The Research On Vision-based Inspectors For Liquid Filling Line

The vision-based inspectors for liquid filling line are the vision based intelligent inspection equipments integrated of optical, mechanical and electronic functions. And they are automatic equipment in modern manufacturing industry. The intelligent inspectors can inspect empty bottle and liquid in bottle instead of human. So studying the intelligent inspector for filling line conforms to the requirements of develops of national economy and the market demand.This article firstly introduces the background and significance of the researches, the survey of machine vision technology, and its application in intelligent inspection. Then the bottle vision-based inspector and the liquid vision-based inspector for filling line are designed respectively according to the bottle inspection demands before liquid filling and the liquid inspection demands after liquid filling. The structures of these inspectors are explained in this article. The several designs of the inspection and control system are compared. The inspection and control system base on DSP is designed on the basis of the comparisons, which can realize high speed inspection. Because the image captured system is one of core system in inspectors, the crucial technology and equipments of image captured system are studied deeply. Then the image captured system which can provide clear image is developed. After inspection the fault production need be separated from production line. This article designs the flexible rejector, which can be sure to winkle out production stably from high speed product line. The inspection software is developed, which can accomplish inspection and control task on inspector. According to studies in this article, the prototypes of the vision-based inspectors for filling line have already been developed,which prove the validity of these designs.This paper mainly researches intelligent inspection methods which are used for liquid filling line. And the different methods which are used to inspect the bottle finish, bottom, wall and liquid in bottle are proposed based on using support vector machines mainly.The quality of bottle finish is important for filled production, so the finish inspection is necessary. Because the position of bottle in captured image need be located in online inspection, the location methods are firstly discussed. The paper presents the location methods based on modified Hough transform for bottle finish after comparing. The inspection method based on experiential rules algorithm is put forward for high speed inspection. This method inspects bottle finish with experiential rules according to the mean gray curve of finish which is got by rounded scan. The quality of bottle finish is judged quickly by this method. This method is concision, but it depends on manual rule and accurate rate of this method is not very high. So the support vector machines(SVM) whose generalization is good are used for finish inspection. But the experiments show the ability of SVM relates to the kernel function and its parameters. And then the inspection method based on support vector machine neural network is put forward to inspect bottle finish. The support vector machine neural network synthesizes the support vector machines and neural networks, which can be optimized effectively. The experiments prove the inspection result of this method is better.The bottle wall and bottom also need be inspected before liquid is filled. The location method based on center probability is presented for bottle wall. The position of region of interest of bottle bottom is confirmed by location methods based on modified Hough transform. After analyzing defection characteristic, the divisional inspection methods are presented. This method divides the region of interest into some smaller region for reducing the effect of noise, and afterward uses expert rules to detect if there are defect in these small regions. But because the defect may be also divided in this method, the accurate of inspection is not very satisfied. Then aiming at shortage of divisional inspection methods, the inspection method based multi-kernel support vector machines ensemble is put forward. The whole defection regions are segmented by modified watershed transform. And then this method can get the features of these regions, and uses the multi-kernel function support vector machines ensemble to classify these features. The multi-kernel function support vector machines ensemble uses ant colony optimization to optimized SVM in ensemble, which can use different kernel function, so its classification ability is good. The accurate rate of this inspection method is proved to be higher.There may be some impurities in liquid after filling, which may harm the consumer. Hence quality of liquid must be inspected. For distinguishing the impurities in liquid and marks on bottle, the motion analysis are applied to process image sequences. The impurities are shown as bright regions in images. Hence a binary image difference is presented. The images are firstly converted to binary image by the methods based on clustering, and the sequent binary images are subtracted to segment the motive regions. After that a matching algorithm is put forward to track these motive regions. The possible position of motive object in next frame is predicted by Kalman filter, and then the tracking window is set up according to this position. The motive regions in this tracing window are associated with the motive object to confirm the object which is matched for this track object, and the track chain is build at the same time. The region and motive features of object are exacted, and these features are classified by fuzzy support vector machines. The fuzzy support vector machines utilize fuzzy theory, which improves the ability of processing complex problem and anti-noise ability. So it can inspect liquid more exactly. At the same time the detection method of filling level is put forward to scale the liquid volume. The method which synthesizes the edge detection and edge linking is used to detect filling level. The filling level is found exactly and quickly by this method, and the high of filling level is calculated to detect liquid volume. The experiments show these methods are effective.