Online monitoring of oil wear debris image based on CNN

Image monitoring of oil wear particles is currently only applicable to microflows and susceptible bubble interference. This paper develops an optical oil-monitoring system that can be used for large-diameter pipes with high flow rates. A shallow wide observation cell equivalent diameter Φ5 mm designed allow a theoretical maximum rate about 8 L/min, which significant improvement over current image generally less than Φ2 pipes. low-magnification (0.8 X – 5 ) stereoscopic microscope head improve the field view depth field, high-speed camera increase range monitored. set experimental platforms also constructed produce bubbles separately. Images are then collected subsequent training verification classification algorithms. motion object extraction algorithm based on background differences Otsu method extract debris images, convolutional neural network (CNN) distinguish between debris. Compared traditional morphological feature method, histogram oriented gradient (HOG) k -nearest neighbor (KNN) algorithm, support vector machine (SVM) CNN eliminates tedious process selection, has better results. The results show effectively collect particle images classify them, accuracy reach 91.8%.