Integrity Detection System Of Hollow Fiber Membrane Component Based On Image Recognition

Hollow fiber column membrane module is a kind of water filtration module containing thousands of hollow fiber membrane filaments,which plays an important role in life and production.The national standard GB/T 36137-2018 has made qualitative requirements for the detection method of hollow fiber membrane integrity.The module is filled with the specified air pressure,and the integrity of the module can be judged by human eyes observing whether there are bubbles on the end face of the module.In the actual production,when the area of the leakage point(bubble)is 1‰～5‰of the end surface area of the column membrane,it needs to be repaired,otherwise the water quality will be affected.This paper designs a hollow fiber membrane module integrity detection system based on image recognition.The detection system uses ov5647 camera sensor at the speed of one frame per second to collect the image of the end face of the module in the 3-5min integrity detection experiment,and then detects whether there are bubbles in the currently collected image.If there are bubbles,the bubble contour range and center point coordinates are determined.The image detection targets of the system are the shell of the module,the film filament and the bubbles appearing on the end face when the module is incomplete.Because the shell and the film filament of the module are round,and the contour is obvious,which is quite different from the background,the Hough circle transform algorithm is used to detect them.Firstly,the image preprocessing is needed.In the preprocessing,the dark channel defogging processing,image graying,power-law transform,image processing are used Median filter and other algorithms remove the noise of the image,improve the contour characteristics of the target,and then extract the edge of the image.The algorithms used include gradient enhancement based on Sobel operator and edge detection based on Canny operator.Finally,the Hough circle transform algorithm is used to detect the center and radius of the module shell and membrane wire.The whole process is realized by OpenCV framework.Bubbles are transparent objects,and the edge characteristics are difficult to capture.In the deep learning object detection,the network feature layer of bubbles is extracted by using the network structure of YOLOv3,and the model that can recognize the contour of bubbles is trained,so that bubbles can be accurately identified in the experimental images continuously collected and the scenes of different end faces of components.In this paper,by changing the number of bubbles on the end of the way to collect multiple groups of test pictures,the final test results show that the integrity of hollow fiber membrane module judgment zero error,the relative error of bubble identification and membrane wire count statistics is kept below 2%.The integrity detection system designed in this paper quantifies the incomplete components by accurately and quickly identifying the image information,so as to provide automatic technical support for the follow-up membrane repair process.