Success Story
The production of bottles and bottle caps in the beverage industry is highly automated, as large quantities must be produced quickly and reliably. "Our aim is to make automated processes as safe and reliable as possible. This is the lever for being able to operate a process at high speed in the long term. Thus, we rely on standardized tools that allow us to manufacture reproducible products reliably and quickly. Based on machine vision, we have developed a solution for inspecting porcelain bottle caps," says Thomas Kelz, responsible for quality technology at K&S Anlagenbau GmbH.
The company from Lengenwang in Bavaria is developing automation solutions for the electrical, automotive, medical, and consumer goods industries. Together with the consulting firm Computer Vision Solutions Consulting & Research, the company has developed an application for the quality inspection of beverage bottle caps.
The challenge was to implement a quality control system that quickly and reliably detects faults in a fully automated process but is also low-maintenance. Low-maintenance means that the system can quickly and easily be switched to other product types. "Automation only brings decisive added value, if the customer has little effort in maintaining the system," says Dr. Heber. In the newly developed system, quality control is carried out inline. To do this, the camera takes a picture of the bottle cap to be checked. This is then checked using the algorithms of the machine vision software MVTec HALCON.
The challenge with print image control is the variability of the print images on the one hand and the requirement to create a clear operating concept for the system operator on the other hand. Possible print image errors could be, for example, incorrect or out-of-focus colors or imprints that are not completely visible. The central position of the print image on the bottle cap is also evaluated as a quality criterion. "One of the major challenges during implementation was that the number of different-looking print images is theoretically infinite. This applies to both the good, i.e., "OK" prints, and the faulty, i.e., "NOK" prints. Thus, we had to ensure that the software was trained in advance to make the right decision at all times," explains Dr. Heber. For training, the image of an OK print is saved in HALCON. This "good image" serves as a template against which each image is compared during the inspection process. Various parameters in HALCON can be used to set the degree to which deviations are still considered OK.
An important component of the system is the graphical user interface (GUI). With this, the entire system can be overviewed, and changes and configurations of the process can be made.
The requirement for the GUI was that it can dynamically display several different apps, which can be controlled both via the GUI and the PLC. In the finished system, the operator can set various display options, configuration modes, and different user levels. This means that it is also possible to assign rights and thus determine which employees are allowed to carry out which activities. Troubleshooting or adjustments, e.g., if a different print needs to be tested, can also be made via the GUI.
Another core component is MVTec HALCON. HALCON is integrated into the K&S Control software using HDevEngine. "HALCON is the industry standard in the field of image processing software and is not limited to just a few algorithms," explains Dr. Heber. HDevEngine acts as an interpreter designed to dynamically load and execute the machine vision algorithms. It also enables changes to the machine vision processes "on the fly" without having to recompile or recertify the entire application. Furthermore, HDevEngine communicates with the GUI in the background. Operators of the system have little or no contact with it. This is because they operate the system via the GUI they are familiar with.
The machine vision software HALCON enables high and robust detection rates. It not only includes HDevEngine, but also many image processing methods that are used to check the print images. Matching technologies, classification, and blob analysis are used in the specific application.
The three technologies interact as follows: after the image has been captured by a camera, shape-based matching localizes the print in the image - sub-pixel accurately and in near real-time. To do this, the pre-trained print image is matched with the scanned image. This works even if the print is rotated, scaled, perspectively distorted, partially covered or outside the image, or subject to non-linear lighting fluctuations. Once the print has been found, the next task is classification. Here, too, it is important that the class features have been defined and trained beforehand. In the K&S Anlagenbau plant, classification is used for quality control. Possible errors include color deviations, missing edges, incorrect edges, or structures. Corresponding deviations are segmented and analyzed using blob analysis.
If the machine vision technologies described find an error, this is displayed in the GUI. The corresponding area is highlighted in color. This also allows users to recognize the nature of the defect.
If the imprint is "OK", the bottle cap is processed further.
"However, the scope of the machine vision software goes beyond this. To ensure that the process is as automated as possible, we have embedded the software in the entire system in such a way that it can communicate with other components. If a bottle cap is not printed correctly, HALCON informs the system accordingly and the component is automatically ejected," explains Dr. Heber. The communication protocols OPC-UA and PROFINET enable consistent data exchange between the various components within the system. In this way, full automation can be achieved.
"The feedback we have received from the customer has been very positive. The system runs flawlessly and can be operated highly autonomously," says Dr. Heber. Specifically, the system can test 120 bottle caps per minute.
