METHOD FOR DETECTING DEFECTS IN THE HORIZONTAL JOINT OF THE MOLD FOR GLASS CONTAINERS

Abstract

The invention relates to a method for detecting defects in the horizontal mold joint (JH) of containers (R) on the finish (B). The method comprises the following steps: placing the container (R) between a light source and a camera; ensuring the rotation of the container (R) on its own axis; acquiring, with the aid of the camera, an image at each rotation increment of the container, such that the number of images per rotational revolution exceeds 36; and analyzing the captured images for each container so that: the profile of the finished edge is detected in each image; the finished edge profiles of the images are compared with a reference finished edge profile to detect deviations between these profiles; and a defect in the horizontal mold joint (JH) of a container is detected when at least one image of said container exhibits a deviation.

Description

METHOD FOR DETECTING DEFECTS IN THE HORIZONTAL JOINT OF THE MOLD FOR GLASS CONTAINERS Technical field of the invention The present invention relates to the technical field of inspection of objects, hollow articles or, in general, transparent or translucent containers such as, for example, glass bottles, jars or flasks. The object of the invention relates more specifically to the field of inspection of such glass containers to detect, in the finish of these containers, the presence of defects in the horizontal joint of the mold. Preliminary technique of the related application In general, a container has a bottom from which a vertical wall rises, terminating in a part called the end. The end is of different types depending on the closure system and includes, as illustrated in Figure 1, the annular S-end surface at the top and a CB-end at the bottom. Figure 1 represents a threaded end, which also includes threads on a cylindrical vertical portion. Glass containers are known to be manufactured using a forming machine called an IS machine, which consists of several independent forming sections fed with drops of malleable glass. These forming sections are equipped with at least one cutting cavity fitted with a preforming mold and the same number of finishing cavities, each of which receives a blow mold in which the containers take their final shape at high temperature. Conventionally, the finish is formed in the preforming mold. When the preformed item is transferred to the blow mold, the finish is already formed, and the preformed item is held in place by the finish. The finish is formed by the finishing molds, which consist of two half-molds that form the vertical wall of the finish and a mold piece called a ring that forms the surface of the finish or sealing surface. The preforming item mold also includes two body half-molds to form the body wall of the preformed item. During the press-and-blow forming process, a punch pushes the glass against the body half-molds. During the blow molding process, the punch is shorter and enters the finishing mold, but it is compressed air that pushes the glass against the body half-molds.Therefore, the exterior of the vertical wall of the finish is formed by two finishing halves, the top surface of the finish is formed by the ring, and the interior of the finish is formed by the punch. Consequently, each finish of vessel R includes, as illustrated in Figure 1, on both sides of its vertical wall, a vertical mold joint JV corresponding to the interface between the two finishing halves and a horizontal mold joint JH corresponding to the interface between the part of the finishing mold called the ring and the two finishing halves. These finishing mold joints are more or less marked or visible on the vessels, depending on the fit of the mold parts that deteriorate with use. It should be noted that the horizontal joint of the JH mold is located slightly below the surface of the S finish of the containers. Therefore, it is necessary to detect defects affecting the horizontal joint of the container mold in order to eliminate containers with defects that could affect their aesthetic appearance or, even more seriously, pose a real danger to the user. In the prior art, AGR INTERNATIONAL proposes, using a DSG400 laboratory monitoring machine that accurately measures the dimensional characteristics of containers, the possibility of detecting defects in the horizontal mold joint known as knockouts and flanges. This machine comprises, in particular, a light source located on one side of the container and a camera located on the other. During image capture, the container is rotated around its vertical axis. This machine does not appear to be designed to detect all defects in the horizontal mold joint. Furthermore, this machine has a limited monitoring rate, which does not allow for monitoring containers at the production rates typical of container manufacturing lines. WO 2013 / 128538 describes a method for detecting defects in the finish of glass vessels, specifically horizontal mold seam defects. This method involves rotating the vessel and projecting a vertical line of light onto the finish. A linear camera collects the light reflected by the finish. The method analyzes the reflected light to identify defects based on variations in the reflected light profile. While this method can only detect edges whose shape causes them to reflect light back towards the camera, it does not allow for quantifying the size of the defects or detecting different types of horizontal mold seam defects. However, it has a very high level of reliability for detecting small defects. Therefore, there is a need for a technique that can detect, at high speed, various defects in the horizontal mold seams of containers, while also offering a very high level of reliability in detecting small defects, typically less than one millimeter. Furthermore, it is important to avoid classifying containers as defective when their horizontal mold seams actually meet the required quality criteria. Brief description of the invention The object of the invention is to satisfy this need by proposing a method for detecting, at high speed, defects in the horizontal mold joint of glass containers. This method offers a very high level of reliability in detecting small defects in the horizontal mold joint, thus avoiding classifying as defective containers whose horizontal mold joint actually meets the required quality criteria. To achieve this objective, the invention proposes a method for detecting defects in the horizontal mold joint of glass containers, each with a vertical axis. The method comprises the following steps for detecting defects in each container: / re)nn / eznz / E / YiAi - placing the container between a light source and a camera to capture images whose optical axis of observation is substantially perpendicular to an axis parallel to the vertical axis of the container and whose field of view comprises at least the left edge of the finish or the right edge of the finish, including at least part of the horizontal joint of the mold; ensure the rotation of the container about itself along the vertical axis in accordance with at least one rotation; capture with the help of the camera, at each increment of rotation of the container, an image so that the number of images per rotational revolution is greater than 36; Analyze the captured images for each container, so that: The images define a finish inspection zone over a finish height that includes at least part of the horizontal mold joint; the edge profile of the finish is detected in the image inspection area; The edge profiles of the finish in the images are compared with a reference edge profile to detect deviations between these edge profiles and the reference edge profile; A defect in the horizontal joint of a container mold is detected when at least one image of said container shows a deviation. According to an advantageous variant of the modality, the camera is arranged in such a way that the field of view comprises at least part of the horizontal joint of the mold and a reference whose position at high altitude is known in relation to the horizontal joint of the mold and in which, in order to define the inspection zone of the finish in the images, the reference whose position at altitude is known in relation to the horizontal joint of the mold is sought, and the inspection zone of the finish is placed in relation to the reference in such a way that the inspection zone of the finish extends over a height that includes the horizontal joint of the mold. For example, in the images, the reference is sought for all or part of the surface of the finish or counter-finish. Typically, there is a camera to capture the images that has a horizontal field of view width between 5 mm and 130 mm and a field of view height between 3 mm and 20 mm. For example, there is an image capture camera that has a resolution greater than 25 pixels / mm. According to a preferred mode feature, the image capture camera is positioned so that its optical axis of observation is sensibly tangent to the left or right edge of the finish. The method according to the invention may use a camera provided with a telecentric lens. According to an advantageous modality feature, there is a light source having vertical and horizontal illumination dimensions between 100% and 200% and preferably 120% of the camera field of view dimensions multiplied by the distance between the light source / re ) nn / eznz / E / YiAi and the camera lens and divided by the distance between the axis of the containers and the camera lens. According to another modality, a telecentric light source is available whose illuminated field has dimensions greater than or equal to the dimensions of the camera's field of view. According to one modality variant, the edge profiles of the finish in the images are compared to a reference edge profile of the finish, using as a reference profile at least one profile obtained from at least one reference container considered to be compliant, to detect deviations. According to another variant of the modality, the edge profiles of the finish in the images are compared with a reference edge profile, using as a reference profile an average profile calculated over several rotation increments of at least one reference vessel considered to be compliant, in order to detect deviations. According to another variant of the mode, the edge profiles of the finish in the images are compared with a reference edge profile of the finish, using as a reference profile a finish edge profile to which a low-pass filter is applied, to detect deviations. According to another variant of the mode, the edge profiles of the finish in the images are compared with a reference edge profile of the finish, applying a high-pass filter to the edge profiles of the finish, to detect deviations. Advantageously, for each image, the edge profile of the finish is compared to the reference edge profile of the finish for at least several heights in the inspection zone, by comparing at least one measurement of the area, width and / or slope with a threshold, and a deviation is detected when at least one of these measurements exceeds that threshold. In general, the method consists of detecting at least one defect in the horizontal joint of the mold, which can be one of the following defects: external nodule, external flange, and displaced ring. According to a preferred modality, at least one detection criterion is defined for each of the following defects: namely external nodule, external rim, and displaced ring, and the method consists of distinguishing the detected defects among these three defects using at least one of these detection criteria. According to an implementation feature, the shape of the deviations detected for various heights in the inspection zone is chosen as the detection criterion, at least for one image. According to another feature of the implementation, the detection criterion is chosen, the angular extent of observation of the deviations corresponding to the number of successive images in which a deviation is detected. To distinguish the displaced ring defect from the nodule and external rim defects, the shape of the detected deviations is analyzed for various heights in the inspection zone. To distinguish the external nodule defect from the external rim defect, the angular extent of observation of the deviations is analyzed, considering that a nodule defect is detected when the angular extent of observation of the deviations is below a fixed maximum limit. ¡ re ) nn / eznz / E / YiAi To distinguish the external nodule defect from the external rim defect, the angular extent of observation of the deviations is analyzed, considering that an external rim defect is detected when the angular extent of observation of the deviations is above a fixed minimum limit. Conventionally, for each container with a defect in the horizontal mold joint, a signal is sent to eject the container from a production line. According to an advantageous feature of the mode, the container is rotated about itself along the vertical axis according to at least one rotational revolution for a maximum duration of 200 ms. Another object of the invention is to propose an inspection device that implements the method according to the invention to detect, in the finish of the glass container, defects in the horizontal joint of the mold. Brief description of the drawings Figure 1 is a large-scale view showing the finished glass vessel, the horizontal mold joint, and each vertical mold joint. Figure 2 is a schematic elevation view showing a device for implementing the method according to the invention, for detecting defects in the horizontal mold joint in the glass container finish. Figures 3A-3B are, respectively, side view of the finished edge and a top view of the finished edge of a glass container having a horizontal mold joint of the external nodule type. Figures 4A-4B are, respectively, side view of the finished edge and a top view of the finish of a glass container having an external flange type horizontal mold joint defect. Figures 5A-5B are, respectively, side plane views of the finished edge and a top plane view of the finish of a glass container having a horizontal mold joint defect of the offset ring type. Figure 6 is an example of an image of the right edge of a glass container finish with a horizontal mold joint defect. Figure 7 is a diagram that explains the extraction, in an image, of the edge profile of a container finish with a defect in the horizontal mold joint. Figure 8 is a diagram that explains the extraction, in an image, of the reference profile of the edge of a conforming container finish. Figure 9 is a diagram that explains the comparison between the edge profile of a vessel finish that has a horizontal mold joint defect of the external flange or external nodule type and the reference edge profile of a vessel finish. Figure 10 is a schematic explaining the comparison between the edge profile of a vessel finish that has a horizontal mold joint defect of the displaced ring type and the reference edge profile of a vessel finish. Figure 11 is a diagram that explains the extraction, in one image, of the reference profile of the finishing edge in the form of a straight line for a thread finish. Figure 12 is an example image of the right edge of a vessel finish with an external nodule defect, showing the vessel profile and a reference profile. Detailed description of the invention As seen more specifically in Figures 1 and 2, the object of the invention relates to a device 1 for implementing a method according to the invention for detecting defects in the horizontal mold joint JH in glass containers R. Conventionally, a container R has a bottom F from which a vertical cylindrical wall V rises along a vertical axis Z, terminating in a part called the end B. In the case of a bottle-type container R, the vertical cylindrical wall V extends from the bottom F, forming the body of the bottle, which is connected to a neck C via a shoulder E. Each vessel R includes, as illustrated in Figure 1, in particular a horizontal mold seal JH corresponding to the interface between the mold portion called the ring and the two mold halves that form the vertical cylindrical wall V of the vessel. By definition, the horizontal mold seal JH is located slightly below the finished surface S, which corresponds to the flat surface of the sealing vessel. The horizontal mold seal JH extends over the entire circumference of the finished surface. According to the invention, the method consists of detecting in each container a defect in the horizontal joint of the mold JH selected from the following defects: external nodule JHK known by the English name (knock out), external flange JHF also known by the English name (flange) and displaced ring JHO also known by its English name (overhang / overmatch). As shown more specifically in Figures 3A and 3B, an external nodule defect (JHK) corresponds to a glass burr at one of the T-shaped intersections of the horizontal mold joint (JH) and the vertical mold joint (JV) between the mold ring and the two mold halves that form the vertical wall. This defect corresponds to a glass burr created by blunt ridges at the interface of the three molds or by a poor fit between them. This external nodule (JHK) takes the form of a pointed or needle-like glass protrusion (often triangular in cross-section) with a small thickness along the vertical Z-axis and a small horizontal extension, often less than one millimeter. Consequently, it has a small circumferential angular extension (θ) taken around the vertical Z-axis. Typically, the external nodule (JHK) has a small circumferential angular extension (θ) of less than a few degrees. As can be seen more specifically in Figures 4A and 4B, a JHF external flange defect corresponds to excess glass resulting from a poor fit between the mold part called the ring and the two mold halves that form the vertical wall. This JHF external flange takes the form of a thin plate of glass along the vertical Z-axis. Furthermore, this JHF external flange has a significant angular extension θ taken around the vertical Z-axis. Typically, the JHF external flange has a significant angular extension θ greater than or equal to 10°, but very easily greater than 30°. It can be much wider, up to 180°. More rarely, the flange may be present around the entire circumference. As can be seen more specifically in Figures 5A and 5B, a JHO displaced ring defect corresponds to excess glass resulting from a displacement of the mold part called the ring and the mold halves that form the vertical wall. This JHO displaced ring takes the form of a very thick glass protrusion along the vertical Z-axis with a significant angular extent θ around the Z-axis. Typically, the ring displacement can occur over 180°, but the resulting step is generally significant over an angular extent θ greater than or equal to 90°. As will be explained in detail in the following description, the object of the invention is to detect a container having a defect in the horizontal mold joint JH selected from one or both of the following defects: external nodule JHK, external flange JHF, and displaced ring JHO. According to an advantageous embodiment, the object of the invention is to distinguish a defect in the horizontal mold joint JH selected from the defects of external nodule JHK, external flange JHF, and displaced ring JHO. To this end, the device 1 for implementing the method for detecting defects in the horizontal joint of mold JH includes a fixed light source 3 arranged on one side of the container R and a fixed image capture camera 4 arranged on the other side of the container R. This camera 4, which includes a lens 4a, is adapted to capture images in which at least part of the horizontal joint of mold JH of the containers R appears. This camera 4 is connected to an analysis and processing unit 5 configured to analyze the images taken and detect a defect in the horizontal joint of mold JH of the containers R. The imaging of a vessel R is performed while the vessel R rotates about its vertical axis Z for at least one rotation, allowing the entire horizontal joint of the mold JH to pass in front of the camera. Each vessel R is supported by a rotation system 6. For example, the rotation system 6 includes a sliding or positioning plane 7 for the bottom F of the vessel R, as well as a drive system 8 of all known types. The drive system 8 is piloted such that each vessel R remains between the light source 3 and the camera 4 for the time required to complete at least one revolution, during which the camera 4 captures images, as will be explained later in the description. For example, the rotation system 6 includes as a drive system a rotating wheel or plate 8 that drives the container into rotation by friction, while the container is secured supported on at least two free wheels or rollers 9. The rollers, for example, form part of a star-shaped transport wheel 10 that transports the containers R in a circular path, to support them successively and slide them onto a placement plane 7, in front of the detection device 1. Advantageously, this star-shaped transport wheel 10 forms part of an inspection machine, which inspects more than 150 containers per minute, including one or more stations for inspecting the containers in motion at the exit of a production line.In other words, the detection device 1 according to the invention can be implemented in addition to the inspection stations conventionally placed to inspect the containers on the production line, without reducing the operating rate of the containers in these stations that integrate a rotation of the containers for this inspection. According to a feature of the invention, the in-line inspection machine inspects between 50 and 500 containers per minute, typically between 150 and 450 containers per minute. For each container, the method includes a transport step consisting of carrying container R and placing it between the light source 3 and the image capture camera 4, and a step to ensure the rotation of container R about its own axis along the vertical Z-axis for at least one rotation, during which the images are acquired. It should be noted that with the star-shaped transport wheel 10, the previously inspected container is carried to the inspection station at the same time as a new container is introduced for inspection. Therefore, the in-line inspection machine is equipped with high-speed handling means. For a rate of 150 containers per minute, the duration of the rotation and inspection step is a maximum of 200 ms.According to a preferred feature of the invention, a complete rotation of a container takes a maximum of 100 ms. It is also commonly chosen that the container completes 1.5 rotations. In other words, the method according to the invention consists of placing the container R between a light source 3 and an image capture camera 4 whose optical axis of observation is substantially perpendicular to an axis parallel to the vertical Z axis of the container and whose field of view comprises at least the left edge of the finish or the right edge of the finish, including at least a part of the horizontal joint of the mold, ensuring the rotation of the container R about itself along the vertical Z axis according to at least one rotation in less than 200 ms, and acquiring by the camera 4, at each increment of rotation of the container, an image so that the number of images per rotational revolution is greater than 36. According to a feature of the invention, the camera 4 is controlled so that an image is captured at each rotation increment of the vessel, resulting in a number of images per rotation exceeding 36. In other words, the method according to the invention aims to acquire at least one image every 10° of rotation of the vessel R. For example, the number of images of a vessel R over 360° ranges from 36 to 96. The rotation increment of the vessel between each captured image represents an angular sector traversed by the finish, ranging, for example, from 10° to less than Γ. Now, a fast camera with an integration time of less than 500 ps and a readout time of 0.5 ms would allow the acquisition of 400 images per rotation in 200 ms. Increasing the acquisition frequency leads to greater accuracy, particularly in estimating the radial length of a nodule.This results in a higher system cost due to the price of a fast camera 4 and the computing power that the analysis and processing unit 5 will need. According to the invention, the light source 3 and the camera 4 are adapted so that the camera can acquire, during the rotation of the container about its vertical axis Z, images I in which at least a portion of the horizontal joint JH of the container R mold appears. As illustrated in Figure 2, the vertical axis Z of the container is conventionally considered to be parallel to the vertical z direction of an orthogonal x, y, z reference frame. This vertical z direction is perpendicular to the transverse y direction passing through the light source 3 and the camera 4, while the lateral x direction is perpendicular to both the transverse and vertical z directions. The container R is placed between the light source 3 and the camera 4, whose optical axis of observation Y extends parallel to the transverse direction y, i.e., along a direction substantially perpendicular to the vertical direction z. This camera 4 has a field of view that extends laterally along the lateral direction x, orthogonal to the vertical direction z and the optical axis of observation Y. The field of view of camera 4 thus extends in the plane defined by the vertical direction z and the lateral direction x. According to a preferred example, the camera 4 is positioned so that its optical axis is substantially tangent to the left or right edge of the finish. In accordance with the vertical z-direction, the field of view of camera 4 comprises at least the horizontal joint of mold JH. The field of view of camera 4 comprises at least the left edge of finish B or the right edge of finish B. According to a preferred embodiment, the field of view of camera 4 comprises, for good resolution, only the left edge of finish B or the right edge of finish B. Of course, the method according to the invention can be carried out with a camera whose field of view is not limited to a right or left edge. This preferred embodiment is particularly suitable for high speeds, given that in-line inspection machines rotate the containers at least once and that, therefore, observation of a single face is sufficient to ensure the detection of defects in the horizontal joint of mold JH of the finish.Furthermore, single-sided observation allows you to make the most of a camera's field of view and place its optical axis tangent to the observed finish edge. This camera 4, acting as a matrix camera, transmits to the analysis and processing unit 5, for each rotation increment, a horizontal projection of the finish. In this projection, at least one edge of the finish is highlighted, designated as the left edge or right edge based on its position in each image captured by the camera (Figure 6). The left or right designation is based on the observational viewpoint of each image, as a vessel of revolution does not strictly have a left or right side. In the example illustrated in Figure 6, image I corresponds to a right edge of vessel R. In each image captured by the camera, at least the left edge of finish B or the right edge of finish B appears, allowing visualization of at least part of the horizontal joint of mold JH and at least the profile P of the finish edge. In other words, as shown in Figure 6, each image I must include, at a height measured along a direction parallel to the vertical z direction, at least part of the contour or profile P of the finish edge, which includes at least part of the horizontal joint of mold JH. Therefore, in each image, an inspection zone Zl is defined, corresponding to a horizontal strip of the image with a height determined to include a portion of the profile P of the finish edge that includes a portion of the horizontal joint of mold JH. According to an advantageous feature of the implementation, camera 4 is positioned so that its field of view includes at least a portion of the horizontal mold joint JH and a reference point whose high-altitude position relative to the horizontal mold joint JH is known. Therefore, after locating this reference point in the images, it is possible to position the finish inspection zone Zl relative to the reference point so that the finish inspection zone extends over a height that undoubtedly includes the horizontal mold joint. However, this image capture is also performed with a margin to visualize this reference and part of the horizontal joint of the JH mold, taking into account the positioning dispersions of the R containers with respect to camera 4 and the manufacturing tolerances of the R containers. This margin is, for example, at least + / - 2 mm when the height of the containers can vary by + / - 2 mm. It should be understood that this reference point presented by the R containers is a notable element, such as a contour, an angle, or a shape that can be located in the images. This notable element is also visible in all images captured during the rotation of the item, and its vertical position in each image can be unambiguously determined. Its vertical distance Dv to the horizontal joint of the mold JH is known in each image, and therefore, this distance is preferably constant. The location of this reference point in the images allows a positional reference system to be assigned to each image, which can then be used in their analysis. Typically, the reference point can be, for example, all or part of the surface S of the top or bottom finish CB when a vessel includes one. When the top surface S is used as the reference point, camera 4 is adjusted so that the height of its field of view includes the top surface S plus a margin to account for positioning variations and manufacturing tolerances of the vessels. Similarly, camera 4 is adjusted so that the height of its field of view includes the horizontal mold joint JH plus a margin. The horizontal width of the field of view of camera 4 is adapted to include the profile P of the finish edge with a margin and the reference, i.e., part of the finish surface in the example considered. According to a preferred variant, the field width is chosen to include only a finish edge plus the necessary margins. For example, camera 4 has a horizontal field width preferably between 5 mm and 80 mm and a field height between 3 mm and 20 mm. This allows for optimization of the camera field, i.e., limiting unnecessary areas of the sensor. According to another variant, the two left and right edges of the finish are included in the field of view, and the field width can reach 130 mm to observe finishes with a diameter of 120 mm and a margin of 10 mm. For example, camera 4 is a matrix camera combined with an optical lens, which together allows the inspection area to be observed with a resolution greater than 25 pixels / mm. The light source 3 is made in any suitable way to ensure backlighting of the container R adapted to the image capture by camera 4. / re ) nn / eznz / E / YiAi According to a preferred feature of the modality, the light source 3 has defined vertical and horizontal illumination dimensions. Conventionally, the vertical dimension is taken along the vertical z-direction, while its horizontal dimension is parallel to the lateral x-direction. In a non-telecentric optical system, the dimension DL of the light source 3 is of the type: DL = CH x DI / DC; where CH is the field of view of camera 4, DI is the distance between the light source 3 and the camera lens 4a, and DC is the distance between the inspected area and the camera lens. More specifically, DC is considered to be the distance between the camera lens (e.g., the optical center of a non-telecentric lens) and the focal plane containing the vertical Z-axis of the container and the point of tangency with the finished edge of an optical beam passing through the optical center. The vertical and horizontal illumination dimensions are the dimensions of the usable area of ​​light source 3 observable by camera 4. According to a feature of the modality, the light source 3 has vertical and horizontal illumination dimensions between 100% and 200% and preferably 120% of the field of view dimensions of camera 4 multiplied by the distance between light source 3 and lens 4a of the camera and divided by the distance between the vertical Z axis of the containers and lens 4a of the camera. According to one variant, the light source 3 can be a telecentric light source whose illuminated field has dimensions greater than or equal to the dimensions of the camera's field of view 4. The light source 3 thus generates, for the camera's field of view, a beam whose mean rays are parallel to the camera's optical axis. In this case, it should be noted that the camera's lens 4a is a telecentric lens. The images captured by camera 4 for each container R are analyzed by the analysis and processing unit 5, which is configured to detect a defect in the horizontal mold joint JH of the containers R. The method for detecting a defect in the horizontal mold joint JH consists of analyzing, for each container, the captured images and, specifically within each image, the inspection zone Zl of the finish over a finish height that includes at least part of the horizontal mold joint. This analysis method involves detecting, within the image inspection zone, the profile P of the finished edge and comparing the finished edge profiles P in the images with a reference profile Pf of the finished edge to detect deviations between these finished edge profiles P and the reference profile Pf of the finished edge.A defect in the horizontal mold joint of a container is detected when at least one image of container R shows a deviation. For each container with a defect in the horizontal mold joint JH, the analysis and processing unit 5 sends a signal indicating the defective quality of the container. This signal can then control an ejector to remove the container from the production line. The detection of the P-profile of the finish edge in each of the captured images can be carried out using any appropriate image processing method. As explained previously, the image analysis is preferably limited to the inspection zone Zl of the finish where part of the finish profile P appears, including the horizontal mold seal. Advantageously, the inspection zone is positioned relative to a reference point on the vessel shown in the image, the high-altitude position of which is known with respect to the horizontal mold seal. This implementation variant limits the analysis of the finish profile to only the portion necessary to characterize the horizontal mold seal, while eliminating other parts of the profile that could be mistaken for a horizontal mold seal, such as those that include, for example, a thread. According to this method, in each image, the reference whose high-altitude position is known with respect to the horizontal joint of the mold is sought, and the finish inspection zone is positioned with respect to the reference in such a way that the finish inspection zone extends over a height that necessarily includes the horizontal joint of the mold JH. Figure 7 illustrates, by way of example, the detection of the profile P of the right edge of a finish in an image I where the finish surface S appears as a reference. The first step is to locate the position of the finish surface S vertically, i.e., along the vertical z-direction. For example, along a vertical line following the left edge of the image, a vertical gray-level transition corresponding to the finish surface S is detected. Therefore, it is possible to position the inspection zone Z1 from this reference so that the image inspection necessarily takes into account the horizontal mold joint JH. However, the reference search method may differ. For example, if the reference is the counter finish, a search for the maximum two-dimensional correlation between the image and a reference window containing a learned model of the counter finish or the visible right or left edge of the counter finish can be implemented. The second step involves extracting the outer contour or profile of the finishing edge P, for example, from the finished surface S to the lower limit of the inspection zone Z1, located below the finished surface and beyond the horizontal mold joint ZH. As illustrated in Figure 7, the profile of the finishing edge P includes the horizontal mold joint JH. Finding an initial profile point for the finishing edge P can involve scanning the image horizontally, at a given height relative to the reference, until a horizontal gray level transition is detected. The extraction of the edge profile P from each image can be performed using any image processing method. For example, in the inspection zone Z1, it might be planned to search, for each height taken in the vertical z direction, for the outer edge of the finish by looking for a white / black transition Ti, or from the outside, a white / black transition. The position of this transition Ti along the lateral x direction is determined relative to the vertical z direction. The search for this transition Ti is performed at each height and along the entire height of the inspection zone Z1. As illustrated on the right of Figure 7, for each image, the edge profile P of the finish can be obtained as a curve referenced in the z, x plane. For example, the profile P consists of all the transitions Ti, for each of which the xy, z coordinates are known. The next step is to compare the finish edge profile P with a reference finish edge profile Pf, obtained by various means. According to a preferred variant of the invention, as shown in Figure 8, the extraction of the reference finish edge profile Pf is performed analogously to the extraction of the finish edge profile P from a container illustrated in Figure 7. For this purpose, for a reference container that includes a reference finish edge profile Pf encompassing a portion of a horizontal joint of the reference mold without defects, an image is captured that allows visualization of this reference finish edge profile Pf and, as a reference, the finished surface.Advantageously, this reference image is made under the same conditions as the images of the containers to be inspected, using the inspection device by placing the reference container between camera 4 and light source 3. Analysis of this image according to the method described above allows the extraction of the reference profile Pf of the finished edge in which the horizontal mold joint appears without defect. According to another approach, the reference profile Pf of the finished edge is obtained by learning from several images of a single vessel or from images of several conforming vessels. For example, any type of mathematical operation can be performed on the images or on the profiles extracted from the images. Therefore, the reference profile Pf can be one or more profiles obtained from a single conforming reference vessel or from several conforming reference vessels. Similarly, the reference profile Pf can be an average profile calculated from several rotation increments of a single conforming reference vessel or from several conforming reference vessels.According to another variant of the modality, the reference profile Pf of the finishing edge can also be obtained from a manufacturing drawing or any geometric model of containers. According to another variant of the invention illustrated in Figure 11, in the case, for example, of thread finishes such as those illustrated in Figure 1, the finish edge includes a cylindrical portion, and therefore the finish edge profile includes a straight portion PD. In this case, the reference profile Pf of the finish edge can be a simple reference line D that passes, in the best case, through the straight portion PD of the finish edge profile P, and therefore through the points Ti of this portion of the finish contour, or a straight line parallel to it. According to another variant illustrated in Figure 12, the invention aims to use, as a reference profile Pf, a profile of the finished edge to which a low-pass filter is applied. In other words, the reference profile Pf corresponds to the profile of the finished edge to which filtering or smoothing has been applied. For example, in Figure 12, the profile P of the finished edge (in white) corresponds to the raw profile extracted from the image, while the reference profile Pf (in gray) corresponds to the moving average of profile P. Of course, different types of low-pass filters can be applied, such as the mean, Gaussian, median, or offset median filter. According to another modality variant, it should be noted that a high-pass filter can be applied to the finished edge profiles to compare the finished edge profiles P of the images with a reference finished edge profile Pf. Applying a high-pass filter, for example, of the gradient type, highlights deviations—that is, strong or rapid variations in the derivative of the finished edge profile—which reflect the presence of a local irregularity. The presence of such deviations corresponds to a defect in the horizontal mold joint. The comparison step for each image, between the finished edge profile P of an image and the reference finished edge profile Pf, detects any deviations between these finished edge profiles and the reference finished edge profile. A defect in the horizontal mold joint of a container is detected when at least one image of that container shows a deviation. This comparison, which leads to the observation or not of a defect in the horizontal mold joint, can be implemented according to various analysis methods. For each image, at least at several altitudes and preferably at all altitudes within the inspection zone Zl, the profile of the finished edge P is compared to the reference profile of the finished edge Pf. This step is intended to compare at least one measurement of area, width, and / or slope with a threshold, and a deviation is detected when at least one of these measurements exceeds this threshold. Typically, an area measurement might correspond to the area between the finished edge profile P and the reference profile Pf of the finished edge. A width measurement might correspond to the difference between the finished edge profile P and the reference profile Pf of the finished edge, taken along the lateral x-direction. A slope measurement might correspond to a measurement of the curve obtained by subtracting the finished edge profile P from the reference profile Pf of the finished edge.Another possible measurement is the difference between two amplitudes of the finished edge profile P and the reference profile Pf of the finished edge, taken at equal or different altitudes. According to one variant, to compare the profile of the finished edge P and the reference profile Pf of the finished edge, the actual position of the finished edge is taken into account, including any tilt in the image related to the tilt of the finish with respect to the Z-axis, due to handling risks during inspection, or a lack of verticality of the container. Such a tilt of the finish is illustrated, for example, in Figure 11. It should be noted that if the reference profile of the finished edge Pf is a straight line D passing through the straight portion PD of the finished profile, it is possible in this simplified variant to determine a deviation by measuring the distance of the points of the finished profile P to the reference straight line D.In the example illustrated in Figure 9, the CEc curve represents the deviation between the finished edge profile P and the reference finished edge profile Pf. This curve evolves along the vertical z-axis, that is, as a function of altitude in the inspection area. Different measurements of the CEc deviation curve allow for estimating the size and / or shape of the deviation, and simultaneously assessing the presence of a defect, its type, and its severity. For example, one can consider the point of maximum amplitude of this curve corresponding to the end of a protrusion furthest from the vertical Z-axis of the vessel. This maximum amplitude corresponds to the peak height of the CEc deviation curve, providing a reliable estimate of the radial length of the nodule or flange, which is a very good physical criterion for assessing the severity of the defect.This amplitude can be compared to a threshold value and a deviation (or defect) is detected if this measurement exceeds that threshold. / re i nn / eznz / E / YiAi The threshold can be adjusted to parameterize the severity of the inspection; for example, the operator can decide that lightly marked nodules are tolerated. If the measurement detects a deviation, it is the area of ​​a surface obtained by subtracting the profile of the finished edge from the reference profile of the finished edge; the measurement is, for example, the integral of the CEc deviation curve. If the measurement is a slope, it can be the maximum slope of the CEc deviation curve, or a deviation between two successive slopes of the CEc deviation curve on either side of a peak or threshold crossing. From the preceding description it is clear that the method according to the invention consists of detecting at least one defect of the horizontal joint of the selected mold JH from the following defects: external nodule JHK, external flange JHP and displaced ring JHO. According to an advantageous variant of the method, the invention allows for distinguishing between three defects: external nodule (JHK), external flange (JHF), and displaced ring (JHO). To this end, at least one detection criterion is defined for each defect (external nodule, external flange, and displaced ring), and at least one of these criteria is used to distinguish between the defects. This makes it possible to identify the type of defect, thus allowing for more precise inspection of the vessels. According to one exemplary approach, the shape of the deviations detected at various heights within the inspection area is chosen as the detection criterion, at least for one image. For example, the shape of a deviation can be described in different ways, such as by measuring slopes or amplitudes taken at several consecutive heights. This shape can be observed in the CEc deviation curve. In Figure 9, the CEc deviation curve has a peak, which characterizes an external nodule or an external rim, but if the defect is a displaced ring, as in Figure 5A, the CEc deviation curve will have a stepped shape, as illustrated in Figure 10. According to another exemplary approach, the detection criterion is the angular extent of observation of the deviations corresponding to the number of successive images in which a deviation is detected; that is, the angle of rotation of the container during which the deviation is detected in the images. The angular extent of observation of the deviations for a defect of the external rim or displaced ring type is very close to the circumferential angular extent of the defect as seen from the top plane, as can be clearly seen in Figures 3B, 4B, and 5B. For an external nodule, the angular extent of observation is greater than its circumferential angular extent.If, for example, the rotation increment between two image captures is 4°, and an external node has a very small circumferential angular extent of 2°, and if it is long enough, then it may be visible in several successive images, respectively two or three, separated in time by one or more rotation increments, and therefore its angular extent of observation is respectively 800 12°. However, each of these criteria can be implemented independently of each other or in combination one after the other; one of these criteria can be applied before the other and vice versa. According to an implementation example, the shape of the deviations detected for / re)nn / eznz / E / YiAi at various altitudes in the inspection zone is analyzed to distinguish the displaced ring defect (JHO) from the external node defect (JHK) and the external flange defect (JHF). Indeed, as clearly seen in Figure 5A, the displaced ring defect (JHO) is distinguished, in particular, by its step or crenellated shape from the external node (Figure 3A) and external flange (Figure 4A) defects, which are characterized by a sharp protrusion. According to another implementation example, the angular extent of observation of deviations is analyzed to distinguish between the external nodule defect (JHK) and the external rim defect (JHF). An external nodule defect is detected when the angular extent of observation of deviations falls below a maximum limit, set, for example, at 30°. In other words, a JHK nodule defect is detected when the number of successive images in which a deviation is detected corresponds to an angular extent of observation below the maximum limit set, for example, at 30°. According to another implementation example, the angular extent of observation of deviations is analyzed to distinguish between an external nodule defect (JHK) and an external rim defect (JHF). An external rim defect is detected when the angular extent of observation of deviations exceeds a minimum limit, set, for example, at 30°. In other words, an external rim defect (JHF) is detected when the number of successive images in which a deviation is detected corresponds to an angular extent of observation exceeding the minimum limit set, for example, at 30°. According to the invention, the thresholds applied to the measurements to detect deviations, the criteria of distinction such as those used in comparing the shapes of the profiles, and / or those used as minimum or maximum values ​​of the angular amplitude of observation, are stored in the analysis and processing unit 5, and preferably adjustable by an operator depending on the containers inspected and the required qualities.

Claims

1. A method for detecting, in the finish (B) of glass vessels (R), each with a vertical axis (Z), defects in the horizontal mold joint (JH); the method comprises, for the detection of defects in each vessel, the following steps: placing the vessel (R) between a light source (3) and a camera (4) to capture images whose optical axis of observation is substantially perpendicular to an axis parallel to the vertical axis (Z) of the vessel and whose field of view comprises at least the left edge of the finish or the right edge of the finish, including at least part of the horizontal mold joint; ensuring the rotation of the vessel (R) about itself along the vertical axis (Z) according to at least one rotational evolution; acquiring with the aid of the camera (4), at each rotational increment of the vessel, an image such that the number of images per rotational revolution is greater than 36;Analyze, for each container, the captured images so that: an inspection zone (Zl) of the finish is defined in the images over a finish height that includes at least part of the horizontal mold joint; the profile (P) of the finish edge is detected in the image inspection zone; the finish edge profiles (P) of the images are compared with a reference finish edge profile to detect deviations between these finish edge profiles and the reference finish edge profile; a defect in the horizontal mold joint (JH) of a container is detected when at least one image of said container shows a deviation.

2. The method according to claim 1, wherein the camera (4) is arranged such that the field of view comprises at least part of the horizontal mold joint (JH) and a reference whose high-altitude position is known relative to the horizontal mold joint, and wherein, in order to define the finish inspection zone (Zl) in the images, the reference whose high-altitude position is known relative to the horizontal mold joint is sought, and the finish inspection zone (Zl) is positioned relative to the reference such that the finish inspection zone extends over a height that includes the horizontal mold joint.

3. The method according to claim 2, according to which the images are searched for as a reference all or part of the surface finish (S) or counter finish (CB).

4. The method according to any of the preceding claims, wherein an image capture camera (4) is provided having a horizontal field of view width between 5 mm and 130 mm and a field of view height between 3 mm and 20 mm.

5. The method according to any of the preceding claims, wherein an image capture camera (4) is provided having a resolution greater than 25 pixels / mm.

6. The method according to any of the preceding claims, wherein the image capture camera (4) is positioned so that its optical axis of observation (Y) is substantially tangent to the left or right edge of the finish.

7. The method according to any of the preceding claims, according to which a camera (4) provided with a telecentric lens is available.

8. The method according to any of the preceding claims, wherein a light source (3) is provided having vertical and horizontal illumination dimensions between 100% and 200% and preferably 120% of the field of view dimensions of the camera (4) multiplied by the distance between the light source and the lens (4a) of the camera and divided by the distance between the axis of the containers and the lens of the camera.

9. The method according to any of the preceding claims, wherein there is a telecentric light source whose illuminated field has dimensions greater than or equal to the dimensions of the camera's field of view.

10. The method according to any of the preceding claims, wherein the finished edge profiles (P) of the images are compared with a reference finished edge profile (Pf), using as the reference profile (Pf) at least one profile obtained from at least one reference container deemed to conform, to detect deviations.

11. The method according to any of claims 1 to 9, wherein the finished edge profiles (P) of the images are compared with a reference finished edge profile (Pf), using as the reference profile (Pf) an average profile calculated over several rotation increments of at least one reference vessel deemed to conform, to detect deviations.

12. The method according to any of the preceding claims, wherein the finish edge profiles (P) of the images are compared with a reference finish edge profile (Pf), using as the reference profile (Pf) a finish edge profile to which a low-pass filter is applied, to detect deviations.

13. The method according to any of claims 1 to 9, wherein the finish edge profiles (P) of the images are compared with a reference finish edge profile (Pf), applying a high-pass filter over the finish edge profiles, to detect deviations.

14. The method according to any of the preceding claims, wherein, for each image, the profile of the finished edge (P) and the reference profile of the finished edge (Pf) are compared for at least several heights in the inspection zone (Zl), comparing at least one measurement of the area, width and / or slope with a threshold and a deviation is detected when at least one of these measurements exceeds said threshold.

15. The method according to any of the preceding claims, wherein the method consists of detecting at least one defect in the horizontal mold joint (JH) selected from the following defects: external nodule (JHK), external flange (JHF), and displaced ring (JHO).

16. The method according to any of the preceding claims, wherein at least one detection criterion is defined for each of the following defects: i.e., external nodule (JHK), external flange (JHF), and displaced ring (JHO), and wherein the method consists of distinguishing the detected defects among these three defects using at least one of these detection criteria.

17. The method according to the preceding claim, according to which the detection criterion is chosen, for at least one image, the shape of the detected deviations for various altitudes in the inspection zone.

18. The method according to any of claims 16 or 17, according to which the detection criterion is chosen to be the angular extent of observation of the deviations corresponding to the number of successive images in which a deviation is detected.

19. The method according to claim 17, according to which the shape of the detected deviations for various heights in the inspection zone is analyzed in order to distinguish the displaced ring defect (JHO) from the external nodule (JHK) and external flange (JHF) defects.

20. The method according to any of claims 16 to 19, wherein the angular extent of observation of deviations is analyzed to distinguish the external nodule defect (JHK) from the external flange defect (JHF), wherein an external nodule defect is detected when the angular extent of observation of deviations is below a fixed maximum limit.

21. The method according to any of claims 16 to 20, wherein the angular extent of observation of deviations is analyzed to distinguish the external nodule defect (JHK) from the external rim defect (JHF), wherein an external rim defect is detected when the angular extent of observation of deviations is above a fixed minimum limit.

22. The method according to any of the preceding claims, according to which, for each container with a defect in the horizontal mold joint, a signal is sent to eject the container from a production line.

23. The method according to any of the preceding claims, according to which the rotation of the container (R) about itself along the vertical axis (Z) is ensured in accordance with at least one rotation for a maximum time of 200 ms.

24. An inspection device implementing a method according to any of the preceding claims, for detecting, in the finish of the glass container, defects in the horizontal mold joint (JH).

Images