📝 Selected Research Papers

Sudao He, Fuyang Chen, Hongtian Chen. A latent representation generalizing network for domain generalization in cross-scenario monitoring, IEEE Transactions on Neural Networks and Learning Systems. (JCR Q1, IF: 10.4) (cite)

• LRGN: A framework for Cross-scenario Monitoring without data of concerned events (such as faults, defect and physical intrusion) in a new scenario.
• Academic Impact: Efficient domain generalization with agnostic embedding space by estimating domain shifts of vibration signals by a sequential- variational generative adversarial network.
• Industry Impact: Verification based on real-field data from distributed optical fiber sensors in Jinliwen line and Nanjing Metro Line S7.

Sudao He, Fuyang Chen, Ning Xu, Hongtian Chen. Online monitoring for non-stationary operation via a collaborative neural network. IEEE Transactions on Instrumentation and Measurement, vol. 71, pp. 1-12, 2022. (JCR Q1, IF: 5.6)(cite)

• CoNN: A Collaborative Neural Network for non-stationary operation monitoring through online adaptation with theoretical guaranteed convergence.
• Academic Impact: Concept-sharing feature layers for adaptation to prior knowledge changes and concept-exclusive classification layers to learn concept-specific posterior knowledge.
• Industry Impact: CoNN is verified based on actual data from Nanjing Metro Line S7.

Sudao He, Fuyang Chen, Bin Jiang. Physical intrusion monitoring via local-global network and deep isolation forest based on heterogeneous signals. Neurocomputing, vol. 441, pp. 25-35, 2021 (JCR Q2, IF: 6)(cite)

• We propose a local–global semi-sharing network for heterogeneous signals in physical intrusion monitoring
• Academic Impact: A deep isolation forest is developed to learn from high-dimensional and extremely-imbalanced data.
• Industry Impact: An F1 score of 0.780 under a class ratio of 83.3 is achieved based on real-field data from Nanjing Metro Line S7.

Sudao He, Wai Kei Ao, Yi-Qing Ni. A Unified Label Noise-Tolerant Framework of Deep Learning-based Fault Diagnosis via A Bounded Neural Network, IEEE Transactions on Instrumentation and Measurement, 2024. (JCR Q1, IF: 5.6)(code)

• A unified framework for label-noise fault diagnosis of mechanical system.
• Academic Impact: Theoretical guarantee and enhancement of label noise tolerance.
• Industry Impact:
  • Applicable to different deep learning models.
  • Implementation of multiple variants and SOTA methods.
  • Real-field tests on automatic train control antenna beam of high-speed train.

Fuyang Chen, Sudao He, Yiwei Li, Hongtian Chen. Data-driven monitoring for distributed sensor networks: an end-to-end strategy based on collaborative learning, IEEE Sensors Journal, vol. 22, no. 22, pp. 21795-21805, 15 Nov.15, 2022. (JCR Q1, IF: 4.3)(cite)

• A collaborative soft-label network for label-noise learning in distributed sensor networks.
• Academic Impact: Introduce local similarities for modifying hard samples through a dual-space smoothing technique.
• Industry Impact: An accuracy of 88.33 is achieved when 40% of the labels are not correct using real-field data from Nanjing Metro Line S7. It can be extended to an unsupervised learning task.