Yongxiang Li

Yongxiang Li

Auto-generated profile, based on publicly available information

Yongxiang Li

Department of Industrial Engineering and Management, Shanghai Jiao Tong University

Research Focus & Metrics

Academic Career Length

21Years (2006-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
1 February 2021 in IEEE Transactions on Instrumentation and Measurement
Verified
Journal Article
Closed access
CSChangqing Shen
Changqing Shen
XWXu Wang
Xu Wang
DWDong Wang
Dong Wang
YLYongxiang Li
Yongxiang Li
JZJun Zhu
Jun Zhu
MGMingming Gong
Mingming Gong
An inconsistent distribution between training and testing data caused by complicated and changeable machine working conditions hinders wide applications of traditional deep learning for machine fault diagnosis. In a target domain, in which labeled samples are not available (testing data), transfer learning can adopt a relevant source domain (training data) to identify the similarity between the two domains and subsequently mitigate the negative effects of a domain shift. Previous studies on transfer learning mainly focused on decreasing the marginal distribution distance of two different domains or narrowing the conditional distribution distance even though marginal and conditional distributions provide different contributions to transfer tasks. The relative importance of the two distributions is difficult to dynamically and quantitatively assess. To align the two distributions (joint distribution) of two different domains, in this article, we propose a dynamic joint distribution alignment network (DJDAN) to evaluate the relative importance of marginal and conditional distributions dynamically and quantitatively. Furthermore, compared with common metrics that use pseudo labels to calculate the conditional distribution distance, the proposed DJDAN uses soft pseudo labels to more accurately measure the conditional distribution discrepancy between different domains. Extensive experiments reveal the superiority and generalization of the proposed DJDAN for bearing fault diagnosis under different working conditions.

22 May 2023 in IEEE Transactions on Neural Networks and Learning Systems
Verified
Journal Article
Closed access
RLRuixian Li
Ruixian Li
JWJianguo Wu
Jianguo Wu
YLYongxiang Li
Yongxiang Li
YCYao Cheng
Yao Cheng
Rolling bearings are critical components in modern mechanical systems and have been extensively equipped in various rotating machinery. However, their operating conditions are becoming increasingly complex due to diverse working requirements, dramatically increasing their failure risks. Worse still, the interference of strong background noises and the modulation of varying speed conditions make intelligent fault diagnosis very challenging for conventional methods with limited feature extraction capability. To this end, this study proposes a periodic convolutional neural network (PeriodNet), which is an intelligent end-to-end framework for bearing fault diagnosis. The proposed PeriodNet is constructed by inserting a periodic convolutional module (PeriodConv) before a backbone network. PeriodConv is developed based on the generalized short-time noise resist correlation (GeSTNRC) method, which can effectively capture features from noisy vibration signals collected under varying speed conditions. In PeriodConv, GeSTNRC is extended to the weighted version through deep learning (DL) techniques, whose parameters can be optimized during training. Two open-source datasets collected under constant and varying speed conditions are adopted to assess the proposed method. Case studies demonstrate that PeriodNet has excellent generalizability and is effective under varying speed conditions. Experiments adding noise interference further reveal that PeriodNet is highly robust in noisy environments.

26 May 2022 in IEEE Transactions on Automation Science and Engineering
Verified
Journal Article
Closed access
WFWei Fan
Wei Fan
ZCZhenqiang Chen
Zhenqiang Chen
YLYongxiang Li
Yongxiang Li
FZFeng Zhu
Feng Zhu
MXMin Xie
Min Xie
Condition monitoring plays a significant role in guaranteeing the reliability and safety of rotating machinery, which aims to detect an incipient fault and assess the degradation tendency. The construction of a health index (HI) is a crucial step to realize above tasks. At present, kurtosis, crest factor, and so on have been recognized as popular HIs to depict the operating condition. However, shortcomings of these classical HIs still exist: 1) classical HIs are prone to be affected by strong white Gaussian noise; 2) classical HIs are not sensitive to incipient faults. To deal with these two shortcomings, a reinforced noise resistant correlation method is proposed in this paper. Firstly, a new signal is constructed using the steps of segmenting and averaging to suppress the interference of noise. Then, a novel correlation function is used to get the hidden period. The proposed HI is constructed based on the discrete version of this correlation function to increase the sensitivity of incipient faults. Subsequently, theoretical values of the proposed HI under healthy states are investigated. The effectiveness of the method is demonstrated using simulated degradation processes and two accelerated degradation datasets of rolling element bearings. Through comparisons with other classical HIs, the proposed HI can simultaneously suppress the interference of strong noise and detect incipient faults. The comparison results identify the effectiveness of the proposed method in monitoring the condition of rotating machinery.

1 August 2022 in IEEE Transactions on Instrumentation and Measurement
Verified
Journal Article
Closed access
ZLZhen Li
Zhen Li
YLYongxiang Li
Yongxiang Li
XYXiaowei Yue
Xiaowei Yue
Enrico ZioEnrico Zio
Enrico Zio
JWJianguo Wu
Jianguo Wu
In complex systems, the operating units often suffer from multiple failure modes, and each failure mode results in distinct degradation path and service life. Thus, it is critical to perform the failure mode diagnostics and predict the remaining useful life (RUL) accordingly in modern industrial systems. However, most of the existing approaches consider the prognostic problem under a single failure mode or treat the failure mode classification and RUL prediction as two independent tasks, despite the fact that they are closely related and should be synergistically performed to enhance the generalization performance. Motivated by these issues, we propose a deep branched network (DBNet) for failure mode classification and RUL prediction. In this approach, the two tasks are jointly learned in a sequential manner, in which the feature extraction layers are shared by both tasks, while the neural network branches into individualized subnetworks for RUL prediction of each mode based on the output of the diagnostic subnetwork. Different from the traditional multitask learning-based methods, where the diagnostics and RUL prediction are performed in parallel, the proposed DBNet innovatively couples these two tasks sequentially to boost the prognostic accuracy. The effectiveness of the proposed method is thoroughly demonstrated and evaluated on an aircraft gas turbine engine with multiple failure modes.

20 January 2021 in IEEE Transactions on Instrumentation and Measurement
Verified
Journal Article
Closed access
WFWei Fan
Wei Fan
YLYongxiang Li
Yongxiang Li
LCLong Chen
Long Chen
ZXZhenying Xu
Zhenying Xu
This article proposes a novel period detection approach named B-spline constructed method for bearing fault detection by fully utilizing the sparsity of vibration signals. A newly constructed separated sparse representation model is proposed for vibration signals to reduce interference generated by strong background noises. In the model implementation, a B-spline dictionary is adopted to represent sparse transients due to its inherent ability to model sparsity and its high flexibility, and then, the sparse transients are estimated through split augmented Lagrangian shrinkage algorithm. A power function is proposed as a criterion for period detection based on the estimated transients. It is proven that its maximum value will be reached when the separation time parameter coincides with the theoretical period, and thus, the power function can be used to detect the fault characteristic period. The effectiveness of the proposed method is verified by both the simulated and practical vibration signals from faulty rolling bearings.

1 January 2021 in IEEE Transactions on Instrumentation and Measurement
Verified
Journal Article
Closed access
RLRuixian Li
Ruixian Li
LZLi Zhuang
Li Zhuang
YLYongxiang Li
Yongxiang Li
CSChangqing Shen
Changqing Shen
Bearing fault diagnosis plays an essential role in the maintenance of rotating machines in industries. Challenges in developing an effective and robust bearing fault diagnosis method include the complexity of vibration data and the external interference caused by the data collection. This study develops an intelligent data-driven method for bearing fault diagnosis in noisy environments, consisting of the feature transformation of vibration data and fault recognition based on transformed features. First, an extension of the Ramanujan filter banks (RFBs) method, scaled-RFB, is introduced to suppress the noises and convert original sequential vibration data into representative RGB images. Next, a strip convolutional neural network (strip-CNN) is developed with strip convolution to recognize the health condition of bearings based on the obtained RGB images. Two vibration datasets collected from Soochow University and a public data source are utilized to validate the effectiveness and robustness of the proposed method individually. Six levels of Gaussian noises are separately added into the two datasets to further demonstrate the performance of the proposed method in noisy environments. Compared with six benchmarking methods, the proposed method can achieve the best performance on bearing fault diagnosis in most scenarios and shows promising performance on datasets with a higher noise level. The average Precision, Recall, and ${F}1$ scores of the proposed method on both datasets are at least 51.79%, 52.49%, and 52.47% higher than that of benchmarking methods respectively when the signal-noise ratio (SNR) is −10 dB.