Method for operating container treatment system, container inspection device for container treatment system

Abstract

A method of operating a container handling system (1) for handling a plurality of container parts (9) for containers (10) and preferably for plastic containers and / or bottles, in which a transport device (6) transports the plurality of container parts (9) as a container part stream along a predetermined transport path from at least one handling device (20, 23, 22, 24, 28, 28) of the container handling system (1), the container parts (9) are transported to at least one further processing device (23, 22, 24, 28, 30, 32) of the container processing system (1), in which at least one sensor device (4) acquires, preferably optically, sensor data, in particular spatially resolved, relating to the container parts (9), and preferably camera images, in order to carry out a container inspection task. According to the invention, a storage variable is determined with respect to the acquired, in particular spatially resolved sensor data, which storage variable characterizes a storage command for storing the acquired, in particular spatially resolved sensor data in a non-volatile storage device (5), the storage variable being determined on the basis of a similarity variable, which storage variable characterizes a storage command for storing the acquired, in particular spatially resolved sensor data in the non-volatile storage device (5). The similarity variable characterizes the similarity of the acquired, in particular spatially resolved sensor data, to predetermined and / or specifiable reference data.

Description

[0001] manual

[0002] The present invention relates to a method for operating a container handling system, a container inspection device for a container handling system, and a container handling system.

[0003] Preferably, the container is a plastic container (especially a PET container), a container whose main component is pulp, and / or a glass container and / or a metal can. The container can be used in the beverage and / or food and / or cosmetics and / or pharmaceutical industries. Examples include metal cans or bottles, such as glass bottles, pulp bottles, and plastic bottles.

[0004] In container handling systems (e.g., container filling systems), various sensors and image processing systems are used for process control. For various process steps, such as injection molding, container cleaning, filling, labeling, sealing, packaging, strapping, and / or shrink packaging, visual inspections are performed subsequently to control the process and / or regulate it in an open or closed loop.

[0005] Typically, control systems, mostly image processing systems, are installed for each process step. These systems require configuration and very complex parameterization. This demands extensive experience and expertise. For fine-tuning and further improvements, the camera images of the controlled container are currently stored in permanent image storage.

[0006] Typically, in current container processing systems, the strategy of temporarily storing images in a camera and permanently storing images on fixed storage is controlled according to different criteria, such as good / bad product evaluation or the last captured image.

[0007] If you want to store images with a new feature, error, or appearance, the disadvantage of the above strategy is that there are no suitable standards or keep strategies for storing these images. Simply put, the machine can only store what it can select using existing standards. Since the machine takes approximately 1,000,000,000 images per day, manually selecting and reviewing all images is impractical.

[0008] A method for performing a setup operation of a container inspection apparatus is known from DE 10 2021 133 164 B3. In this method, a sensor device acquires spatially resolved sensor data related to a container to be inspected, and a real-time evaluation device evaluates the spatially resolved sensor data of a single container in real time using a settable real-time container inspection model. Furthermore, multiple spatially resolved sensor data sets are provided on a non-volatile storage device. During the setup operation, the setup device retrieves the stored multiple spatially resolved sensor data sets and evaluates a test container inspection model based on the retrieved multiple spatially resolved sensor data sets.

[0009] The purpose of this invention is to overcome the shortcomings known from the prior art and to provide a method for operating a container handling system, a container inspection device for a container handling system, and a container handling system. This invention proposes a storage strategy for sensor data collected during container inspection, which selects important sensor data to store on a storage device to further improve inspection accuracy.

[0010] According to the invention, this objective is achieved by the subject matter of the independent claims. Advantageous embodiments and further improvements of the invention are the subject matter of the dependent claims.

[0011] In a method of operating a container handling system according to the invention to process multiple container components for containers, and preferably for plastic containers and / or bottles, a transport device is provided that transports the multiple container components (particularly container components to be processed and / or processed) as a container component flow along a predetermined transport path from at least one processing device of the container handling system to at least another processing device of the container handling system. Preferably, the container handling system includes the transport device.

[0012] Preferably, the container is a plastic container (especially a PET container), a container whose main component is pulp, and / or a glass container and / or a metal can. The container can be used in the beverage and / or food and / or cosmetics and / or pharmaceutical industries. Examples include metal cans or bottles, such as glass bottles, pulp bottles, and plastic bottles.

[0013] "Container components for containers" can also be understood as the container itself. Therefore, a container component may be, for example, a preform that is produced into a fully formed container through a molding process, or it may be a fully formed container.

[0014] Container components for containers can also be container fittings, such as (preferably resealable) container closures (e.g., screw caps or bottle caps), (PET and / or plastic) caps, labels, (laser or direct printing) markings, fillers and / or (finished) container packaging, container assemblies or the like (and combinations thereof).

[0015] Therefore, the transport equipment can be, for example, a feed section (e.g., a feed rail) for the container sealing section, which transports the container sealing section from the collection equipment to the sealing equipment for sealing the container.

[0016] Preferably, the container component is an object that can be transported as a single unit by the transport equipment and transported independently of other transported container components.

[0017] Preferably, the (first) processing device and / or (at least one) other processing device (particularly in each case) perform at least one processing step on the container component (and preferably on the container).

[0018] (First) The processing steps of the processing equipment and / or (at least one) other processing equipment may be selected from a set of processing steps, which includes: an injection molding process for producing injection molded parts (e.g., plastic preforms); a cleaning process; a crushing process and / or a slitting process (e.g., within the framework of a recycling process); a molding process (particularly (stretch) blow molding); a (laser) marking process; a personalization process (e.g., affixing QR codes); a rejection process (in which container parts are rejected from the container part stream of the container processing system); and a sorting process (in which containers are sorted according to type). The process includes: sorting components; filling; sealing; (especially) printing; labeling; laser decoration; bundling (for container components, preferably containers); packaging (especially applying primary and / or secondary packaging, such as heat-sealing several containers into a package and / or bundling unit), especially heat shrink packaging; determining the composition of a substance or mixture of substances, for example, by means of mass spectrometry and / or odor sensor equipment, to determine, for example, the composition of the gaseous environment and / or air within or at a container component, preferably within or at a container; and similar processes and combinations thereof.

[0019] In addition, at least one sensor device (the sensor device of the container handling system and / or the container inspection apparatus described in more detail below) acquires sensor data (preferably camera images) related to the container components (to be inspected and / or transported) to perform container inspection tasks, preferably during the operation of the container handling system.

[0020] Preferably, the transport equipment transports multiple (to be processed and / or processed) container components to a sensor device (for inspecting the container components to perform a container inspection task), which in each case acquires sensor data, particularly spatially resolved data, related to the transported container components. The sensor device and / or a container inspection apparatus including at least one of the sensor devices (described in more detail below) may be arranged between the processing equipment and the at least one other processing equipment.

[0021] Preferably, at least certain areas of the container component (to be inspected), preferably areas of the container component that can be observed or seen from at least one viewing direction, are mapped into the sensor data.

[0022] It is conceivable that, for each container component to be inspected and / or transported to the sensor device, sensor data is acquired or collected individually or on a per-container-component basis (in a separate acquisition step for each of the sensor devices).

[0023] A single sensor device can be set up to collect the necessary sensor data related to the container components in order to perform container inspection tasks. However, it is also conceivable to set up several sensor devices for this purpose, which collect sensor data related to the container components from several shooting directions and / or collect data on container components transported on different tracks in a multi-track transportation area.

[0024] The sensor data is preferably spatially resolved sensor data, which specifically reflects the characteristics to be acquired (e.g., color values ​​and / or grayscale values ​​and / or brightness values) of a certain area of ​​the container component. Preferably, the spatially resolved sensor data indicates (or is able to indicate) a sensor data curve that depends on at least one spatial and / or geometric coordinate, and preferably depends on at least two spatial and / or geometric coordinates.

[0025] Sensor data, such as sensor data captured by a camera, can be, for example, color values ​​and / or grayscale values ​​and / or brightness values.

[0026] Sensor data acquired by a LIDAR device can be RGB values ​​and / or intensity values, which are acquired and stored for each data point together with or depending on its X, Y, and Z position values.

[0027] It is also conceivable that the sensor data is frequency-resolved sensor data.

[0028] Therefore, for example, for each sensor data point, at least one intensity value can be acquired that depends on the frequency of the radiation acquired by the sensor device, so that the sensor data indicates a sensor value curve that depends on the frequency.

[0029] It is also conceivable that the sensor data is spectrometer sensor data, preferably generated by a mass spectrometer, which is used, for example, as an odor sensor. It is conceivable that this is used to analyze, for example, the composition of gases and / or air and / or air mixtures (e.g., a gaseous environment in a container). Here, the acquired sensor data can, for example, indicate an intensity curve that depends on the mass-to-charge ratio of the atoms and molecules contained in the gaseous environment.

[0030] Preferably, the acquisition of sensor data, particularly spatially resolved sensor data, is optical. More preferably, the spatially resolved sensor data is a camera image.

[0031] Preferably, sensor data related to the container component to be inspected is acquired while these individual container components are being transported at a constant, and in particular, unreduced, transport speed—that is, while the container components are in motion. In other words, the container components are not slowed down and / or stopped in order to acquire sensor data. This provides the advantage of high throughput and production speed for the container handling system.

[0032] For optical acquisition of sensor data, a container component can be provided, preferably with illumination of the container, such as incident light illumination and / or transmitted light illumination.

[0033] According to the invention, (preferably by a container inspection device) storage parameters are determined with respect to the acquired, particularly spatially resolved sensor data (especially in a computer-implemented method step), which characterize storage instructions for storing the acquired, particularly spatially resolved sensor data in (at least one) non-volatile storage device.

[0034] The storage parameter is preferably a binary parameter, wherein one possible value of the storage parameter preferably indicates that the collected, particularly spatially resolved, sensor data needs to be stored on a non-volatile storage device.

[0035] Preferably, another value (preferably a second value) of the storage parameter indicates that the acquired, particularly spatially resolved, sensor data does not need to be stored on a non-volatile storage device. In this case, it is preferable to subsequently store only the acquired, particularly spatially resolved, sensor data temporarily, and delete it from the storage device, for example, after storing a (specific) amount of more acquired, particularly spatially resolved, sensor data.

[0036] Preferably, by assigning / allocating a value to a storage parameter (which indicates that the collected sensor data needs to be stored on a non-volatile storage device), the operation of storing the collected sensor data on a non-volatile storage device is triggered, and particularly preferably executed.

[0037] The non-volatile storage device can be a storage device that is a (fixed) component of at least one sensor device and / or container inspection device. It is conceivable that this storage device is a ring memory in which the oldest sensor data is overwritten when the storage capacity is reached (therefore the stored sensor data is only available for a limited time).

[0038] Non-volatile storage devices can also be storage devices for container processing systems, which are designed, for example, as fixed memory.

[0039] It is conceivable that the collected sensor data would first (especially temporarily, preferably only temporarily) be stored on the storage device of the sensor device.

[0040] Preferably, depending on the (determined) storage parameters, the acquired, particularly spatially resolved sensor data, is transferred from a storage device for temporary storage of the acquired, particularly spatially resolved sensor data (e.g., a storage device for the sensor device) to a storage device for permanent storage (preferably non-volatile) of the acquired, spatially resolved sensor data (this storage device may be, for example, a fixed memory and / or an external storage device associated with the sensor device and / or, for example, a storage device for a container inspection device and / or a container handling system and / or an external storage device associated with the container handling system).

[0041] "Non-volatile" can also mean that the selected image or sensor data and / or sensor data stored on a non-volatile storage device are only retained for a specific period of time, i.e., they can be deleted after a set time or after a parameterized operation (by an assembly worker). "Non-volatile" can also mean that the image or sensor data does not need to be retained or stored after the machine is turned off.

[0042] "Non-volatile" should also mean that the image data to be "held" and / or the sensor data to be stored must be available for parameterization in the simplest case.

[0043] It is also conceivable that, additionally or alternatively, "non-volatile storage device" is understood to mean that the image data to be "held" and / or the sensor data to be stored is still retained even if the storage device is not powered.

[0044] According to the present invention, storage parameters are determined based on similarity parameters, which characterize the similarity between acquired, particularly spatially resolved sensor data and predetermined and / or predeterminable reference data.

[0045] The reference data is preferably stored in the storage device of the container handling system, and particularly preferably in the container inspection device that includes the sensor device.

[0046] Specifically, the collected, especially spatially resolved, sensor data is compared with reference data, and similarity parameters are determined based on the comparison results.

[0047] Preferably, the similarity parameter is a non-discrete parameter, which in particular cannot take only two values ​​or a finite (fixed) number of values. Preferably, the similarity parameter is a continuous parameter. Preferably, the similarity parameter represents the degree of similarity.

[0048] In other words, to extend the machine's retention strategy or image storage function, in the case of a sensor device designed as a camera, this image storage function can retain images in the camera and fixed memory based on image similarity. For example, images of the errors or features being sought can be used.

[0049] In particular, in the case of multiple sensor data, especially multiple spatially resolved sensor data, the multiple sensors can be ordered according to their respective similarity to the reference data using a similarity parameter determined with respect to predetermined reference data in each case. That is, it is feasible to distinguish between two or more different acquired sensor data in terms of their similarity to the reference data.

[0050] Preferably, the predetermined plurality of acquired, particularly spatially resolved, sensor data are sorted according to a similarity parameter determined for reference data (preferably for a predetermined plurality of reference data). Preferably, the storage parameters of the sensor data in these sensor data are determined depending on the order in which the plurality of acquired, particularly spatially resolved, sensor data are sorted.

[0051] That is, for example, storage parameters representing or characterizing the sensor data to be stored on a non-volatile storage device can be allocated to five sensor data that have the highest similarity to the reference data compared to the remaining sensor data among a plurality of acquired, particularly spatially resolved sensor data and have reached the top five in the sorting.

[0052] This provides the advantage that, for example, images or sensor data of specific defects associated with container component types that are rare in the container component flow can be pre-defined as reference data, and that among multiple acquired, especially spatially resolved, sensor data, the examined sensor data that are closest to that specific defect image can be automatically searched.

[0053] Here, for example, the operator can predetermine how many sorted sensor data points, particularly spatially resolved ones, to be stored in non-volatile storage (by assigning / allocating corresponding storage parameters).

[0054] The data from multiple sensors to be sorted can be sensor data collected directly and sequentially.

[0055] It is also conceivable that the multiple sensor data to be sorted are (particularly preferred) all sensor data collected within a certain time period, preferably within a time period that can be predetermined (by the operator), especially spatially resolved sensor data. These multiple sensor data to be sorted can be stored simultaneously in a storage device.

[0056] Preferably, the multiple sensor data sets to be sorted are not (at least not entirely) simultaneously stored on the (shared) storage device (especially since the applicant collects a very large amount of sensor data daily, thus quickly reaching the storage capacity). Preferably, when the collected sensor data stored on the storage device reaches a predetermined quantity and / or when a predetermined storage requirement is reached, in the case of further sensor data for the multiple sensor data sets to be sorted, a corresponding similarity parameter for these further sensor data sets is determined. Based on the determined similarity parameter, a decision is made whether to store these further sensor data on the storage device and instead delete other sensor data already stored on the storage device for the multiple sensor data sets to be sorted (or not to store these further sensor data on the storage device and therefore not consider them in further sorting).

[0057] The advantage of this method is that, by using a similarity parameter, a predetermined number of the acquired sensor data from the multiple sensor data to be sorted are always stored on the storage device, which are closest to the (predetermined) sorting or storage criteria among the sensor data observed so far from the multiple sensor data to be sorted.

[0058] Another advantage is that this sorting method can be used in continuous operation. That is, once the relevant sensor data is acquired, sorting can begin one after another based on similarity parameters associated with the reference data. There is no need to wait until all the sensor data for the multiple sensors to be sorted is actually available.

[0059] In a preferred method, the reference data is preferably spatially resolved reference sensor data acquired by a sensor device. The sensor device can be, in this context, a sensor device that acquires sensor data, particularly spatially resolved sensor data (within which storage parameters need to be determined).

[0060] Additionally or alternatively, the sensor device may be a sensor device that is different from (e.g., structurally identical to) the sensor device that acquires sensor data, particularly spatially resolved sensor data (of a container handling system).

[0061] That is, sensor data acquired by structurally identical sensor devices from different container handling systems can be used as reference sensor data. This provides the advantage of, for example, checking whether very rare defects or features arising from faults in different container handling systems that occur in the container components being handled also occur in the container handling system under consideration.

[0062] Additionally or alternatively, the reference sensor data can be spatially resolved sensor data acquired by the sensor device and modified manually and / or by image processing methods (preferably automatically). That is, the sensor data can be modified (particularly manually) to have additional (particularly predetermined) defects. In this way, it is possible to check whether the sensor data acquired by the sensor device of the container handling system exhibits the same or similar defects.

[0063] It is also conceivable to generate reference sensor data using an AI-based reference sensor data generation model (machine learning), which has been trained, for example, using multiple existing sensor data of container parts with known defect types (and multiple existing sensor data of defect-free container parts) to generate sensor data about defects.

[0064] It is also conceivable that, for example, in the case of integrating a new type of container component, data provided by the developers of the new type of container component, especially the collected sensor data (e.g., as the result of simulation or as sensor data collected by means of (third-party) external sensor devices related to the container processing system), could be used as reference data.

[0065] In another preferred approach, a similarity parameter characterizes the similarity between acquired, particularly spatially resolved, sensor data and a predetermined and / or predeterminable plurality of, particularly spatially resolved, reference sensor data. That is, a similarity parameter representing or characterizing the similarity to several reference sensor data can also be considered.

[0066] For example, as a similarity parameter characterizing the similarity between acquired, particularly spatially resolved sensor data and predetermined and / or predeterminable, particularly spatially resolved reference sensor data, the largest or smallest similarity parameter among all similarity parameters obtained with respect to exactly one reference sensor data from the multiple reference sensor data can be selected.

[0067] Determining similarity parameters for multiple reference sensor data can be used, for example, to select or store reference sensor data from multiple acquired sensor data that have the lowest possible similarity to predetermined reference sensor data (e.g., to container components with known defects, particularly those with all known defects to date, and to defect-free container components) in order to, for example, discover new defect types that have not been known to date.

[0068] In another preferred method, the operator of the container handling system, preferably using the human-machine interface of the container handling system, predetermines reference data, preferably spatially resolved reference sensor data and / or multiple, particularly spatially resolved reference sensor data. This provides the advantage that the operator can select, for example, the collected sensor data suggested for the operator as reference sensor data using an input device (e.g., designed as a touchscreen) of the container handling system (e.g., by means of an inspection device and / or the container handling equipment, particularly by means of a display device).

[0069] However, it is also conceivable that operators could transmit such reference data via the human-machine interface of the container inspection device and / or container handling system, with regard to which similarity parameters need to be determined. In this way, reference data can be determined in a user-friendly manner.

[0070] In another preferred method, a rejection parameter is determined based on the acquired, particularly spatially resolved, sensor data. This rejection parameter characterizes a rejection instruction for removing an associated container component from the container component stream.

[0071] Preferably, the decision to remove a container component from the container component stream depends on a rejection parameter. For the corresponding rejection parameter, which represents a positive rejection instruction for removing an associated container component from the container component stream, the rejection device of the container handling system (automatically) removes the associated container component from the container component stream.

[0072] The determination of the rejection parameters can be performed by the container inspection device (in a computer-implemented method step). However, it is also conceivable that a separate, particularly processor-based, rejection decision device could be provided (especially as part of a container processing system).

[0073] In particular, the similarity parameter is not the same as the rejection parameter, so not all sensor data related to the container parts to be rejected is stored.

[0074] Preferably, the similarity parameter is determined independently of the elimination decision and / or the elimination parameter is not considered when determining the similarity parameter (preferably the reverse is also true).

[0075] Specifically, the storage parameters are not determined based on one or more determined rejection parameters. That is, not all sensor data related to the container component to be rejected, as well as the sensor data subsequently acquired in each case, are stored; instead, the storage parameters are determined based only on at least one or more determined similarity parameters.

[0076] In particular, in the case of the proposed method, it is important that the determination of storage parameters does not depend solely on whether container components should be removed. In other words, the storage strategy is independent of the selected or predetermined removal criteria.

[0077] In another preferred method, the acquired, particularly spatially resolved, sensor data is used as reference sensor data associated with at least one rejected and / or to be rejected container component. This provides the advantage that this storage strategy also examines acquired sensor data similar to these reference sensor data associated with non-rejected container components. This allows for checking whether the predetermined rejection criteria are correct or should be further adjusted.

[0078] In another preferred method, storage parameters characterizing a positive storage instruction are determined if the acquired, particularly spatially resolved, sensor data shows a relatively high similarity to predetermined and / or predeterminable reference data, and / or if such relatively high similarity is determined. This can be particularly advantageous, for example, for identifying (and storing) sensor data of rare container component types in a container component stream, or for identifying container components with similar defects or characteristics as described above.

[0079] In another preferred method, a plurality of reference sensor data are predetermined, and if the relatively low similarity between the acquired, particularly spatially resolved, sensor data and the predetermined plurality of reference sensor data is determined and / or if this relatively low similarity is determined, then storage parameters characterizing the positive storage instruction are determined. This provides a highly advantageous possibility for discovering so-called “blind spots,” i.e., features or defects of container components that have not yet been discovered, compared to methods known from the prior art. Since these features or defects are still unknown, they have a low similarity to all known sensor data mapped in the predetermined plurality of reference sensor data. This can be used, for example, for further improvements, fine-tuning of the processing, or, for example, for discovering aging or faults in the container processing system.

[0080] In another preferred method, the predetermined plurality of reference sensor data includes both reference sensor data related to container components to be removed from the container component stream and reference sensor data related to container components not to be removed from the container component stream. This is particularly advantageous if, as described above, it is necessary to discover defects or features that have been previously unknown at the container components.

[0081] In another preferred method, a set of container component features is predefined (for the container processing system and / or container inspection device), and a similarity parameter is determined based on this set of container component features. Preferably, the set of container component features is stored on the storage device of the container processing system and / or container inspection device. Preferably, the set of container component features is transferred to the container processing system and / or container inspection device (particularly from external storage devices and / or external servers).

[0082] Preferably, the set of container component features can be accessed, and in particular, individual container component features within that set can also be accessed. That is, the set of container component features is not inherently implicit in the image evaluation algorithm (similar to a "black box"), but is stored in such a way that they can be accessed independently and separately. Particularly preferably, the set of container component features can also be replaced individually (particularly independently of other software modules). It is also conceivable that the set of container component features can be output and / or transmitted solely for its own sake.

[0083] In another preferred approach, the set of container component features is a set of container component features automatically obtained (particularly extracted by a neural network) within the framework of a machine learning method performed on the training container inspection task.

[0084] Here, the training container inspection task is preferably different from the container inspection task. Preferably, container component features are extracted during or in the course of a machine learning method performed with respect to the training container inspection task. By executing this machine learning method, a (trained) algorithm or (trained) (machine learning) model should be obtained for performing or executing the training container inspection task.

[0085] Preferably, the set of extracted container component features is a set of container component features that are (automatically) extracted within the framework of the machine learning method performed on the training container inspection task.

[0086] Here, the extracted container component features are not, in particular, predetermined features, nor are they selections of predetermined features (e.g., user-defined features). The extracted container component features are, in particular, abstract features that represent or characterize, for example, contrast, parameters characterizing the frequency of straight lines (e.g., the number of straight lines), brightness or brightness curves, the shape of contours (lines) and / or boundary lines, angles, shapes, number of angles, curvature, combinations thereof, etc.

[0087] Preferably, (in particular, all) container component features are automatically generated, preferably automatically generated within the framework of machine learning methods (and in particular not selected).

[0088] Preferably, the set of extracted container component features is not adjusted and / or changed, even when defining (new and / or further and / or adjusted) container inspection tasks and / or prescribing new / modified reference data or reference sensor data.

[0089] It is conceivable that, for example, the user of the container processing system predetermines the number of container component features in the set of container component features to be extracted or to be extracted, and / or the number of container component features in the set of container component features to be extracted or to be extracted is, for example, predetermined by the user of the container processing system. Preferably, when determining the set of container component features to be extracted, a predetermined number of container component features is considered. For example, the user of the container processing system may transmit this predetermined number of container component features to an external server, which determines the set of container component features to be extracted. However, it is also conceivable that the number of container component features is predetermined by the manufacturer of the container processing system, and is particularly not influenced by the container processing system (or its operator).

[0090] Using AI-based similarity metrics, similarity to reference images or reference sensor data can be determined. Preferably, only similar images or acquired sensor data are retained and stored.

[0091] This offers the advantage of being able to selectively collect images with specific features, errors, or appearances. If a customer complains about the inability to recognize a container, images of similar containers can be selectively collected and incorporated into improvements to the recognition system.

[0092] For example, this allows for the targeted and efficient collection of training images for AI applications.

[0093] Preferably, a training dataset can be generated based on predetermined reference sensor data and / or multiple reference sensor data to train a machine learning recognition model to identify the types of defects and / or container components.

[0094] Preferably, the training container inspection task is a different container inspection task. This preferably reuses the expensive feature extraction (preferably using a neural network) and therefore eliminates the need for further adjustments. However, it is also conceivable that the training container inspection task is the container inspection task to be performed.

[0095] In another preferred approach, the set of container component features is a set of container component features extracted within the framework of a supervised learning method. Preferably, a set of training data is used to perform the supervised learning method, this set of training data including sensor data related to the container components, which is labeled or marked with the inspection results of a predetermined training container inspection task to be obtained under appropriate conditions.

[0096] In another preferred approach, the supervised learning method is the K-Nearest Neighbors algorithm (also abbreviated as "k-NN" or "KNN"). Advantageously, this is a simple algorithm that can easily adapt to newly added training patterns. The K-Nearest Neighbors algorithm only requires a K value and a distance metric (migration metric), which is far fewer values ​​required compared to other machine learning algorithms.

[0097] In another preferred approach, conventional machine learning algorithms are used (as a learning method), such as decision tree learning (where decision trees are nonparametric supervised learning algorithms, preferably having a hierarchical tree structure, such as for classification and regression tasks), random forests or random decision forests, logistic regression, k-means clustering, and support vector machines (abbreviated as SVM).

[0098] In another preferred method, the feature space is stretched by the provided set of extracted container component features, or the feature space is stretched by the provided set of extracted container component features. In other words, a feature space can be formed by stretching the provided set of extracted container component features / the feature space is stretched by the provided set of extracted container component features.

[0099] Preferably, a distance metric (spread metric) related to the feature space is provided.

[0100] Preferably, a similarity parameter is determined using a distance metric (spacing metric). Here, sensor data, particularly spatially resolved sensor data, and / or reference sensor data are preferably represented in a feature space (as feature vectors). The spacing between these two feature vectors is preferably determined using a distance metric (spacing metric). This spacing, or a parameter characterizing this spacing, is preferably used as a similarity parameter between the acquired, particularly spatially resolved sensor data, and the reference sensor data.

[0101] The reference data may, for example, already be a representation of (reference) sensor data in the feature space. That is, the reference data may, for example, already be a feature vector (or a representation of that feature vector). Reference data that has already been used as a (pre-determined) feature vector indicator has the advantage of significantly smaller data volume.

[0102] In other words, a distance metric (spacing metric) associated with the feature space is preferably provided, wherein additionally or alternatively, the real-time evaluation device uses the distance metric (spacing metric) as a similarity measure between sensor data of different container components, preferably different containers, particularly spatially resolved sensor data, and / or as a similarity measure between the acquired sensor data and (e.g., predetermined within the framework of setting a container inspection task) reference sensor data.

[0103] Preferably, (within the framework of performing container inspection tasks) the similarity between the collected, particularly spatially resolved, sensor data of one container component and sensor data of another container component (e.g., intended as reference sensor data) is assessed using distance metrics.

[0104] Preferably, based on the extracted set of container component features, a feature vector is created for all acquired (particularly spatially resolved) sensor data, i.e., for example, for each captured camera image. The feature vector can be, for example, a 256-dimensional vector.

[0105] Preferably, the feature vector has a maximum of 512 dimensions, more preferably a maximum of 256 dimensions, and particularly preferably a maximum of 128 dimensions. More preferably, the feature vector has a minimum of 16 dimensions, more preferably a minimum of 32 dimensions, more preferably a minimum of 64 dimensions, and particularly preferably a minimum of 128 dimensions. However, in principle, it is also possible to use feature vectors with more than 512 dimensions or feature spaces corresponding to higher dimensions.

[0106] Preferably, (particularly within the framework of performing container inspection tasks), the distance between the feature vector of sensor data collected about the first container component and the feature vector of sensor data collected about the other container component can be determined by means of a distance metric.

[0107] Preferably, within the framework of performing a container inspection task, the distance between the feature vector of the sensor data collected for the first container component and the feature vector determined for the reference sensor data can be determined using a distance metric.

[0108] In this way, distance metrics can be used to determine the feature vector most similar to a predetermined feature vector (and the corresponding sensor data collected).

[0109] Preferably, a feature vector (and the corresponding collected sensor data) is determined with the shortest possible distance between the predetermined distance metric and the predetermined feature vector.

[0110] However, it is also conceivable to determine eigenvectors (and corresponding sensor data) that have the largest possible (or maximum) gap with respect to a predetermined distance metric and the eigenvectors and / or average and / or high-frequency eigenvectors of the sensor data acquired to date (or acquired within a specific time period). In this way, it is feasible to identify very rare sensor data (e.g., data with very rare defects and / or rare container component types).

[0111] Here, distance metric is preferably used as a similarity metric to assess the similarity between the acquired, particularly spatially resolved, sensor data and reference data, particularly reference sensor data, in order to determine the similarity parameter.

[0112] In another preferred approach, Euclidean metric and / or cosine similarity in the feature space are used as distance metrics. Cosine similarity (also known as "cosine similarity" or "cosine distance") is a measure of the similarity between two vectors, where the cosine value of the angle between the two vectors is determined. Cosine similarity can be specifically understood as the degree of directional consistency between two vectors. Specifically, the cosine similarity between two vectors a and b can be calculated by dividing the standard dot product of vectors a and b by the Euclidean norm of a and the Euclidean norm of b, i.e.: Cosine similarity = (a·b) / (||a||||b||).

[0113] A relatively small distance between two feature vectors (in the feature space) obtained using a distance metric is preferably considered as low similarity between two sensor data points corresponding to the respective feature vectors. Conversely, a relatively large distance between two feature vectors (in the feature space) obtained using a distance metric is preferably considered as high similarity between two sensor data points corresponding to the respective feature vectors.

[0114] In another preferred method, the container inspection task is a classification task selected from a set of classification tasks, including: classifying container components and preferred containers as defective or non-defective (preferably binary classification) (good / defective); identifying and / or classifying the defect types of container components and preferred containers; identifying and / or classifying different types of container components and preferred containers (e.g., about ten different bottle types); identifying and / or classifying the outline and / or color of container components and preferred containers; and classifying the defects at the inspected container components (especially by means of the first...). Identifying and / or classifying error-free / error-prone execution of at least one processing step in a processing device; identifying relatively rare container component types and / or defect types in a container component flow (“rare” specifically means occurring at a frequency of less than 1 / 1000); label control; liquid level control; inspecting for bubbling behavior of liquids within containers; identifying container cracks and / or bottle neck breaks and / or container bottom area ruptures; identifying foreign particles attached to or present in or within container components and / or containers; and similar classification tasks and combinations thereof.

[0115] Preferably, the working operation is a continuous (production) operation of the container inspection device and / or a continuous (production) operation of the container handling system (e.g., a container filling system) having the container inspection device. Specifically, the working operation can be a production operation. Specifically, the working operation is not a testing operation and / or maintenance operation and / or setup operation, in which the transport speed of the container or container component (as it passes through the container inspection device) is reduced compared to the transport speed in the working operation.

[0116] Preferably, the sensor device is selected from the group comprising: image capturing devices, such as cameras (preferably black-and-white and / or color cameras); CMOS sensors (CMOS is an abbreviation for Complementary metal-oxide-semiconductor); CCD sensors; 3D sensors; X-ray based image capturing devices; optical elements; thermal imaging cameras; stereo cameras; LiDAR cameras; odor sensors; and / or (mass) spectrometers; and similar devices and combinations thereof.

[0117] In a preferred embodiment, the transport device transports the container from a first processing device to another (or second) processing device (and / or the transport device is particularly suited to and designed to transport the container from the first processing device to another (or second) processing device).

[0118] Preferably, the first processing equipment and / or the other processing equipment is selected from a group comprising: injection molding apparatus for producing injection molded parts (e.g., preforms); cleaning apparatus for cleaning containers and / or container components; crushing equipment for crushing containers; filling apparatus for filling containers; molding apparatus, particularly blow molding machines, for forming plastic preforms into plastic containers; sealing equipment for sealing containers; labeling apparatus; marking equipment; sorting equipment; packaging equipment (for sealing and / or heat shrink packaging); equipment for bundling container components and / or containers; determining equipment for determining the composition of a substance or mixture of substances, such as determining the composition of the gaseous environment and / or air within or at a container component, preferably within or at a container, such as mass spectrometers and / or odor sensor devices; and similar equipment and combinations thereof.

[0119] Preferably, the container component flow is a sequential or sequential (particularly continuous) flow of container components (along the transport path). For example, the container component flow can be a flow of containers that are sequential or sequential (along the transport path). Here, the container component flow can be guided or transported in a single-channel or multi-channel manner (with the aid of transport equipment) in certain areas, preferably within the entire container inspection device (as a dense flow). Preferably, at least one sensor device is assigned to each channel of the container component flow, and each container component in the container component flow located in that channel is sampled.

[0120] The transport equipment may also be a high-density transport vehicle, used for transporting multiple container components, preferably containers, preferably in a multi-channel and / or unordered manner. The transport equipment may also be a buffer, used for buffering multiple container components, preferably containers, preferably in a multi-channel and / or unordered manner.

[0121] The container components, preferably the container, can preferably be transported or guided at least in sections and preferably along the entire transport area (by the transport equipment) upright or vertically.

[0122] Preferably, the transport equipment is suitable for and intended to guide or transport multiple container components, preferably containers, at least in segments, preferably along the entire transport area under back pressure.

[0123] Preferably, the transport equipment is adapted and designed (at least within the transport area) to transport and / or guide at least one container component (preferably at least one container) per hour, preferably at least 5,000 (especially those to be inspected) container components (preferably containers) per hour, preferably at least 20,000 (especially those to be inspected) container components (preferably containers) per hour, preferably at least 100,000 (especially those to be inspected) container components (preferably containers) per hour, preferably at least 140,000 (especially those to be inspected) container components (preferably containers) per hour, particularly preferably at least 180,000 (especially those to be inspected) container components (preferably containers) per hour, and this process is carried out within the operating operations of the handling equipment and / or container handling system. Preferably, the transport equipment is adapted and designed (at least within the transport area) to transport and / or guide up to 180,000 (especially preferably up to 200,000) (especially those to be inspected) container components (preferably containers) per hour, and this process is carried out within the operating operations of the handling equipment and / or container handling system and / or container inspection device.

[0124] Preferably, the transport equipment in a single-lane transport area is suitable and designed (at least within the single-lane transport area) to transport and / or guide at least 100,000 container components (preferably containers) per hour and / or at most 180,000 (particularly preferred, up to 200,000) container components (preferably containers) per hour, and the process is carried out within the working operation of the handling equipment and / or container handling system and / or container inspection device.

[0125] The container can be a preform, which is produced into a fully formed container through a molding process, and / or these preforms have so far been produced only through a primary molding step. The container can also be a fully formed container, which has reached its final shape, for example, through a molding process of a preform (e.g., an injection molded part or injection preform).

[0126] The container may be empty, still needing to be filled, and / or needing to be recycled and / or needing to be refilled. The container may be a filled container. Additionally or alternatively, the container may be a container sealed using (in particular exactly one) a container seal and / or capable of being sealed using (in particular exactly one) a container seal.

[0127] The container can be a single-use container or a reusable container.

[0128] The container is preferably (preferably sealed) and especially capable of being sealed, for containing liquids and / or flowable substances, such as pastes and / or ointments and / or gels, for example substances from the food industry, cosmetics industry or pharmaceutical industry.

[0129] It is also conceivable that the container is a container for holding liquids and / or solids, such as a container for holding contact lenses.

[0130] Preferably, the external storage device is a cloud-based (preferably non-volatile) storage device and / or an external server (containing the storage device), wherein the storage device is accessed, in particular, via the Internet (and / or via public and / or private networks, particularly at least partially wired and / or wireless). The external server should be specifically understood as an external server associated with container inspection devices and / or real-time evaluation devices and / or configuration devices, particularly a backend server.

[0131] External servers, for example, are backends, particularly for container inspection device manufacturers or service providers, adapted to manage spatially resolved sensor data (particularly sensor data from multiple sensor devices and / or multiple container inspection devices), and / or execute machine learning methods related to the (training) of the container inspection method to be performed, and / or configure and adjust the container inspection device. The functionality of the backend or external server can be executed on an (external) server farm. The (external) server can be a distributed system.

[0132] The present invention further relates to a method, particularly a computer-implemented method, for determining a set of features of container components, and particularly for performing feature extraction on the set of container component features, for use in a container inspection apparatus to perform a container inspection task, the method comprising the following steps:

[0133] - Provide training on container inspection tasks;

[0134] - Provide a training dataset that includes multiple sensor data related to multiple container components used for the container, and each includes a label that indicates the expected result of the training container inspection task;

[0135] - Implement machine learning methods based on the training dataset for training container inspection tasks, with a preference for supervised machine learning methods;

[0136] - Extract container component features obtained in machine learning methods. Feature extraction can be performed by the algorithm of a machine learning method as the first processing stage. It represents a favorable possibility for significantly improving processing efficiency and reducing the influence of irrelevant information. Here, feature extraction is automatically learned during the training phase of the machine learning method.

[0137] Here, machine learning methods can be performed by (especially) deep neural networks, such as “convolutional neural networks” (CNNs). They automatically learn to extract meaningful features, such as edges, shapes, and textures, from (raw) sensor data, for example, throughout the training process (e.g., in computer vision methods for container inspection tasks).

[0138] Here, the training container inspection task is preferably different from the container inspection task to be performed and / or executed at the container inspection device. Preferably, container component features are extracted in or during a machine learning method performed with respect to the training container inspection task. By executing this machine learning method, a (trained) algorithm or (trained) (machine learning) model should be obtained for performing or executing the training container inspection task.

[0139] Preferably, the set of extracted container component features is a set of container component features that are (automatically) extracted within the framework of the machine learning method performed on the training container inspection task.

[0140] The fact that a container inspection task can be trained using a method different from the container inspection task itself has the advantage that a single training process or machine learning method (i.e., a training process or machine learning process performed within the framework of training the container inspection task) is sufficient for feature extraction. The extracted container component features obtained here are then used for other container inspection tasks.

[0141] For example, the feature vector of the bottle extracted from the (camera) image can be used in several inspection tasks / classifications.

[0142] Two examples of inspection tasks / categories:

[0143] 1. Does the feature vector show the brown bottle?

[0144] 2. Does the feature vector show sealed bottles?

[0145] Both can be identified in the image, and therefore can also be identified in the extracted feature vector.

[0146] In the first case, the feature vectors of brown and other colored bottles are used to "train" a machine learning algorithm (such as kNN). In the second case, the feature vectors of sealed and unsealed bottles are used.

[0147] These features only need to be extracted once per image.

[0148] In a preferred method, training data is used in the training process or machine learning approach. This training data includes multiple, particularly spatially resolved, sensor data (of the container) collected by at least one sensor device. This provides the advantage that the training process has been specifically coordinated for the container inspection apparatus to be configured, thereby allowing for direct consideration, for example, of the specific conditions of that particular container inspection apparatus, such as the optical characteristics of the sensor devices or specific lighting conditions within the container inspection apparatus.

[0149] Preferably, sensor data, particularly spatially resolved sensor data, is provided / configured for use as training data and (acquired by at least one sensor device) has (container) type characteristics and / or classification characteristics (depending on the classification task, such as defect type classification). Preferably, the sensor data, particularly the spatially resolved sensor data, is stored together with the (container) type characteristics and / or classification characteristics to which they are respectively assigned as a training dataset (particularly on external and / or non-volatile storage devices). Preferably, multiple training datasets are generated in this manner. The classification characteristics can here be categories for training container inspection tasks (or the result categories of the aforementioned container inspection tasks). That is, the sensor data, particularly the spatially resolved sensor data, assigned to container components can be classified according to the types of defects, etc., present therein.

[0150] Training the network requires a certain amount of labeled and / or classified sensor data (e.g., images) for each application, ranging from 1,000 to 100,000 (e.g., 10,000 labeled and / or classified images per application). This labeling and / or classification work can be performed locally or centrally by image processing experts.

[0151] Preferably, (additionally or alternatively) as training data, (preferably only) sensor data (or derived data thereof) of the container components, particularly spatially resolved sensor data, collected by the sensor devices of (at least) another, preferably structurally identical, (preferably from the same manufacturer) container inspection apparatus. This provides the advantage that a large amount of sensor data can be provided and used before the container inspection system is put into formal operation.

[0152] Alternatively, one could consider using locally resolved sensor data (or data derived from it) that is (fully or partially) synthetically generated or augmented as training data. This offers the advantage that, for example, rare defect type categories can be modeled, and machine learning models can be trained in an efficient manner.

[0153] Preferably, the training or machine learning method is performed using supervised learning. However, it is also possible to train the machine learning method using unsupervised learning, reinforcement learning, or random learning.

[0154] Preferably, the extracted container component features are provided and / or transmitted to a container inspection device or a real-time evaluation device of the container inspection device (as described above).

[0155] Preferably, the method is executed on an external server associated with the container processing system, and / or the set of extracted container component features is stored on an external storage device associated with the container processing system. Preferably, the stored set of extracted container component features can be retrieved (upon approval) by an operator of the container processing system.

[0156] In one advantageous method, the number of container component features to be extracted is determined by a predetermined set of container component features.

[0157] The training process or machine learning method can be performed locally (on the container inspection device) and / or centrally and / or location-independently and / or on an external server associated with the container inspection device and / or container processing system.

[0158] In one advantageous approach, machine learning methods and / or feature extraction are performed spatially separate from the container processing system, particularly outside the site of the container processing system.

[0159] Preferably, the set of extracted container component features obtained by the proposed method is used in the method for operating the container handling system as proposed above according to a preferred embodiment.

[0160] The present invention also relates to a container inspection device for processing a container handling system for containers and, preferably, for plastic containers and / or bottles. Here, the container inspection device is adapted, intended, and / or configured to perform container inspection tasks within the container handling system.

[0161] Preferably, the container handling system has a transport device adapted and designed to transport multiple container components as a flow of container components along a predetermined transport path from at least one handling device of the container handling system to at least another handling device of the container handling system.

[0162] Here, the container inspection device has at least one sensor device that is suitable for and designed to acquire, preferably optically, sensor data, particularly spatially resolved sensor data, and preferably camera images, related to container components in order to perform the container inspection task.

[0163] According to the invention, the container inspection apparatus is adapted and intended to determine storage parameters with respect to acquired, particularly spatially resolved, sensor data, which characterize storage instructions for storing the acquired, particularly spatially resolved, sensor data in a preferably non-volatile storage device. Here, "non-volatile" can be understood as described above.

[0164] Here, storage parameters are determined based on similarity parameters, which characterize the similarity between the acquired, particularly spatially resolved, sensor data and predetermined and / or pre-predictable reference data.

[0165] That is, within the framework of the present invention, it is also proposed to store sensor data in a storage device based on sensor data similarity. In particular, the similarity with reference sensor data can also be determined here by means of an AI-based similarity metric (based on the extracted set of container component features) as described above. Preferably, only similar sensor data is retained and stored.

[0166] Preferably, the container inspection apparatus is adapted, suited, and / or designed to perform, individually or in combination, all the method steps described above in the context of sensor data acquisition and / or the determination and / or similarity parameters of reference data and / or stored parameters. Conversely, the method can be configured, individually or in combination, with all the features described within the framework of the container inspection apparatus.

[0167] The present invention also relates to a container processing system for processing multiple container components for a container, the container processing system including a container inspection device according to the above (preferred) embodiment, and including a processing device, at least one other processing device, and a transport device for transporting the container components from the processing device to the at least one other processing device.

[0168] Preferably, the container handling system is adapted, suited, and / or designed to perform the methods described above for operating the container handling system, and to perform, individually or in combination, all the method steps already described in the context of the methods above. Furthermore, the container handling system and / or the handling equipment and / or the at least one other handling equipment and / or transport equipment may have or be equipped with at least one of the aforementioned features, individually or in combination with other features.

[0169] The present invention also relates to a computer program or computer program product, including program means, particularly program code, which represents or encodes at least some and preferably all of the method steps of the method according to the invention, and preferably represents or encodes an embodiment of the preferred embodiments described, and is designed for execution by a processor device.

[0170] The present invention also relates to a data storage device that stores at least one embodiment or a preferred embodiment of a computer program according to the present invention.

[0171] This invention has been described with respect to containers or container components for containers. Here, the invention can also be transferred to injection-molded parts (injection preforms) or, more generally, to articles to be processed in a handling system (e.g., contact lenses to be manufactured and / or packaged), whose processing progress and / or characteristics and / or (defects / quality) status are controlled by at least one sensor device (for acquiring, in particular, spatially resolved sensor data, associated with each individual article to be inspected). The applicant reserves the right to claim rights to the subject matter involved.

[0172] The present invention also relates to a method of operating an article handling system to handle a plurality of injection preforms and / or articles, wherein a transport device transports the plurality of injection preforms and / or articles as part flows along a predetermined transport path from at least one handling device of the article handling system to at least another handling device of the article handling system.

[0173] Here, at least one sensor device acquires, preferably optically, sensor data, particularly spatially resolved, related to the preform and / or the article, and preferably camera images, in order to perform the article inspection task.

[0174] According to the present invention, storage parameters are determined with respect to the acquired, particularly spatially resolved sensor data, which characterize storage instructions for storing the acquired, particularly spatially resolved sensor data in a non-volatile storage device, wherein the storage parameters are determined based on a similarity parameter, which characterizes the similarity between the acquired, particularly spatially resolved sensor data and predetermined and / or predeterminable reference data.

[0175] The present invention also relates to an article inspection device for an article handling system that processes a plurality of injection preforms and / or articles to perform an article inspection task in the article handling system, wherein the article handling system has a transport device adapted and intended to transport the plurality of injection preforms and / or articles as part flows along a predetermined transport path from at least one handling device of the article handling system to at least another handling device of the article handling system.

[0176] Here, the item inspection device has at least one sensor device that is suitable for and designed to acquire, preferably optically, sensor data, particularly spatially resolved, and preferably camera images related to the preform and / or the item in order to perform the item inspection task.

[0177] According to the invention, an article inspection device is adapted and intended to determine storage parameters with respect to acquired, particularly spatially resolved sensor data, which characterize storage instructions for storing the acquired, particularly spatially resolved sensor data in a non-volatile storage device, wherein the storage parameters are determined based on a similarity parameter that characterizes the similarity between the acquired, particularly spatially resolved sensor data and predetermined and / or predeterminable reference data.

[0178] The other features described above within the framework of the container component are applied in a similar manner to the injection preform and / or article.

[0179] For further advantages and implementation methods, please refer to the accompanying drawings:

[0180] in:

[0181] Figure 1 A schematic diagram of a container handling system according to a preferred embodiment of the invention is shown; and

[0182] Figure 2 A camera image illustrating the method according to the invention is shown according to a preferred embodiment.

[0183] Figure 1 A schematic diagram of a container processing system 1 according to the invention for processing a container component 10 is shown, according to a first embodiment. The container component in this figure is a container 10 designed as a bottle.

[0184] Reference numeral 12 indicates the equipment arranged at container component 10 (the container in this figure). Figure 1 In the illustrated embodiment, an identification device is shown as an example of equipment, which is arranged on the bottle 10. For example, this is a (printed) QR code. Reference numeral 14 marks the container seal, which is another component of the container 10.

[0185] exist Figure 1 In the embodiment shown, a plastic preform is provided and fed by a transport device 6 into a heating device 20, where it is heated, and then expanded into a (plastic) bottle 10 in a blow molding apparatus as further processing equipment. The arrangement of the blow molding apparatus within the container processing system 1 is indicated by reference numeral 23. The bottle 10 may be equipped with an identification device 12, for example, by a personalization device (e.g., a printing device), thereby producing a bottle with the identification device 12.

[0186] The container component 9 can be transported from one processing unit to the next processing unit within the container handling system 1 by at least one transport device 6, and can also be transported within the processing unit. In this figure, the processing units shown are an inspection device 21, a filling device 22 for filling the bottles 10 with the product, a sealing device 24, a drying device 28, a labeling device 30, and a packaging device 32 for packaging the bottles 10 (in downstream order of the bottle transport direction).

[0187] Reference numeral 2 in each case indicates another container inspection device (e.g., arranged at the end of the production line and between the sealing device 24 and the drying device 28), which, for example, checks the filling height in the bottle and / or the correct arrangement of the seal on the bottle 10 and / or the safety ring and / or the correct labeling and / or packaging or other production data of the bottle 10.

[0188] In each case, reference numeral 4 indicates the sensor device, which in this figure is a camera, by means of which sensor data associated with each (inspected) container component 9 is collected, acquired, or recorded individually.

[0189] In each case, reference numeral 3 indicates a real-time evaluation device, by means of which sensor data collected by the sensor device 4 of the corresponding container inspection device 2 is evaluated to perform (predetermined and / or specified) container inspection tasks.

[0190] In a preferred proposed method, the detected sensor data is evaluated using a set of extracted features. The set of extracted features used is the result of (trained) feature extraction by a neural network pre-trained on a similar (inspection) task of image classification using extensive (training) data. However, in the final step of evaluating the extracted features, a classical classification method is also applied.

[0191] Reference numeral 50 denotes an internal server or internal storage device, and reference numeral 52 denotes an external server or external storage device, particularly a cloud-based storage device. For example, AI-based feature extraction can be performed on external server 52. Preferably, the set of extracted container component features can be stored on external and / or internal storage devices 50 / 52.

[0192] Reference numeral 5 in the attached figure indicates a storage device, which is a (fixed) component of the container inspection device 2 in this figure. Sensor data acquired by the sensor device 4 can be stored on the storage device 2.

[0193] Preferably, a retention strategy / image storage function is provided, which can retain images on storage device 5 and / or in camera 4 and / or in fixed memory due to image similarity. For reference, images of the errors or features being sought can be used, for example. Similarity to reference images can be determined using a preferred AI-based similarity metric. Preferably, only similar images are retained and stored.

[0194] Preferably, the container inspection device 2 can determine a similarity parameter that characterizes the similarity between the acquired image or sensor data and predetermined and / or predeterminable reference data (e.g., reference images). Depending on the determined similarity parameter, a storage parameter (e.g., via the container inspection device) can be determined that characterizes whether the acquired sensor data should be stored on the storage device 5.

[0195] Figure 2 Twelve camera images representing the method according to the invention are shown according to a preferred embodiment.

[0196] In particular, these (along with other camera images not shown) have been used to evaluate similarity. Figure 2 The results of sorting camera images based on their similarity (in descending order of similarity) are shown.

[0197] These camera images are captured within a frame that inspects the bottom of the container through its opening. The bottom is illuminated by a lighting device using transmitted light.

[0198] exist Figure 2 The first image marked with the reference numeral RSD in the upper left corner of the attached plane is used here as a reference image. Therefore, the distance between this image and itself, determined by means of a distance metric such as Euclidean distance, is 0.

[0199] Figure 2 The other camera images shown are sorted according to their respective distances from the reference image as determined by a distance metric (from left to right, then from top to bottom), thus exhibiting increasing distances, i.e. decreasing similarity.

[0200] The last camera image located in the lower right corner of the attached plane, and... Figure 2 The image shown has the largest spacing of 0.1848 compared to the other images, and is therefore the eleventh neighboring image of the reference image.

[0201] These camera images clearly demonstrate the high performance of the proposed method. The rough sample image RSD shows the bottom of a container with an embossed "BA". The camera images most similar to this one, determined by the proposed method—the first and second neighboring images ("Neighbour 1") and the second neighboring image ("Neighbour 2")—also show the same embossed "BA" (with decreasing sharpness). The third neighboring image ("Neighbour 3") shows a droplet in the center, its shape resembling the inner contour of the "B" in a circular pattern.

[0202] Figure 2 The proposed method, using a distance metric in a feature space (stretched by features extracted in an AI-based training method) to evaluate similarity, where each image is represented by a feature vector, allows all container bottoms with "BA" embossing to be sorted among their four nearest neighbor images. If the ten most similar container bottom camera images are always stored in the storage device, then the three container bottoms with "BA" embossing in the container stream will be stored in the storage device and available for operator retrieval.

[0203] The applicant reserves the right to claim rights to all features essential to the invention disclosed in the application, whether individually or in combination, provided that such features are novel compared to the prior art. Furthermore, it should be noted that features that may be advantageous in themselves are also described in individual figures. Those skilled in the art will readily recognize that a particular feature described in a figure may be advantageous even without employing other features in that figure. Additionally, those skilled in the art will recognize that advantages can also be derived by combining several features shown in a single figure or different figures.

[0204] List of reference numerals

[0205] 1 Container handling system

[0206] 2.21 Container Inspection Device

[0207] 3. Real-time evaluation equipment

[0208] 4. Sensor devices

[0209] 5. Storage devices

[0210] 6. Transportation equipment

[0211] 10 containers

[0212] 9 Container Components

[0213] 12. Equipment, Direct Printed Components

[0214] 14. Equipment, container sealing section

[0215] 20. Processing equipment, this section contains heating equipment.

[0216] 23. Processing equipment, this is the printing equipment.

[0217] 22. Processing equipment, this is the filling equipment.

[0218] 24. Processing equipment, this is the capping equipment.

[0219] 28. Processing equipment, this is the drying equipment.

[0220] 30. Processing equipment, this is the labeling equipment.

[0221] 32. Processing equipment, this section contains packaging equipment.

[0222] 50 Internal servers and storage devices

[0223] 52 External servers and storage devices

[0224] RSD reference sensor data

Claims

1. A method of operating a container handling system (1) to handle a plurality of container components (9) for a container (10), wherein a transport device (6) transports the plurality of container components (9) as a container component flow along a predetermined transport path from at least one processing device (20, 23, 22, 24, 28, 30) of the container handling system (1) to at least another processing device (23, 22, 24, 28, 30, 32) of the container handling system (1), wherein at least one sensor device (4) acquires sensor data associated with the container components (9) to perform a container inspection task. Its features are, Storage parameters are determined for the collected sensor data, the storage parameters representing storage instructions for storing the collected sensor data on a non-volatile storage device (5), wherein the storage parameters are determined based on similarity parameters, the similarity parameters representing the similarity between the collected sensor data and predetermined and / or predeterminable reference data.

2. The method according to claim 1, wherein the reference data is reference sensor data collected by the sensor device.

3. The method according to claim 1, characterized in that, The similarity parameter characterizes the similarity between the collected sensor data and a plurality of predetermined and / or pre-determinable reference sensor data.

4. The method according to claim 3, characterized in that, The reference data is pre-defined by the operator of the container handling system (1).

5. The method according to claim 1, characterized in that, The rejection parameter is determined based on the collected sensor data. The rejection parameter characterizes the rejection instruction for removing the associated container component (9) from the container component stream, wherein the similarity parameter is determined independently of the rejection decision made.

6. The method according to claim 1, characterized in that, As reference sensor data, sensor data collected with respect to at least one rejected and / or to be rejected container component (9) are used.

7. The method according to claim 1, characterized in that, If the relatively high similarity between the collected sensor data and predetermined and / or predeterminable reference data is determined and / or the relatively high similarity is determined, then the storage parameters characterizing the positive storage instruction are determined.

8. The method according to claim 3, characterized in that, Multiple reference sensor data are predetermined, and if the relatively low similarity between the acquired sensor data and the predetermined multiple reference sensor data is determined and / or the relatively low similarity is determined, then storage parameters characterizing the positive storage instruction are determined.

9. The method according to claim 3, characterized in that, The predetermined plurality of reference sensor data includes both reference sensor data related to container components (9) to be removed from the container component stream and reference sensor data related to container components (9) not to be removed from the container component stream.

10. The method according to claim 1, characterized in that, A set of container component features is defined, and the similarity parameter is determined based on the set of container component features.

11. The method according to claim 10, characterized in that, The set of container component features is a set of extracted container component features automatically obtained within the framework of the machine learning method performed on the training container inspection task.

12. The method according to claim 11, characterized in that, The set of container component features is a set of container component features extracted within the framework of a supervised learning method, wherein the supervised learning method is preferably the K-nearest neighbor algorithm.

13. The method according to claim 12, characterized in that, The feature space is stretched by the provided set of extracted container component features and / or the feature space is stretched by the provided set of extracted container component features, and a distance metric associated with the feature space has been provided and / or will be provided, wherein the distance metric is used as a similarity metric to evaluate the similarity between the acquired sensor data and the reference data to determine the similarity parameter.

14. The method according to claim 1, characterized in that, The container inspection task is a classification task, which is selected from a set of classification tasks, including: classifying container components as defective or non-defective; identifying and / or classifying the defect types of the container components; identifying and / or classifying different types of the container components; identifying and / or classifying the outline and / or color of the container components; identifying and / or classifying the error-free / error-prone execution of at least one processing step performed on the inspected container component; and combinations thereof.

15. A container inspection apparatus (2) for processing a plurality of container components (9) for a container (10) in a container processing system (1) to perform a container inspection task in the container processing system (1), wherein the container processing system (1) has a transport device (6) adapted and intended to transport the plurality of container components (9) as a container component flow along a predetermined transport path from at least one processing device (20, 23, 22, 24, 28, 30) of the container processing system (1) to at least another processing device (23, 22, 24, 28, 30, 32) of the container processing system (1), wherein the container inspection apparatus (2) has at least one sensor device (4) adapted and intended to acquire sensor data associated with the container components (9) for performing the container inspection task. Its features are, The container inspection device (2) is adapted and intended to determine storage parameters with respect to the acquired sensor data, the storage parameters being characterized by storage instructions for storing the acquired sensor data on a non-volatile storage device (5), wherein the storage parameters are determined based on a similarity parameter, the similarity parameter being characterized by the similarity between the acquired sensor data and predetermined and / or predeterminable reference data.

16. A container processing system (1) for processing a plurality of container components (9) for a container (10), the container processing system comprising a container inspection device (2) according to the preceding claims, and comprising a processing device, at least one other processing device, and a transport device for transporting the container components from the processing device to the at least one other processing device.

Images