an intermediate bridge has attracted increasing attention due to its advantages in combining health condition monitoring and RUL prediction capabilities. Unfortunately, there has not been a comprehensive review to systematically analyze and summarize the recent advances in HI-dependent prognostic approaches. To fill this gap, this paper comprehensively reviews the RUL prediction methods with HI dependence for rotating machinery to help researchers quickly outline the current research status and state-of-the-art methods. Firstly, several run-to-failure datasets used to validate the performance of prognostic algorithms are briefly introduced. Secondly, three types of HI-based prognosis methods for rotating machinery, including similarity-based prognostic methods, model and data-based statistical methods, and intelligent time series prediction methods, are analyzed comprehensively. Finally, the current main challenges and prospects are presented to conclude this paper. It is anticipated that the target audience of this paper will be practitioners and researchers involved in the field of machinery PdM, providing them with the necessary basic knowledge and technical overview.
- Open Access
Semantic Ads Retrieval at Walmart eCommerce with Language Models Progressively Trained on Multiple Knowledge Domains
Sponsored search in e-commerce poses several unique and complex challenges. These challenges stem from factors such as the asymmetric language structure between search queries and product names, the inherent ambiguity in user search intent, and the vast volume of sparse and imbalanced search corpus data. The role of the retrieval component within a sponsored search system is pivotal, serving as the initial step that directly affects the subsequent ranking and bidding systems. In this paper, we present an end-to-end solution tailored to optimize the ads retrieval system on Walmart.com. Our approach is to pretrain the BERT-like classification model with product category information, enhancing the model's understanding of Walmart product semantics. Second, we design a two-tower Siamese Network structure for embedding structures to augment training efficiency. Third, we introduce a Human-in-the-loop Progressive Fusion Training method to ensure robust model performance. Our results demonstrate the effectiveness of this pipeline. It enhances the search relevance metric by up to 16% compared to a baseline DSSM-based model. Moreover, our large-scale online A/B testing demonstrates that our approach surpasses the ad revenue of the existing production model.
Open AccessSemantic Ads Retrieval at Walmart eCommerce with Language Models Progressively Trained on Multiple Knowledge Domains
Sponsored search in e-commerce poses several unique and complex challenges. These challenges stem from factors such as the asymmetric language structure between search queries and product names, the inherent ambiguity in user search intent, and the vast volume of sparse and imbalanced search corpus data. The role of the retrieval component within a sponsored search system is pivotal, serving as the initial step that directly affects the subsequent ranking and bidding systems. In this paper, we present an end-to-end solution tailored to optimize the ads retrieval system on Walmart.com. Our approach is to pretrain the BERT-like classification model with product category information, enhancing the model's understanding of Walmart product semantics. Second, we design a two-tower Siamese Network structure for embedding structures to augment training efficiency. Third, we introduce a Human-in-the-loop Progressive Fusion Training method to ensure robust model performance. Our results demonstrate the effectiveness of this pipeline. It enhances the search relevance metric by up to 16% compared to a baseline DSSM-based model. Moreover, our large-scale online A/B testing demonstrates that our approach surpasses the ad revenue of the existing production model.
- 13 January 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 74 (00189456) , 1-12
Unsupervised Health Indicator Fusing Time and Frequency Domain Information and Its Application to Remaining Useful Life Prediction
The prediction of component remaining useful life (RUL) is essential in making an appropriate maintenance plan for equipment. Constructing a reliable health indicator (HI) is crucial for RUL prediction. HI can be generated by quantifying distribution discrepancies. Most existing methods construct HIs based on the time domain, whereas in certain cases, time-domain data contain fewer degradation characteristics than frequency-domain data. To enhance the applicability and quality of HIs under different conditions, this paper presents a novel unsupervised approach for generating HI from both the time and frequency domains. Considering the frequency-domain data characteristics of mechanical vibration signals, an exponential mixture model is first applied to extract the frequency-domain distribution characteristics. Furthermore, a Gaussian mixture model is used to mine time-domain distribution characteristics. Subsequently, a distribution contact ratio metric is employed to respectively generate the time and frequency domain HIs by quantifying the discrepancies between baseline distribution and data distributions at different degradation moments. The final HI is constructed by weighting the time-domain and frequency-domain HIs. RUL prediction is achieved using the proposed HI and a variant of recurrent neural network. Lastly, the efficiency and superiority of this approach are validated using multiple gear life-cycle datasets, and the presented HI exhibits a higher RUL prediction accuracy than classical and advanced unsupervised HIs.
- 13 January 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 74 (00189456) , 1-12
Unsupervised Health Indicator Fusing Time and Frequency Domain Information and Its Application to Remaining Useful Life Prediction
The prediction of component remaining useful life (RUL) is essential in making an appropriate maintenance plan for equipment. Constructing a reliable health indicator (HI) is crucial for RUL prediction. HI can be generated by quantifying distribution discrepancies. Most existing methods construct HIs based on the time domain, whereas in certain cases, time-domain data contain fewer degradation characteristics than frequency-domain data. To enhance the applicability and quality of HIs under different conditions, this paper presents a novel unsupervised approach for generating HI from both the time and frequency domains. Considering the frequency-domain data characteristics of mechanical vibration signals, an exponential mixture model is first applied to extract the frequency-domain distribution characteristics. Furthermore, a Gaussian mixture model is used to mine time-domain distribution characteristics. Subsequently, a distribution contact ratio metric is employed to respectively generate the time and frequency domain HIs by quantifying the discrepancies between baseline distribution and data distributions at different degradation moments. The final HI is constructed by weighting the time-domain and frequency-domain HIs. RUL prediction is achieved using the proposed HI and a variant of recurrent neural network. Lastly, the efficiency and superiority of this approach are validated using multiple gear life-cycle datasets, and the presented HI exhibits a higher RUL prediction accuracy than classical and advanced unsupervised HIs.