Haisong Xu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Haisong Xu has a total of 8 co-authors. Here are the top co-authors by number of shared publications:
- Xiaolin Shi
College of Mechanical Engineering and Automation, Liaoning University of Technology, Jinzhou, China
3 shared publications Fan Yang
Institute of Intelligent Manufacturing, School of Mechanical Engineering, Northwestern Polytechnical University, Xi'an, China
2 shared publications- Han Zhang
College of Mechanical Engineering and Automation, Liaoning University of Technology, Jinzhou, China
2 shared publications - Yi Li
College of Mechanical Engineering and Automation, Liaoning University of Technology, Jinzhou, China
2 shared publications Han Zhang
School of Mechanical Engineering and Automation, Liaoning University of Technology, Jinzhou, China
1 shared publicationLiping Ma
Institute of Intelligent Manufacturing, School of Mechanical Engineering, Northwestern Polytechnical University, Xian, China
1 shared publication
Latest Publications
- Open Access
- 28 July 2025
- Published by Public Library of Science (PLoS) in PLOS ONE
- Vol. 20 (7) , e0328892
Small target detection of bearing surface defects based on improved YOLOv5s
In the context of industrial automation, the accurate detection of small defects on bearing surfaces (dents, bruise, scratch) is crucial for the safe operation of equipment. However, traditional detection methods have problems such as insufficient feature extraction for small targets and sensitivity to background interference. Based on this, an improved YOLOv5s small target detection method ECN-YOLOv5s for bearing surface defects is proposed. First, an efficient channel attention mechanism (ECA) is inserted in the layer before the SPPF in the backbone network, which realizes efficient computation of channel attention and reduces redundant computation while decreasing the model complication. Second, using the Content-Aware ReAssembly of Features (CARAFE) module instead of the original nearest-neighbor up-sampling module achieves light weighting while allowing for better aggregation of contextual information within a larger sensory field, which effectively improves the diversity and effectiveness of the model. Finally, an original loss function is replaced with Normalized Wasserstein Distance (NWD) to decrease the susceptibility to small object positional deviations. Experimental results in our homemade dataset indicate that the mean average precision (mAP) of ECN-YOLOv5s for bearing defect identification is 92.7%. This is an improvement of 2.4% versus the former YOLOv5s model. Compared with some mainstream detection models, the modified model in this study also provides better effects. Thus, the model is able to meet the requirements for the detection of small defects in bearings in industry.
- 28 July 2025
- Published by Public Library of Science (PLoS) in PLOS ONE
- Vol. 20 (7) , e0328892
Open AccessSmall target detection of bearing surface defects based on improved YOLOv5s
In the context of industrial automation, the accurate detection of small defects on bearing surfaces (dents, bruise, scratch) is crucial for the safe operation of equipment. However, traditional detection methods have problems such as insufficient feature extraction for small targets and sensitivity to background interference. Based on this, an improved YOLOv5s small target detection method ECN-YOLOv5s for bearing surface defects is proposed. First, an efficient channel attention mechanism (ECA) is inserted in the layer before the SPPF in the backbone network, which realizes efficient computation of channel attention and reduces redundant computation while decreasing the model complication. Second, using the Content-Aware ReAssembly of Features (CARAFE) module instead of the original nearest-neighbor up-sampling module achieves light weighting while allowing for better aggregation of contextual information within a larger sensory field, which effectively improves the diversity and effectiveness of the model. Finally, an original loss function is replaced with Normalized Wasserstein Distance (NWD) to decrease the susceptibility to small object positional deviations. Experimental results in our homemade dataset indicate that the mean average precision (mAP) of ECN-YOLOv5s for bearing defect identification is 92.7%. This is an improvement of 2.4% versus the former YOLOv5s model. Compared with some mainstream detection models, the modified model in this study also provides better effects. Thus, the model is able to meet the requirements for the detection of small defects in bearings in industry.
- Open Access
- 7 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 117210-117223
A Lightweight Method for Detecting Bearing Surface Defects Based on Deep Learning and Ontological Reasoning
Bearing as a critical component of machinery and equipment, its state of health directly affects the operating efficiency and security of the equipment. Therefore, the quality control of bearings must be very strict. Aiming at the different sizes and textures of the defect types on the surface of the outer ring of the bearing, and the fact that most of the target detection algorithms relying on deep learning show low speed and low precision in the detection of defects on the surface of the bearing, this paper presents a lightweight method for detecting the defects on the surface of the bearing based on deep learning and ontological reasoning. In this paper, YOLOv5s is chosen as the baseline model. First, the dynamic convolution is fused with the Ghost module and the combined structure is embedded into the C3 module, thus constructing a new module named C3-GhostDynamicConv (C3-GDConv) module, which achieves network lightweighting while maintaining efficient computation. Secondly, adding the lightweight attention mechanism SGE (Spatial Group-wise Enhancement) to the neck network almost eliminates the need to add additional parameters and computation. Without adding significant computational burden, it can significantly enhance the model’s performance and realize the efficient identification of bearing surface defects. Finally, a knowledge ontology of bearing defects is constructed by combining ontological reasoning techniques to give reasoning results on whether bearing defects can be repaired or not. The experimental results show that the frame rate per second (FPS) value of the improved model GDS-YOLOv5s for bearing defect recognition in this paper is 130, which is an improvement of 11% compared to the original YOLOv5s model. GFLOPs and the number of parameters decreased by 7% and 10% respectively, and mAP@0.5 reached 92.4%. The recognition accuracy is ensured while keeping the model lightweight. Combined with ontological reasoning, it is able to determine the repairability of defects, which provides a reliable basis for the maintenance decision of bearings. This study offers an efficient and intelligent scheme for bearing defect inspection and subsequent processes, which has important engineering application value.
- 7 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 117210-117223
Open AccessA Lightweight Method for Detecting Bearing Surface Defects Based on Deep Learning and Ontological Reasoning
Bearing as a critical component of machinery and equipment, its state of health directly affects the operating efficiency and security of the equipment. Therefore, the quality control of bearings must be very strict. Aiming at the different sizes and textures of the defect types on the surface of the outer ring of the bearing, and the fact that most of the target detection algorithms relying on deep learning show low speed and low precision in the detection of defects on the surface of the bearing, this paper presents a lightweight method for detecting the defects on the surface of the bearing based on deep learning and ontological reasoning. In this paper, YOLOv5s is chosen as the baseline model. First, the dynamic convolution is fused with the Ghost module and the combined structure is embedded into the C3 module, thus constructing a new module named C3-GhostDynamicConv (C3-GDConv) module, which achieves network lightweighting while maintaining efficient computation. Secondly, adding the lightweight attention mechanism SGE (Spatial Group-wise Enhancement) to the neck network almost eliminates the need to add additional parameters and computation. Without adding significant computational burden, it can significantly enhance the model’s performance and realize the efficient identification of bearing surface defects. Finally, a knowledge ontology of bearing defects is constructed by combining ontological reasoning techniques to give reasoning results on whether bearing defects can be repaired or not. The experimental results show that the frame rate per second (FPS) value of the improved model GDS-YOLOv5s for bearing defect recognition in this paper is 130, which is an improvement of 11% compared to the original YOLOv5s model. GFLOPs and the number of parameters decreased by 7% and 10% respectively, and mAP@0.5 reached 92.4%. The recognition accuracy is ensured while keeping the model lightweight. Combined with ontological reasoning, it is able to determine the repairability of defects, which provides a reliable basis for the maintenance decision of bearings. This study offers an efficient and intelligent scheme for bearing defect inspection and subsequent processes, which has important engineering application value.
- Open Access
- 15 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 70116-70127
Semantic Segmentation of Assembly Images Combining Deep Learning and Ontological Reasoning
With the advancement of industrial intelligence, semantic segmentation of assembly images is increasingly widely used in automated production and other fields. Aiming at the problem that traditional image segmentation methods often neglect the semantic association and structural information between objects in the image, this paper proposes a semantic segmentation method for assembly body images that combines deep learning and ontology reasoning. DeepLabV3+ is used as the base model, and MobileNetV2 backbone network and Convolutional Block Attention Module (CBAM) are introduced to accelerate the computational speed of the network and at the same time enhance the fusion ability of multi-scale features of the network to further improve the semantic segmentation accuracy. However, due to the poor interpretability and reliability of the data-driven deep learning method, an ontology module with inference capability is introduced and semantic rules are written based on the relevant knowledge of the domain to correct the misclassification of deep learning. Finally, we make the reducer model through Solidworks as the dataset for the experiment. The experimental results show that the improved network model improves 11.92% and 6.53% over the original network model in mIoU and mPA, respectively. The constructed ontology model of assembly information is then referenced and the misclassified parts are corrected by reasoning. The effectiveness of the semantic segmentation method proposed in this paper is verified.
- 15 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 70116-70127
Open AccessSemantic Segmentation of Assembly Images Combining Deep Learning and Ontological Reasoning
With the advancement of industrial intelligence, semantic segmentation of assembly images is increasingly widely used in automated production and other fields. Aiming at the problem that traditional image segmentation methods often neglect the semantic association and structural information between objects in the image, this paper proposes a semantic segmentation method for assembly body images that combines deep learning and ontology reasoning. DeepLabV3+ is used as the base model, and MobileNetV2 backbone network and Convolutional Block Attention Module (CBAM) are introduced to accelerate the computational speed of the network and at the same time enhance the fusion ability of multi-scale features of the network to further improve the semantic segmentation accuracy. However, due to the poor interpretability and reliability of the data-driven deep learning method, an ontology module with inference capability is introduced and semantic rules are written based on the relevant knowledge of the domain to correct the misclassification of deep learning. Finally, we make the reducer model through Solidworks as the dataset for the experiment. The experimental results show that the improved network model improves 11.92% and 6.53% over the original network model in mIoU and mPA, respectively. The constructed ontology model of assembly information is then referenced and the misclassified parts are corrected by reasoning. The effectiveness of the semantic segmentation method proposed in this paper is verified.