| The development of online image visual ferrography technology provides an important way to monitor the wear status of mechanical equipment.The online visual ferrograph(OLVF)is used to capture the ferromagnetic wear particles in the lubricating oil,and the visual online ferrograph video images are used to capture the ferromagnetic wear particles in the lubricating oil.,can obtain information such as the shape and quantity of abrasive particles,and judge the wear condition of mechanical parts based on this.However,when processing and analyzing the collected video images of abrasive particles,it is found that due to the unavoidable presence of bubbles in the lubricating fluid,the information of wear particles in the video image is severely interfered by the bubbles and is difficult to extract,and it is difficult to extract in the case of high bubble interference.It is difficult to classify and segmentation abrasive particles,which affects the monitoring effect of equipment wear status.Therefore,the following studies are carried out to address the above problems:1.In order to obtain the online ferrography video data required in this article,a gear’s life cycle wear test was carried out.A gear accelerated wear test bench was used to monitor the gear wear in real time,and a bypass cycle was added to the gear box so that the lubricating oil in the gear box can enter the On-line visual ferrograph(OLVF)complete s the collection of online ferrograph video images.2.The collected online ferrograph video image has the characteristics of dense bubbles and fast moving speed,and the bubbles block the wear particles,which affects the information extraction and segmentation of the wear particles.First,the method of moving object detection is used to obtain the bubble position,and then the adjacent frame information is used to process it to realize the suppression of bubble interference in the online ferrograph video image.3.Based on the gray histogram distribution of the image after suppressing bubbles,the wear particle segmentation coefficient is selected,and the adaptive threshold value of each frame of image is determined to realize the rapid segmentation of the wear particle and the background.Then calculate the wear particle coverage index of the gear’s life cycle based on the collected images.In the actual application of the algorithm,the problem that the high-brightness abrasive grains are difficult to segment is optimized,and the accuracy of the wear particles segmentation is improved,so that the IPCA value is closer to the actual wear particles area.Draw a time series IPCA index curve for the experimental process,and briefly analyze the relationship between gear speed and load and abrasive coverage area index.4.Because massive wears and cutting wears only exist in the severe wear stage,there is an uneven distribution of various abrasive grains in the wear cycle of parts.Aiming at the problem of the small number of massive abrasive grains and cutting abrasive grains,the problem is increased.Wear particle image data,and produce 600 data set samples,build a high-efficiency convolutional neural network(FECNN)model,and complete the classification training of wear particles.Adding the Dropout operation after the convolutional layer to reduce the robustness of the model.The accuracy of the verification set reaches 92.26%,and use the FECNN model to classify the wear particle images under strong interference,proving that the model can achieve the classification of the wear particle images with strong interference... |
PDF Full Text Request