Kun Li
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Kun Li has a total of 10 co-authors. Here are the top co-authors by number of shared publications:
- Yulong Xing
Standards and Metrology Research Institute, China Academy of Railway Sciences Corporation Limited, Beijing 100081, China
2 shared publications - Zisheng Wang
College of Mechanical and Electrical Engineering, Central South University, Changsha 410083, China
2 shared publications - Cong Peng
School of Electronic and Information Engineering, Beijing Jiaotong University, Beijing, PR China
1 shared publication - Congcong Liu
Standards and Metrology Research Institute, China Academy of Railway Sciences Corporation Limited, Beijing 100081, China
1 shared publication - Jieming Zhao
Key Laboratory of Luminescence and Optical Information of Ministry of Education, Beijing Jiaotong University, Beijing 100044, China
1 shared publication - Qi Wang
Standards and Metrology Research Institute, China Academy of Railway Sciences Corporation Limited, Beijing 100081, China
1 shared publication
Latest Publications
- Open Access
- 22 September 2026
- Published by MDPI AG in Electronics
- Vol. 15 (19) , 4360
HIC-LLM: Health-Indicator Consistency Scoring and Optional Evidence-Grounded Large Language Model Reporting
Pack-level state-of-health assessment can conceal the cell that first reaches its lower-voltage limit and constrains a series-connected backup battery string. Cell-level localization is therefore required, but the scarcity of labeled field events and high-resolution telemetry does not support training or validating a supervised, unsupervised, or semi-supervised model. General-purpose language models introduce a separate risk of unsupported causal interpretation. These limitations motivate a two-stage framework: a training-free health-indicator consistency (HIC) score performs all numerical localization through within-event, one-sided median-absolute-deviation normalization, and an optional language model converts only the fixed ranking and structured evidence into an author-audited maintenance report with bounded mechanisms and explicit uncertainty. The primary evidence comprises two capacity-test reports from the same 24-cell, 2-V, 100-Ah pack: a manually terminated reference test and one lower-voltage-limit event. A separate CR-BTM workbook contributes 46 low-current pack-level records and 1104 cell-monitor records, but lacks a unique pack or device identifier, calendar date, capacity integration, and termination reason; it is therefore treated only as a candidate ancillary dataset with compatible measurement characteristics. In the lower-voltage-limit event, Cell #6 reached 1.800 V, released capacity was 86.11 Ah versus 100.46 Ah in the reference event, and HIC ranked Cell #6 first with a score of 7.983, compared with 1.145 for the runner-up. Cell #6 also had the largest lower-tail score at every recorded pre-termination time, including the record approximately 0.57 h before termination. Direct final-voltage and voltage-drop rules also rank Cell #6 first; consequently, the result is an engineering consistency verification of one field-confirmed event, not evidence of predictive superiority or population-level performance.
- 22 September 2026
- Published by MDPI AG in Electronics
- Vol. 15 (19) , 4360
Open AccessHIC-LLM: Health-Indicator Consistency Scoring and Optional Evidence-Grounded Large Language Model Reporting
Pack-level state-of-health assessment can conceal the cell that first reaches its lower-voltage limit and constrains a series-connected backup battery string. Cell-level localization is therefore required, but the scarcity of labeled field events and high-resolution telemetry does not support training or validating a supervised, unsupervised, or semi-supervised model. General-purpose language models introduce a separate risk of unsupported causal interpretation. These limitations motivate a two-stage framework: a training-free health-indicator consistency (HIC) score performs all numerical localization through within-event, one-sided median-absolute-deviation normalization, and an optional language model converts only the fixed ranking and structured evidence into an author-audited maintenance report with bounded mechanisms and explicit uncertainty. The primary evidence comprises two capacity-test reports from the same 24-cell, 2-V, 100-Ah pack: a manually terminated reference test and one lower-voltage-limit event. A separate CR-BTM workbook contributes 46 low-current pack-level records and 1104 cell-monitor records, but lacks a unique pack or device identifier, calendar date, capacity integration, and termination reason; it is therefore treated only as a candidate ancillary dataset with compatible measurement characteristics. In the lower-voltage-limit event, Cell #6 reached 1.800 V, released capacity was 86.11 Ah versus 100.46 Ah in the reference event, and HIC ranked Cell #6 first with a score of 7.983, compared with 1.145 for the runner-up. Cell #6 also had the largest lower-tail score at every recorded pre-termination time, including the record approximately 0.57 h before termination. Direct final-voltage and voltage-drop rules also rank Cell #6 first; consequently, the result is an engineering consistency verification of one field-confirmed event, not evidence of predictive superiority or population-level performance.
- Open Access
A Spiking Neural Network with Attention and Residual Mechanisms for Compound Fault Detection
To address the challenges of severe multi-source coupling, easily masked spiking features, and limited selection of key responses in compound fault signals, this paper proposes a compound fault detection method based on a spiking attention residual network (SARN). This method uses the original time-domain vibration signal as input and constructs an end-to-end spiking neural network framework. A hierarchical spiking attention module is designed to enhance multi-level spiking features from both temporal response and feature channel perspectives, thereby highlighting fault-sensitive information and suppressing redundant responses. Furthermore, a cross-layer spiking residual gating mechanism is introduced to mitigate effective information attenuation in spiking neural networks and improve the representation capability of weak fault features. Simultaneously, a multi-label detection strategy is employed to jointly identify multiple fault attributes, thereby improving the recognition rate of coupled compound fault modes. Verification results show that the proposed method achieves high performance in compound fault detection tasks, and compared with other popular methods, it exhibits better feature separability and detection stability.
Open AccessA Spiking Neural Network with Attention and Residual Mechanisms for Compound Fault Detection
To address the challenges of severe multi-source coupling, easily masked spiking features, and limited selection of key responses in compound fault signals, this paper proposes a compound fault detection method based on a spiking attention residual network (SARN). This method uses the original time-domain vibration signal as input and constructs an end-to-end spiking neural network framework. A hierarchical spiking attention module is designed to enhance multi-level spiking features from both temporal response and feature channel perspectives, thereby highlighting fault-sensitive information and suppressing redundant responses. Furthermore, a cross-layer spiking residual gating mechanism is introduced to mitigate effective information attenuation in spiking neural networks and improve the representation capability of weak fault features. Simultaneously, a multi-label detection strategy is employed to jointly identify multiple fault attributes, thereby improving the recognition rate of coupled compound fault modes. Verification results show that the proposed method achieves high performance in compound fault detection tasks, and compared with other popular methods, it exhibits better feature separability and detection stability.