Lei He

The intelligent identification of wear particles in ferrography is a critical bottleneck that hampers the development and widespread adoption of ferrography technology. To address challenges such as false detection, missed detection of small wear particles, difficulty in distinguishing overlapping and similar abrasions, and handling complex image backgrounds, this paper proposes an algorithm called TCBGY-Net for detecting wear particles in ferrography images. The proposed TCBGY-Net uses YOLOv5s as the backbone network, which is enhanced with several advanced modules to improve detection performance. Firstly, we integrate a Transformer module based on the self-attention mechanism with the C3 module at the end of the backbone network to form a C3TR module. This integration enhances the global feature extraction capability of the backbone network and improves its ability to detect small target wear particles. Secondly, we introduce the convolutional block attention module (CBAM) into the neck network to enhance salience for detecting wear particles while suppressing irrelevant information interference. Furthermore, multi-scale feature maps extracted by the backbone network are fed into the bidirectional feature pyramid network (BiFPN) for feature fusion to enhance the model’s ability to detect wear particle feature maps at different scales. Lastly, Ghost modules are introduced into both the backbone network and the neck network to reduce their complexity and improve detection speed. Experimental results demonstrate that TCBGY-Net achieves outstanding precision in detecting wear particles against complex backgrounds, with a mAP@0.5 value of 98.3%, which is a 10.2% improvement over YOLOv5s. In addition, we conducted comprehensive ablation experiments, to validate the contribution of each module and the robustness of our model. TCBGY-Net also outperforms most current mainstream algorithms in terms of detection speed, with up to 89.2 FPS capability, thus providing favorable conditions for subsequent real-time online monitoring of changes in wear particles and fault diagnosis in ship power systems.

The wear particle classification algorithm proposed is based on an integrated ResNet50 and Vision Transformer, aiming to address the problems of a complex background, overlapping and similar characteristics of wear particles, low classification accuracy, and the difficult identification of small target wear particles in the region. Firstly, an ESRGAN algorithm is used to improve image resolution, and then the Separable Vision Transformer (SepViT) is introduced to replace ViT. The ResNet50-SepViT model (SV-ERnet) is integrated by combining the ResNet50 network with SepViT through weighted soft voting, enabling the intelligent identification of wear particles through transfer learning. Finally, in order to reveal the action mechanism of SepViT, the different abrasive characteristics extracted by the SepViT model are visually explained using the Grad-CAM visualization method. The experimental results show that the proposed integrated SV-ERnet has a high recognition rate and robustness, with an accuracy of 94.1% on the test set. This accuracy is 1.8%, 6.5%, 4.7%, 4.4%, and 6.8% higher than that of ResNet101, VGG16, MobileNetV2, AlexNet, and EfficientV1, respectively; furthermore, it was found that the optimal weighting factors are 0.5 and 0.5.

For mechanical equipment, the wear particle in the lubrication system during equipment operation can reflect the lubrication condition, wear mechanism, and severity of wear between equipment friction pairs. To solve the problems of false detection and missed detection of small, dense, and overlapping wear particles in the current ferrography wear particle detection model in a complex oil background environment, a new ferrography wear particle detection network, EYBNet, is proposed. Firstly, the MSRCR algorithm is used to enhance the contrast of wear particle images and reduce the interference of complex lubricant backgrounds. Secondly, under the framework of YOLOv5s, the accuracy of network detection is improved by introducing DWConv and the accuracy of the entire network is improved by optimizing the loss function of the detection network. Then, by adding an ECAM to the backbone network of YOLOv5s, the saliency of wear particles in the images is enhanced, and the feature expression ability of wear particles in the detection network is enhanced. Finally, the path aggregation network structure in YOLOv5s is replaced with a weighted BiFPN structure to achieve efficient bidirectional cross-scale connections and weighted feature fusion. The experimental results show that the average accuracy is increased by 4.46%, up to 91.3%, compared with YOLOv5s, and the detection speed is 50.5FPS.

The intelligent recognition technology for ferrography images is one of the important methods for diagnosis fault and state detection of machines. In allusion to these questions for the influences of wear particle images’ blurring, background intricacy, wear particle overlapping and lack of light, and others which lead to be the reason for the difficulty of achieving accurate identification, missed detection, and false detection, an intelligent recognition algorithm for ferrography wear particle based on convolutional block attention module (CBAM) and YOLOv5 is proposed. Firstly, it needs enhancement to improve contrast for ferrography wear particle images and lower background interference by adaptive histogram homogenization algorithm. Then, under the framework of YOLOv5 algorithm, the depthwise separable convolution is added to improve the detection speed of the network, and the detection accuracy of the entire network is improved by optimizing the loss function. Moreover, increase weight ratio on wear particle in images by adding a convolution block CBAM model and increase feature representative capability in detection network with YOLOv5 algorithm detection network, which can improve detection accuracy for wear particle. Finally, compare the algorithm with the three classical homologous series object detection algorithm. The experimental results show that the detection accuracy of the model can reach 96.7%, and the detection speed is 32 FPS for the images with a resolution of . It can be developed and applied to the fault diagnosis and condition monitoring of mechanical equipment.