Research On Soap Bubble Air Tightness Detection Technology Based On Machine Vision

Airtightness is an important performance indicator for equipment such as high-pressure gas cylinders,heat exchangers,and refrigerators.If there is a leakage in the process of running this type of equipment,it could have a detrimental effect on the performance of the equipment,and in severe cases,it could result in significant safety incidents and property loss.It’s crucial to study and explore airtightness testing technology.There are two key issues with airtightness testing technology: the first is how to determine whether there is a leak or not,and the second is how to locate the source of the leakage point.Traditional detection methods including the soap bubble method are effective and can react in a short time,yet have high labour costs and high false alarm rates.Moreover,It’s difficult for them to locate leakage points.Emerging airtightness testing methods,such as element tracing method,ultrasound method,etc,possess high accuracy,fast response,and a high degree of automation,most of which support the location of leakage points.However,they require a high-standard testing environment and specialized knowledge,and the equipment involved costs a lot.With the development of artificial intelligence and computer science,machine vision technology breathe flesh life into the automation and intelligence of airtightness testing.This article takes industrial sealed equipment as the research object.With industrial sealed equipment as the research object,this article aims to combine the soap bubble method with machine vision to detect leak bubbles to determine the airtightness problem of the equipment.It solves the defects of manual detection in traditional soap bubble method,and achieves the fast and precise location of leakage points in accordance with the law of change of bubble centroid.The research work of this article is as follows:(1)A airtightness testing platform was designed and built.Taking industrial leakage components as templates,various simulated leakage devices with adjustable leakage rates and variable leakage point patterns were designed using machining techniques such as scratches and lasers.(2)The mechanism of bubble growth at leak points was studied.The present work established growth models and mechanical models for both single and multiple bubbles and summarized the laws of system pressure,bubble coverage area,and centroid variation during the process of bubble growth.Additionally,the components and characteristics of the "bubble liquid" were analyzed,and the composition of the foaming agent was optimized to improve the characteristics of different types of bubbles in the image.(3)The Dynamic object detection methods were studied.Given the contour missing and hole problems in traditional object detection algorithms,an improved Vi Be algorithm was proposed.Five-frame difference method,on the basis of the improved Canny operator,was used to help the Vi Be algorithm establish a background model.Implement adaptive threshold segmentation using the Otsu algorithm.Useing a secondary judgment mechanism to suppress ghosting by judging foreground pixels.The change rate of the foreground target centroid was introduced to adjust the fixed background update rate in the Vi Be algorithm.The foreground targets obtained by the fiveframe difference method and the improved Vi Be algorithm were subjected to logical operation and morphological operation to obtain a more complete foreground target,improving the defects of the naive Vi Be algorithm.(4)The bubble target extraction methods were studied.In order to eliminate the interference part of the foreground image,a KANN-DBSCAN algorithm based on an optimized circular cluster number was proposed.It achieved accurate separation of the interference cluster and the real bubble cluster,solving the problem that the traditional DBSCAN algorithm cannot choose parameters independently and the low operational efficiency of the naive KANN-DBSCAN algorithm.(5)The methods for for determining airtightness and locating leakage points were studied.The present work proposed a method for determining airtightness based on the variation of the coverage area of leakage bubbles,in order to distinguish whether the cluster was formed by leakage bubbles or not.Also,a method for locating leakage points,based on the variation of the leakage bubble centroid,was presented,relying on the characteristics of the change in centroid to locate.To address the problem of initial bubble image contours being missing,a method for improving the accuracy of leakage points localization was put forward,using an improved LSTM neural network to predict changes in centroid coordinates.