Yiming He
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Research Interests
Yiming He has a total of 35 co-authors. Here are the top co-authors by number of shared publications:
- Weiming Shen
Huazhong University of Science and Technology
12 shared publications - Xiaoxi Hu
Tsinghua University
8 shared publications - Yuhan Huang
State Key Laboratory of Intelligent Manufacturing Equipment and Technology, Huazhong University of Science and Technology, Wuhan, China
7 shared publications - Chengzhi Jiang
School of Information Technology and Artificial Intelligence, Zhejiang University of Finance and Economics, China
3 shared publications - Jingming Cao
School of Automation and Intelligence, Beijing Jiaotong University, Beijing, China
3 shared publications - Dandan Peng
Department of Electrical and Electronic Engineering, The Hong Kong Polytechnic University, Hong Kong, China
2 shared publications
Latest Publications
- 15 July 2026
- Published by SAGE Publications in Structural Health Monitoring
JMSANet: A lightweight network for industrial robot multi-joint anomaly detection
Joint anomaly detection from multi-joint signals is a key component in maintaining motion-control quality in industrial robots. In practical applications, this task requires models that achieve strong detection performance while remaining lightweight. However, existing methods often struggle to achieve a satisfactory balance between these two objectives. To address this issue, this paper proposes the Joint-aware Multi-Scale Attention Network (JMSANet), a lightweight architecture for joint-level anomaly detection. Specifically, JMSANet includes: (1) a joint position encoding (JPE) front-end that embeds joint identity into the stacked multi-joint input and performs pointwise projection to reduce inter-joint feature confusion; (2) a lightweight joint-aware hierarchical downsampling (LJHD) backbone built upon depthwise separable lightweight convolution blocks, which extracts multi-scale temporal representations via stage-wise downsampling with a compact footprint; and (3) a multi-scale aggregation (MSA) head that fuses multi-scale features through a lightweight multi-scale attention mechanism over global average pooling, enabling lightweight cross-scale integration. Experimental results on in-situ vibration data collected from an industrial robot show that JMSANet achieves competitive Macro-Precision, Macro-Recall, and Macro-F1 while maintaining a lightweight model footprint. Sensitivity analysis and ablation studies further verify the contributions of JPE, the hierarchical backbone, and the MSA head, supporting the effectiveness of the proposed design for the lightweight industrial robot anomaly detection task. The code is publicly available at https://github.com/Nashizumi45/JMSANet-A-Lightweight-Network-for-Industrial-Robot-Multi-Joint-Anomaly-Detection .
- 15 July 2026
- Published by SAGE Publications in Structural Health Monitoring
JMSANet: A lightweight network for industrial robot multi-joint anomaly detection
Joint anomaly detection from multi-joint signals is a key component in maintaining motion-control quality in industrial robots. In practical applications, this task requires models that achieve strong detection performance while remaining lightweight. However, existing methods often struggle to achieve a satisfactory balance between these two objectives. To address this issue, this paper proposes the Joint-aware Multi-Scale Attention Network (JMSANet), a lightweight architecture for joint-level anomaly detection. Specifically, JMSANet includes: (1) a joint position encoding (JPE) front-end that embeds joint identity into the stacked multi-joint input and performs pointwise projection to reduce inter-joint feature confusion; (2) a lightweight joint-aware hierarchical downsampling (LJHD) backbone built upon depthwise separable lightweight convolution blocks, which extracts multi-scale temporal representations via stage-wise downsampling with a compact footprint; and (3) a multi-scale aggregation (MSA) head that fuses multi-scale features through a lightweight multi-scale attention mechanism over global average pooling, enabling lightweight cross-scale integration. Experimental results on in-situ vibration data collected from an industrial robot show that JMSANet achieves competitive Macro-Precision, Macro-Recall, and Macro-F1 while maintaining a lightweight model footprint. Sensitivity analysis and ablation studies further verify the contributions of JPE, the hierarchical backbone, and the MSA head, supporting the effectiveness of the proposed design for the lightweight industrial robot anomaly detection task. The code is publicly available at https://github.com/Nashizumi45/JMSANet-A-Lightweight-Network-for-Industrial-Robot-Multi-Joint-Anomaly-Detection .
- Open Access
- 24 April 2026
- Published by Elsevier in Engineering
- Vol. 63, 26-30
Evolution of Agent Concept in Intelligent Manufacturing
- 24 April 2026
- Published by Elsevier in Engineering
- Vol. 63, 26-30
Open AccessEvolution of Agent Concept in Intelligent Manufacturing
- 17 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Transportation Electrification
- Vol. PP (4) , 1
TKS-TDM: A Temporal Key Point Selection Model for Efficient Fault Diagnosis of ZDJ9-RTA Using Multi-Sensor Signals
Railway Turnout Actuators (RTAs), especially the ZDJ9-RTA, are critical electromechanical devices in electrified railway systems whose failures can cause severe service disruptions. Accurate and efficient fault diagnosis is therefore essential for railway safety and reliability. Most existing RTA diagnosis methods rely on single-axis measurements, which limit spatial information capture and reduce robustness. For the first time in the electrified railway industry, we achieve RTA fault diagnosis using multi-sensor tri-axial vibration signals to fully exploit multi-directional responses. However, directly processing multi-sensor full-length sequences introduces substantial redundancy and computational burden, hindering real-time deployment. To address this, we propose an end-to-end Temporal Key Selection– Transformer Diagnostic Model (TKS-TDM) that balances diagnostic performance and efficiency by selecting a small set of fault-informative timestamps from multi-sensor signals and performing classification on these sparse representations. We validate TKS-TDM on two authoritative datasets from China Railway Signal & Communication Corp (Xi’an)—a professional Railway Test Line dataset and an in-service Field Deployment dataset with realistic noise. Compared with the strongest baseline, TKS-TDM improves accuracy by 1.60 and 0.89 percentage points on the Railway Test Line and Field Deployment datasets, respectively, while reducing computation from 16.76B to 0.17B FLOPs (a 98.99% reduction) and analyzing less than 0.5% of the temporal data, demonstrating a clear accuracy–efficiency gain for practical deployment. Generalization experiments across the two datasets further confirm the deployment potential of TKS-TDM across heterogeneous railway environments. Code is available at:https://github.com/huang-yu-han/TKS-TDM.git.
- 17 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Transportation Electrification
- Vol. PP (4) , 1
TKS-TDM: A Temporal Key Point Selection Model for Efficient Fault Diagnosis of ZDJ9-RTA Using Multi-Sensor Signals
Railway Turnout Actuators (RTAs), especially the ZDJ9-RTA, are critical electromechanical devices in electrified railway systems whose failures can cause severe service disruptions. Accurate and efficient fault diagnosis is therefore essential for railway safety and reliability. Most existing RTA diagnosis methods rely on single-axis measurements, which limit spatial information capture and reduce robustness. For the first time in the electrified railway industry, we achieve RTA fault diagnosis using multi-sensor tri-axial vibration signals to fully exploit multi-directional responses. However, directly processing multi-sensor full-length sequences introduces substantial redundancy and computational burden, hindering real-time deployment. To address this, we propose an end-to-end Temporal Key Selection– Transformer Diagnostic Model (TKS-TDM) that balances diagnostic performance and efficiency by selecting a small set of fault-informative timestamps from multi-sensor signals and performing classification on these sparse representations. We validate TKS-TDM on two authoritative datasets from China Railway Signal & Communication Corp (Xi’an)—a professional Railway Test Line dataset and an in-service Field Deployment dataset with realistic noise. Compared with the strongest baseline, TKS-TDM improves accuracy by 1.60 and 0.89 percentage points on the Railway Test Line and Field Deployment datasets, respectively, while reducing computation from 16.76B to 0.17B FLOPs (a 98.99% reduction) and analyzing less than 0.5% of the temporal data, demonstrating a clear accuracy–efficiency gain for practical deployment. Generalization experiments across the two datasets further confirm the deployment potential of TKS-TDM across heterogeneous railway environments. Code is available at:https://github.com/huang-yu-han/TKS-TDM.git.
- 1 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 26 (8) , 12398-12407
JFDNet: A Robust Industrial Robot Multijoint Fault Detection Network Under Sensor Fault and Environmental Noise Scenarios
Despite recent progress in deep learning for robot joint fault detection, most existing methods are trained on clean laboratory data and degrade significantly under real-world disturbances. These disturbances corrupt vibration signals and obscure fault-related cues, thereby reducing precision and recall. To address this problem, we propose the Joint Fault Detection Network (JFDNet), an end-to-end model for robust per-joint fault detection in multi-joint industrial robots under sensor fault and environmental noise scenarios. JFDNet features a Single-Joint Tri-Axis Wavelet (SJTAW) backbone that uses Trainable Discrete Wavelet Convolution (TDWC) and an Axis-Interaction Module (AIM) to suppress disturbances and preserve joint-specific fault patterns. A Dynamic Inter-Joint Relation (DIJR) neck further models cross-joint dependencies via a Bi-Directional Mamba Mixer (BDMM) flanked by attention modules. Extensive experiments demonstrate that JFDNet outperforms existing methods, delivering strong results across standard evaluation metrics while maintaining robustness. This work paves the way for deploying robot fault detection systems under sensor fault and environmental noise scenarios.
- 1 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 26 (8) , 12398-12407
JFDNet: A Robust Industrial Robot Multijoint Fault Detection Network Under Sensor Fault and Environmental Noise Scenarios
Despite recent progress in deep learning for robot joint fault detection, most existing methods are trained on clean laboratory data and degrade significantly under real-world disturbances. These disturbances corrupt vibration signals and obscure fault-related cues, thereby reducing precision and recall. To address this problem, we propose the Joint Fault Detection Network (JFDNet), an end-to-end model for robust per-joint fault detection in multi-joint industrial robots under sensor fault and environmental noise scenarios. JFDNet features a Single-Joint Tri-Axis Wavelet (SJTAW) backbone that uses Trainable Discrete Wavelet Convolution (TDWC) and an Axis-Interaction Module (AIM) to suppress disturbances and preserve joint-specific fault patterns. A Dynamic Inter-Joint Relation (DIJR) neck further models cross-joint dependencies via a Bi-Directional Mamba Mixer (BDMM) flanked by attention modules. Extensive experiments demonstrate that JFDNet outperforms existing methods, delivering strong results across standard evaluation metrics while maintaining robustness. This work paves the way for deploying robot fault detection systems under sensor fault and environmental noise scenarios.
- Top Cited Papers
- 27 February 2026
- Published by Elsevier in Neurocomputing
- Vol. 678
Strengthen the weak, align the strong: A federated enhanced iterative learning framework for cross-machine fault diagnosis
- 27 February 2026
- Published by Elsevier in Neurocomputing
- Vol. 678
Top Cited PapersStrengthen the weak, align the strong: A federated enhanced iterative learning framework for cross-machine fault diagnosis
- 4 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Reliability
- Vol. 75 (00189529) , 943-953
Heterogeneous Fault Diagnosis of Industrial Motors via Multimodel Collaborative Neural Architecture Search
Heterogeneous fault diagnosis scenarios are inevitable in industrial systems due to potential communication issues, sensor failures, and different downstream enterprise diagnostic capabilities. Traditional manual models with only a single type structure are difficult to handle heterogeneous inputs with different signal dimensions. This article proposes a multimodel collaborative neural architecture search framework for heterogeneous fault diagnosis. Unlike traditional neural architecture search that focuses on parameter-level fine-grained search, a model-level coarse-grained search space is adopted to fully utilize the existing knowledge and architectures, which greatly reduces search time and computational consumption. An avoidance-based diversity search strategy is specifically developed for this pattern to avoid repeated combinations of models and positions during the search process. The proposed automated multimodel design and optimization method not only endows intelligent diagnostic systems with universal processing capabilities for heterogeneous signals, but also enables self-evolution of the system without the need for time-consuming and labor-intensive manual design for multiple tasks. The proposed method was validated on nine tasks in a real industrial motor production line and achieved over 90% F1 scores.
- 4 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Reliability
- Vol. 75 (00189529) , 943-953
Heterogeneous Fault Diagnosis of Industrial Motors via Multimodel Collaborative Neural Architecture Search
Heterogeneous fault diagnosis scenarios are inevitable in industrial systems due to potential communication issues, sensor failures, and different downstream enterprise diagnostic capabilities. Traditional manual models with only a single type structure are difficult to handle heterogeneous inputs with different signal dimensions. This article proposes a multimodel collaborative neural architecture search framework for heterogeneous fault diagnosis. Unlike traditional neural architecture search that focuses on parameter-level fine-grained search, a model-level coarse-grained search space is adopted to fully utilize the existing knowledge and architectures, which greatly reduces search time and computational consumption. An avoidance-based diversity search strategy is specifically developed for this pattern to avoid repeated combinations of models and positions during the search process. The proposed automated multimodel design and optimization method not only endows intelligent diagnostic systems with universal processing capabilities for heterogeneous signals, but also enables self-evolution of the system without the need for time-consuming and labor-intensive manual design for multiple tasks. The proposed method was validated on nine tasks in a real industrial motor production line and achieved over 90% F1 scores.
- 12 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 75 (00189456) , 1-12
A Novel Portable Edge-Intelligent System for Cross-Individual Fault Diagnosis
Intelligent fault diagnosis has received widespread attention and is of great significance for the safety and stability of industrial systems. However, most current studies are limited to algorithm design and lack systematic deployment schemes, especially for design of intelligent diagnostic instruments. Besides, the current verification benchmark based on source device fault diagnosis (SDFD) ignores potential individual differences, resulting in degraded algorithm generalization for unseen individual devices and combinations of compound faults. This paper proposes a novel design scheme of portable edge-intelligent system including a portable intelligent diagnostic instrument (PIDI) and a novel refined trigonometric activation representation transformer (RTART). The designed PIDI not only inherits the functions of data acquisition and storage, but also conveniently supports real-time inference and updates of deep learning algorithms, as well as human-machine interaction capabilities. A novel RTART model is proposed to adapt to the characteristics of vibration signals through TrigAR, and enhance cross-individual feature generalization ability via multi-scale region pruning strategy (MSRP). Finally, the measurement and validation benchmark for intelligent fault diagnosis is rethought to establish cross-individual fault diagnosis (CIFD). The superiority of the proposed RTART is first validated on public datasets based on CIFD, achieving cross-individual accuracy (CIA) rates of 96.04%, which is superior to the compared advanced models. The designed PIDI embedded with RTART is deployed on real industrial production lines, achieving a CIA of 92.55%, further verifying the effectiveness and application potential of the proposed edgeintelligent system.
- 12 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 75 (00189456) , 1-12
A Novel Portable Edge-Intelligent System for Cross-Individual Fault Diagnosis
Intelligent fault diagnosis has received widespread attention and is of great significance for the safety and stability of industrial systems. However, most current studies are limited to algorithm design and lack systematic deployment schemes, especially for design of intelligent diagnostic instruments. Besides, the current verification benchmark based on source device fault diagnosis (SDFD) ignores potential individual differences, resulting in degraded algorithm generalization for unseen individual devices and combinations of compound faults. This paper proposes a novel design scheme of portable edge-intelligent system including a portable intelligent diagnostic instrument (PIDI) and a novel refined trigonometric activation representation transformer (RTART). The designed PIDI not only inherits the functions of data acquisition and storage, but also conveniently supports real-time inference and updates of deep learning algorithms, as well as human-machine interaction capabilities. A novel RTART model is proposed to adapt to the characteristics of vibration signals through TrigAR, and enhance cross-individual feature generalization ability via multi-scale region pruning strategy (MSRP). Finally, the measurement and validation benchmark for intelligent fault diagnosis is rethought to establish cross-individual fault diagnosis (CIFD). The superiority of the proposed RTART is first validated on public datasets based on CIFD, achieving cross-individual accuracy (CIA) rates of 96.04%, which is superior to the compared advanced models. The designed PIDI embedded with RTART is deployed on real industrial production lines, achieving a CIA of 92.55%, further verifying the effectiveness and application potential of the proposed edgeintelligent system.
- Top Cited Papers
- 12 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 13 (7) , 13005-13018
SMNet: A Novel Compositional Generalization Model for Industrial Robot Multijoint Fault Diagnosis
Compound fault diagnosis in multijoint industrial robots is a critical yet underexplored problem in industrial Internet of Things (IIoT), where the simultaneous degradation of multiple joints poses a severe challenge for reliable operation. Unlike conventional methods limited to single-fault scenarios, this article addresses the compositional generalization challenge—requiring models trained only on simple faults to accurately recognize unseen higher order fault compositions. To this end, we propose StateMix Network (SMNet), a multistage architecture that preserves atomic joint-level representations before compositional diagnosis. Specifically, a single-joint feature extraction (SJFE) backbone extracts clean joint-private features, which are then fused by an attention-guided dilated fusion (AGDF) neck employing parallel cascaded dilated convolution blocks (CDCBs) bracketed by a dual-path attention (DPA) mechanism for scale- and context-aware integration. Finally, a Mamba-based sequence mixer models long-range cross-joint dependencies to capture global fault dynamics. Extensive experiments on in situ vibration data from a single six-joint industrial robot platform, under a strict train-on-simple/evaluate-on-complex protocol, demonstrate that SMNet consistently outperforms representative baselines in macro-precision, recall, and $F1$ -score, particularly on unseen triple- and quadruple-joint compositions. Ablation and sensitivity analyses further validate the effectiveness of each module. This work presents a diagnostic approach that effectively generalizes from simple to complex fault scenarios in industrial robots.
- 12 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 13 (7) , 13005-13018
Top Cited PapersSMNet: A Novel Compositional Generalization Model for Industrial Robot Multijoint Fault Diagnosis
Compound fault diagnosis in multijoint industrial robots is a critical yet underexplored problem in industrial Internet of Things (IIoT), where the simultaneous degradation of multiple joints poses a severe challenge for reliable operation. Unlike conventional methods limited to single-fault scenarios, this article addresses the compositional generalization challenge—requiring models trained only on simple faults to accurately recognize unseen higher order fault compositions. To this end, we propose StateMix Network (SMNet), a multistage architecture that preserves atomic joint-level representations before compositional diagnosis. Specifically, a single-joint feature extraction (SJFE) backbone extracts clean joint-private features, which are then fused by an attention-guided dilated fusion (AGDF) neck employing parallel cascaded dilated convolution blocks (CDCBs) bracketed by a dual-path attention (DPA) mechanism for scale- and context-aware integration. Finally, a Mamba-based sequence mixer models long-range cross-joint dependencies to capture global fault dynamics. Extensive experiments on in situ vibration data from a single six-joint industrial robot platform, under a strict train-on-simple/evaluate-on-complex protocol, demonstrate that SMNet consistently outperforms representative baselines in macro-precision, recall, and $F1$ -score, particularly on unseen triple- and quadruple-joint compositions. Ablation and sensitivity analyses further validate the effectiveness of each module. This work presents a diagnostic approach that effectively generalizes from simple to complex fault scenarios in industrial robots.
- Open Access
- 1 January 2026
- Published by Institution of Engineering and Technology (IET) in IET Collaborative Intelligent Manufacturing
- Vol. 8 (1) , e70080
An LLM‐Enabled Human–Machine Collaborative Fault Diagnosis Framework via Industrial AI Agents
Intelligent fault diagnosis (FD) of mechanical equipment is vital for ensuring the safety and stability of industrial systems. As modern industrial environments continue to expand in scale and complexity, systematic FD increasingly requires not only accurate signal pattern recognition but also flexible human–machine collaboration for task configuration, model selection and decision optimisation. Traditional static diagnostic systems based on small‐scale specialised models (SSMs) are often inadequate for dynamic, multisource and heterogeneous diagnostic tasks. To address these challenges, this paper proposes a large language model (LLM)‐enabled human–machine collaborative fault diagnosis framework that combines multimodal task interpretation, signal‐source identification and dynamic selection of registered SSMs. The LLM operates at the task‐cognition level, whereas structured task checking, model applicability validation and human review for uncertain cases prevent uncertain cognitive outputs from directly triggering deterministic diagnosis. Experiments on two public datasets and an industrial spindle‐motor dataset show that the agent achieves 87.33% SSM selection accuracy and 75.00% end‐to‐end diagnosis accuracy without additional labelled samples for training a separate router, outperforming deterministic rule routing. These results indicate a feasible architecture for heterogeneous industrial diagnosis that combines flexible LLM‐based task understanding, deterministic SSM diagnosis and explicit human review for uncertain tasks.
- 1 January 2026
- Published by Institution of Engineering and Technology (IET) in IET Collaborative Intelligent Manufacturing
- Vol. 8 (1) , e70080
Open AccessAn LLM‐Enabled Human–Machine Collaborative Fault Diagnosis Framework via Industrial AI Agents
Intelligent fault diagnosis (FD) of mechanical equipment is vital for ensuring the safety and stability of industrial systems. As modern industrial environments continue to expand in scale and complexity, systematic FD increasingly requires not only accurate signal pattern recognition but also flexible human–machine collaboration for task configuration, model selection and decision optimisation. Traditional static diagnostic systems based on small‐scale specialised models (SSMs) are often inadequate for dynamic, multisource and heterogeneous diagnostic tasks. To address these challenges, this paper proposes a large language model (LLM)‐enabled human–machine collaborative fault diagnosis framework that combines multimodal task interpretation, signal‐source identification and dynamic selection of registered SSMs. The LLM operates at the task‐cognition level, whereas structured task checking, model applicability validation and human review for uncertain cases prevent uncertain cognitive outputs from directly triggering deterministic diagnosis. Experiments on two public datasets and an industrial spindle‐motor dataset show that the agent achieves 87.33% SSM selection accuracy and 75.00% end‐to‐end diagnosis accuracy without additional labelled samples for training a separate router, outperforming deterministic rule routing. These results indicate a feasible architecture for heterogeneous industrial diagnosis that combines flexible LLM‐based task understanding, deterministic SSM diagnosis and explicit human review for uncertain tasks.
- 16 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 74 (00189456) , 1-13
A Novel Method for Zero-Shot Compound Fault Diagnosis Under Low-Quality Annotation
Due to the measurement equipment and human errors, low-quality annotation in real-world industrial datasets is inevitable. Challenges such as sample imbalance, noisy labels and fault mode scarcity caused by low-quality annotation make zero-shot compound fault diagnosis (ZCFD) a challenging yet valuable task in industrial applications. This paper proposes a novel diagnostic framework to address ZCFD challenges under low-quality annotation. Specifically, a distribution-guided channel attention module is introduced to prevent the feature weight matrix from falling into local optima. A dilated convolutional network is employed as the backbone network for multi-scale feature fusion and fault decoupling. A unidirectional labeling bias guided cross-entropy loss function is designed to mitigate the negative impact of noisy labels during gradient computation. The proposed unidirectional labeling bias guided cross-entropy and distribution-guided channel attention dilated convolutional network (USCE-DCADCN) method has been evaluated on both the inter-shaft bearing fault dataset and the robot compound fault dataset to demonstrate its superior performance on ZCFD tasks under low-quality annotation.
- 16 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 74 (00189456) , 1-13
A Novel Method for Zero-Shot Compound Fault Diagnosis Under Low-Quality Annotation
Due to the measurement equipment and human errors, low-quality annotation in real-world industrial datasets is inevitable. Challenges such as sample imbalance, noisy labels and fault mode scarcity caused by low-quality annotation make zero-shot compound fault diagnosis (ZCFD) a challenging yet valuable task in industrial applications. This paper proposes a novel diagnostic framework to address ZCFD challenges under low-quality annotation. Specifically, a distribution-guided channel attention module is introduced to prevent the feature weight matrix from falling into local optima. A dilated convolutional network is employed as the backbone network for multi-scale feature fusion and fault decoupling. A unidirectional labeling bias guided cross-entropy loss function is designed to mitigate the negative impact of noisy labels during gradient computation. The proposed unidirectional labeling bias guided cross-entropy and distribution-guided channel attention dilated convolutional network (USCE-DCADCN) method has been evaluated on both the inter-shaft bearing fault dataset and the robot compound fault dataset to demonstrate its superior performance on ZCFD tasks under low-quality annotation.