Study On Feature Extraction And Analysis Of Ferrography Wear Particles Based On Object Detection

Ferrography is an oil analysis technology that separates,detects and analyzes wear particles from lubricating oil or hydraulic oil,which can monitor and diagnose the wear status of equipment.Wear particle image analysis is the key to ferrographic analysis.In view of the shortcomings of traditional methods of wear particle identification and analysis,this paper takes the ferrogram prepared by the rotary ferrograph as the research object,and adopts object detection to realize the automatic recognition and feature extraction analysis of ferrography wear particles,which improves the efficiency of ferrography.Combining the working principle of the rotary ferrograph and the law of wear particle deposition,the deposition behavior of ferrography wear particles is studied.The magnetic field distribution of the magnetic head is simulated and analyzed by using COMSOL Multiphysics software.Combined with the force analysis,the equal probability deposition law of wear particles on the rotary ferrogram is explored,and the distribution characteristics of wear particles on the ferrogram are analyzed.Based on the wear debris group theory,a practical wear debris group suitable for the rotary ferrogram is constructed,and the group characteristic analysis area of wear particles on the whole ferrogram is determined,and the field of view required for the whole ferrogram image shooting is determined.According to the classification of wear particles,the images of wear particles are sorted,and the five abnormal wear particles to be detected in this paper are determined,and then a training data set suitable for wear particle detection is produced by using image augmentation technology.According to the characteristics of the wear particle image,the Faster R-CNN object detection algorithm with outstanding comprehensive effect is selected as the basic framework,and WP-Faster R-CNN object detection model suitable for ferrography wear particle detection is constructed by using MATLAB deep learning toolbox.Based on transfer learning,super parameters are optimized from training method,initial learning rate adjustment,anchor frame size modification and other aspects to improve the performance of the model.Finally,the experimental results are compared and analyzed,which proved that the WP-Faster R-CNN model constructed in this paper can detect and identify abnormal wear particles in various wear particle images better,and the detection speed is also faster.Based on the K-means color clustering analysis method,the background of the wear particle image is removed,and the secondary traversal region labeling method is used to segment the removed background wear particle image,and finally a single wear particle image is obtained.According to the idea of the wear particle group feature analysis method and the saliency characteristics of the wear particle image,the image group characteristic parameters of wear particles to be extracted in this paper are defined.The extracted image group characteristic parameters of wear particle are processed,and the membership functions of the characteristic parameters of each wear particle in different wear stages are determined by using the two-level fuzzy comprehensive evaluation method,establish a description system for image group feature parameters of wear particles,and finally realize the quantitative processing of ferrography qualitative analysis.A ferrography wear particle detection and feature analysis system is designed by using App Designer toolbox of MATLAB.The working condition simulation test shows that,the system can detect and recognize the abnormal wear particles in the wear particle image by using the trained object detection model,and realize the wear condition discrimination of equipment based on the established description system for image group feature parameters of wear particles.This paper has 66 figures,21 tables and 99 references.