Packaging machine for packaging products in packaging

Abstract

Packaging machine for packaging products in packaging, the packaging machine comprising a transfer device for transferring a product to a package, a detection device for detecting indicative data for a relative position of the inserted product and the package, a correction device for changing a relative position of an inserted product and the package, and a control unit, wherein the control unit is configured to control the correction device based on the indicative data to change the relative position of the inserted product and the package.

Description

[0001] The present invention relates to a packaging machine for packaging products in packaging according to claim 1 and to a method for changing a relative position of a product placed in packaging and the packaging in a packaging machine according to claim 9. State of the art

[0002] Packaging machines for packing products into packaging are well known from the prior art. These typically include a transfer device, such as a loading station, for placing the products into packaging such as packaging trays. The products can be, for example, one or more slices of sausage or cheese, or larger pieces of product such as cheese cubes can also be placed into packaging.

[0003] The packaging containing the product is then usually sealed by placing a sealing film onto the packaging and sealing it at a predefined sealing seam, creating a gas-tight connection between the packaging and the sealing film.

[0004] However, misalignments can occur between the product and the packaging during insertion. This might be because the packaging's position doesn't exactly match the intended position, or the product itself isn't inserted in the correct position. Particularly with disc-shaped products, this can result in one or more slices protruding from the packaging, obscuring the intended seal. Consequently, the sealing process can be incomplete and therefore of insufficient quality, leading to product rejects.

[0005] To minimize rejects, it is common practice to manually monitor the correct placement of products into packaging and, if necessary, to manually align the products within the packaging. However, this approach is prone to errors, as operators may inadvertently overlook or misjudge incorrect product positioning. While this method can reduce rejects, a certain error rate remains, which is particularly dependent on production speed and the operator's attention. Task

[0006] Based on the known state of the art, the technical problem to be solved is therefore to specify a packaging machine for packaging products in packaging and a method for changing a relative position of a product placed in a packaging and the packaging in a packaging machine, which achieve a reduced reject rate. Solution

[0007] This problem is solved according to the invention by the packaging machine for packaging products in packaging according to claim 1 or by the method for changing the relative position of a product placed in packaging and the packaging in a packaging machine according to claim 9. Advantageous embodiments of the invention are described in the dependent claims.

[0008] The packaging machine according to the invention for packaging products in packaging comprises a transfer device for transferring a product to a package, a detection device for detecting indicative data for a relative position of the inserted product and the package, a correction device for changing a relative position of an inserted product and the package, and a control unit, wherein the control unit is configured to control the correction device based on the indicative data to change the relative position of the inserted product and the package.

[0009] The indicative data can be, in particular, image data or positional data of the packaging and / or the product inside, or can include such data, such that identification of the product's position relative to the packaging's position is preferably possible based on the indicative data (especially preferably solely based on the indicative data). Preferably, the three-dimensional position of the product in space relative to the position of the packaging in space can be detected by the recognition device, or the indicative data can be indicative for this position.

[0010] According to the invention, the correction device is designed such that it can move at least the product or the packaging or both, for example by lifting the product and moving the packaging, or by changing the position of the product relative to the packaging during lifting.

[0011] The control unit can be designed as a computer with a processor and assigned memory and have suitable programming to use the indicative data to generate control commands for the correction unit, and thus change the relative position of the inserted product and the packaging.

[0012] In the following, packaging is understood to mean a packaging tray or at least any type of unsealed or unsealed packaging into which a product (such as one or more slices of sausage or cheese, or chunky products like cheese cubes) can be placed.

[0013] This packaging machine allows for reliable post-packaging adjustments to the product's position, further minimizing rejects due to subsequent improper sealing or closing. Simultaneously, it reduces the potential for errors, as manual operator intervention is no longer required.

[0014] It may be provided that the control unit is trained to control the correction unit based on the indicative data and a target relative position of the inserted product and packaging.

[0015] The target relative position can specify either the product's position relative to the packaging in a plane, such as the transport plane or a parallel plane of packaging, or it can specify a three-dimensional relative arrangement of the product relative to the packaging. By comparing the target relative position with the indicative data, it can then be determined whether and to what extent the product's position relative to the packaging needs to be changed, and the correction unit can be controlled accordingly. This can be achieved using known deterministic algorithms that do not employ a conventional network, thus reducing scrap with less programming effort.

[0016] It may be provided that the control unit is designed to determine a necessary change in position of an inserted product and to control the correction device based on the necessary change in position.

[0017] A necessary position change is understood to mean the change in the position of the inserted product that moves it from its current position to a predetermined position, such as the target relative position of the inserted product to the packaging, so that the packaging can be reliably closed in a subsequent sealing or closing step. To determine the necessary position changes, one can consider, for example, the minimum required position change to achieve reliable sealing or closing of the packaging. It is not essential that this necessary position change moves the product into a perfectly ideal position, but only that its position is changed to such an extent that reliable sealing becomes possible. This embodiment also further reduces rejects.

[0018] In one embodiment, the control unit comprises a neural network trained to determine the necessary change in position.

[0019] The neural network can be specifically designed to recognize patterns in images or corresponding indicative data that specify the relative position of objects. The neural network can be trained to determine the necessary position change by processing the indicative data, for example, by comparing the indicative data with known, correctly positioned products in packaging. This can be done by creating a difference image between the recognized product and the correctly positioned products in packaging known to the neural network, and then processing this image. The necessary position change can then be determined as output through this processing within the neural network.Since neural networks are trained on a large amount of training data for pattern recognition, this can further increase the accuracy in determining the necessary positional adjustment and also the accuracy in detecting incorrectly positioned products, thus reducing waste even further. The use of a neural network is particularly advantageous for deformable products, such as sausage slices, because an incorrectly positioned product is often bent or deformed, making it more difficult to determine the necessary positional adjustment. This can be taken into account during the neural network's training, so that such effects of incorrect product positioning relative to the packaging are also considered when determining the necessary positional adjustment.

[0020] The detection device can include a camera or a laser scanner.

[0021] The camera can be a two-dimensional camera or a 3D camera designed to capture two-dimensional or three-dimensional images, respectively. Capturing two-dimensional images generates less data to process, whereas capturing three-dimensional images can be advantageous when determining the product's position relative to its packaging in space. A laser scanner is particularly well-suited for determining data on the three-dimensional, relative arrangement of the product and its packaging. With these configurations of the detection device, reliable detection of a product's actual position is achieved, allowing for even more precise subsequent adjustments to the position, thus further reducing waste.

[0022] In one embodiment, the correction device comprises a multi-axis robot. The multi-axis robot can, in particular, be a collaborative robot and / or a six-axis collaborative robot, or be designed as a linear unit with an adjustable gripper. With this embodiment, precise and reliable relative movement of the product to the packaging is possible, so that incorrectly positioned products can be repositioned very accurately, thereby further reducing rejects.

[0023] The packaging machine may include a transport device for moving packages containing the product along a transport direction, with the transport device running at least partially alongside the correction device. The transport of the packages within the transport device can be continuous, allowing for changes in the relative position of the product and the packaging during operation, thereby increasing the machine's throughput while minimizing rejects.

[0024] The packaging machine can include a traversing device for moving the correction device relative to the transport device along at least one dimension.

[0025] It can be provided that the correction device can be moved by the traversing device along the transport device, in particular parallel to the transport direction, at a speed equal to the transport speed of the packages containing the product. With this embodiment, the relative movement between the correction device and the package containing the product is reduced, which enables highly precise changes to the product's position, thereby further reducing rejects.

[0026] According to the invention, a method for changing the relative position of a product placed in packaging and the packaging in a packaging machine is further provided, wherein the packaging machine comprises a transfer device for transferring a product to packaging, a detection device for detecting data indicative of a relative position of the product placed in packaging and packaging, a correction device for changing the relative position of the product placed in packaging and packaging, and a control unit, wherein the control unit controls the correction device based on the indicative data to change the relative position of the product placed in packaging and packaging.

[0027] This method can reduce rejects due to incorrectly positioned products.

[0028] It can be provided that the control unit controls the correction unit based on indicative data and a target relative position of the inserted product and the packaging. With this embodiment, particularly dimensionally stable products can be reliably repositioned.

[0029] The control unit can determine the necessary positional change of the inserted product and, based on this change, control the correction device. By determining the necessary positional changes, which can define a two-dimensional or three-dimensional translation of the product, reliable repositioning and thus a reduction in scrap are possible.

[0030] The control unit can include a neural network trained to determine the necessary position change, with the neural network then determining this change. Particularly with shape-changing products, neural networks can detect incorrect positioning of the product relative to its packaging and reliably determine the necessary position change, thereby further reducing waste.

[0031] The detection device may include a camera or a laser scanner and / or the correction device may include a multi-axis robot.

[0032] These devices enable precise determination of the product's position relative to the packaging and precise adjustment of the product's position relative to the packaging, thereby further reducing waste.

[0033] The packages containing the product can be transported along a single transport direction by a transport device of the packaging machine, and the transport device can run at least partially alongside the correction device. The transport device can be operated continuously, which increases the throughput of the packaging machine while simultaneously reducing waste.

[0034] The packaging machine can include a traversing device for moving the correction device relative to the transport device along at least one dimension, and the method can include a method or correction device dependent on the position of the packaging in the transport device. This allows the correction device to move along with the packaging and products, thus reducing the relative movement between the correction device and the product, which can improve the result of repositioning a product.

[0035] All embodiments described here can be combined with each other. Brief description of the characters Fig. Figure 1 shows a packaging machine according to one embodiment. Fig. Figure 2 schematically shows the change in the position of the product relative to the packaging. Fig. Figure 3 shows a flowchart of a training process for a neural network to determine the necessary position change. Detailed description

[0036] Fig. Figure 1 shows a schematic view of a packaging machine 100 for packaging products 131 in packagings 130.

[0037] The packaging machine can be designed in the usual way and, in particular, include a transport device 105 for transporting the packaging 130 and a transfer device 107 for transferring products 131 to the packaging 130. The designs of the transport device 105 and the transfer device 107 are, in principle, arbitrary.

[0038] The transport device 105 may be designed as a conveyor belt (such as an endless rubber conveyor belt or a conveyor chain) for the transport of packages 130 in at least one row and optionally in multiple rows. Package 130 is understood below to mean a packaging tray or at least any type of unsealed packaging into which a product 131 (such as one or more slices of sausage or cheese, or chunky products like cheese cubes) can be inserted.

[0039] The transfer device 107 can, for example, comprise a conveyor belt or other conveying device for the product that is substantially inclined in the direction of transport T of the transport device. This conveyor belt can include one or more carriers and / or retaining elements (not shown here) so that the product 131 does not inadvertently slide down towards the transport device 105 due to the inclined plane. Alternatively or additionally, it can be provided that the coefficient of static friction between the product 131 and the contact surface of the transfer device with the product is chosen to be as high as possible, so that a transition to sliding of the product on the transfer device is avoided as far as possible.

[0040] Downstream of the transfer device 107 in the transport direction T of the packaging, a sealing station 104 can be arranged, which can apply a sealing film 141 to a product-filled packaging 130 and bond it to the packaging 130, for example by applying heat, so that the packaging is sealed. The design of such a sealing station is generally known to those skilled in the art.

[0041] To detect any misalignment of the product relative to the packaging, as shown schematically here with packaging 132, in which the product lies at an angle within the packaging and protrudes from it, the packaging machine 100 is provided, according to the invention, to include a detection device 101 downstream of the transfer device 107. The detection device 101 is designed to detect or generate indicative data for the relative position of the inserted product and the packaging. This indicative data can, for example, be image data. Thus, the detection device 101 can, for instance, include a camera oriented towards the packaging transported in the transport device, which can capture images of the packaging and the product inserted therein.The images can represent the indicative data, or the indicative data can be obtained through further image processing (e.g., contrast adjustment or scaling). The camera can, for example, be designed to capture two-dimensional images, producing a two-dimensional image of the packaging and the products inside from above. Alternatively, the recognition device can include a 3D camera, preferably also positioned above the transport device, which can generate a three-dimensional image of the packaging and the product inside. This three-dimensional image can then represent the indicative data.

[0042] Alternatively, the recognition device 101 can comprise a laser scanner or a device designed to scan the three-dimensional structure of the packaging and the product inside. This generates indicative data in the form of data representing the contour of the packaging and the product inside. Since determining the correct positioning between the product and the packaging does not depend on other information, such as the color of the packaging and the product, this embodiment can be particularly preferred because the processing of the corresponding data, for example using a neural network, can be especially reliable.

[0043] According to the invention, the packaging machine 100 further comprises a control unit 180, which can, for example, be designed as the central control unit of the packaging machine and can, in principle, be configured as a computer with a processor and associated memory. The control unit is further configured to receive the indicative data from the recognition device.

[0044] Furthermore, the packaging machine includes a correction device 102. The correction device 102 is designed to effect a change in the relative position / arrangement between a product 131 placed in the packaging and the packaging 130. For this purpose, the correction device 102 can, for example, include a gripping element, a suction cup, or a similar device with which the product can be grasped, lifted, and moved at least in a plane parallel to the transport device. Alternatively or additionally, it can also be provided that the correction device 102 allows the placed product to be moved in space.While the movement of the product is generally described here, it can also be additionally or alternatively provided that the correction device 102 lifts the product and can move the packaging relative to the product, or that the product and packaging can be moved relative to each other by the correction device.

[0045] The design of the correction device is not fundamentally restricted, provided it can effect a change in the relative position between the packaging and the product. However, preferred embodiments of the correction device include at least one collaborative and / or multi-axis robot, enabling a highly accurate and stepless change in the product's position relative to the packaging. The robot can be a six-axis robot, in particular. The use of collaborative robots, or so-called co-bots, can also be advantageous, as these can operate with exceptional efficiency, fully autonomously, and automatically.

[0046] According to the invention, the control unit 180 is further configured to control the correction device 102 based on indicative data such that the position of an inserted product, such as a mispositioned product, is changed within the packaging before the packaging containing the product is fed to the sealing station. In particular, the control unit may include a deterministic algorithm or a trained neural network to determine the necessary change in the product's position from its actual position within the packaging, which can be determined from the indicative data, to a target relative position that the product should assume before the packaging is sealed. The target relative position may, in particular, be a target relative position between the packaging and the product.This target relative position can either be a single position or encompass a range that specifies a permissible positioning of a product within the packaging, particularly within the packaging cavity. This range could, for example, be the entire packaging cavity, as long as it is ensured that the inserted product does not protrude beyond the edge of the packaging cavity, as this could cause problems during subsequent sealing.

[0047] The control unit 180 is preferably configured to determine the necessary position change using the algorithm or neural network 181 and can derive control commands for the correction device from this in order to control it based on the control commands, so that the correction device 102 changes the position of the product relative to the packaging.

[0048] Since a change in the position of a product within its packaging requires a certain amount of time, one embodiment provides for the operation of the transport device to occur in cycles. This allows packages containing products to come to a standstill relative to the correction device, enabling the device to make any necessary adjustments to the position of one or more products within the packages at each cycle. Because there is no relative movement between the correction device as a whole and the products, simpler control of the correction device is possible.

[0049] Alternatively, the correction device can be moved along a direction by means of a traversing device 106, the direction preferably comprising at least one direction parallel or substantially parallel to the transport direction T. It is preferred if the traversing device 106 can move the correction device 102 parallel to the transport direction T and at the same speed. This also reduces or completely compensates for the relative movement between the packaging and the inserted product and the correction device 102. The traversing device can, in particular, comprise one or more rails on which the correction device can be movably arranged. In one embodiment, the traversing device can extend alongside the transport device or, alternatively, be arranged above the transport device.

[0050] Alternatively or additionally, the traversing device 106 can also be designed to move the correction device in a direction transverse to the transport direction T. This allows either multiple lines (each with its own transport devices) of different packaging machines to be served, or, for example, within a single packaging machine, the correction device to reliably reach packages transported in multiple rows without requiring the correction device itself to have long lever arms. This improves the stability of the correction device and / or the accuracy with which it can change the position of packages relative to the product inside.

[0051] It may be provided that the route along which the correction device can make changes to products placed in packages is selected in such a way that a change in the position of products placed in several successive packages or in several packages transported in the same row is also possible before they reach the sealing stations 104.For example, if the average time to change the position of a product within a package is one second, and it is intended that up to five packages transported in a row transverse to the transport direction are processed by the correction device, it may be provided that the length of the transport path reachable by the correction device and / or the transport speed of the packages in the transport device are selected such that the correction device has at least six seconds available to reposition products in the packages.This ensures that even if, for example, a temporary malfunction in the transfer device or other equipment of the packaging machine 100 results in a significant number of incorrectly positioned products in packaging, this deviation can be reliably reduced, thus reducing the reject rate.

[0052] In general, it can be provided that an available interaction length of the correction device with packagings within the transport device is selected such that it is at least equal to the product of the transport speed of the packagings in the transport device, the average time required to change the position of a product, and the maximum number of products to be repositioned in a given group of packagings, or is greater than this product.

[0053] Fig. Figure 2 shows a top view of a package 230 and a product 240 placed in a packaging recess 231 of the package 230, as well as a process of correcting the position of the product 240 relative to the package 230.

[0054] In the left-hand representation of the Fig. Figure 2 shows that product 240 is not positioned within the packaging recess 231, but rather protrudes partially from it. The packaging recess is understood here as a depression extending below the rim 233 of the packaging 230, such that product 240 rests partially on the rim 233 and partially extends into the packaging recess 231. Product 240 is, in the case of... Fig. The embodiment shown in Figure 2 is a flexible product, such as a slice of cheese or a slice of sausage. However, this is not to be understood as a limitation. The contact at different height levels, on the one hand on the edge 233 and on the other hand in the packaging recess 231, causes a deformation of the product 240, so that the shape of the product 240 deviates from its circular shape, which is only assumed here as an example.

[0055] In the left-hand representation of the Fig. 2. The product 240 rests so far on the edge 233 that it at least partially covers a designated sealing seam 232. The designated sealing seam 232 is the part of the packaging in which the Fig. The sealing station described in section 1 is intended to bond the sealing film to the packaging. If the product partially rests on this intended sealing seam 232, reliable sealing of the packaging cannot be achieved in this area, and the packaging will either not be sealed at all or will be sealed with insufficient quality, or generally closed.

[0056] With the packaging machine according to the invention, Fig. Based on the indicative data, the control unit 180 can now make a necessary position change Δr of the product 240 from its initial position in the left-hand representation of the Fig. 2. A target relative position 241, in which the product is positioned in the packaging recess 231, is determined. This necessary position change Δr can then be used to control the correction device, changing the position of the product 240 relative to the packaging 230 and thus transferring it into the packaging recess. It can be provided that the necessary position change Δr is determined based on the current position 240 of the product and exactly one predetermined position 241 of the product within the packaging recess. However, this places high demands on the change in the product's position, although even a slight deviation would not have a significant negative impact on the sealing result as long as the product is completely within the packaging recess.Therefore, an alternative approach is to determine the necessary position change Δr such that every point of the product lies within a predefined area (which can be understood as the target relative position), such as an area corresponding to the packaging recess. However, due to the potential deformation of the product, this does not necessarily equate to a translational movement in the plane depicted here. Instead, the determined, necessary position change Δr must also take into account that the shape of the product changes during the displacement, in particular that the product's dimensions in the depicted plane increase when it is completely moved into the packaging recess.

[0057] This circumstance can be taken into account by suitable deterministic algorithms, which, for example, also include a model of the physical behavior of the product under deformation, so that the necessary change in position can be determined.

[0058] Alternatively, a specially trained neural network can be used to determine the necessary position change. This network is trained to determine necessary transformations between two states. The first state here is the current position of the product relative to the packaging, as shown in the left-hand diagram. Fig. 2. The second state is a position of the product completely within the packaging recess as the objective, whereby taking into account a possible deformation of the product, the necessary change in position is to be determined or the transformation is determined by the neural network, which transfers the position of the product 240 into the objective position.

[0059] Corresponding neural networks, which can determine transformations between two states, such as between two images, taking into account certain properties like the physical properties of the objects within the image, are generally known to those skilled in the art. These neural networks can use as input the indicative data already discussed above, which, for example, shows an image of the packaging with the product inside, or corresponding point clouds, which are captured using a laser scanner and depict the three-dimensional structure of the packaging and the product inside.

[0060] Optionally, information about the product's intended position relative to the packaging can be used as input data. One advantage of using neural networks is the ability to achieve the necessary positional change even without precisely defining the product's intended position within the packaging or specifying a physical model to describe its behavior under deformation. This is accomplished by selecting the neural network's training data in such a way that the relevant information is implicitly provided to the network during training. The neural network is trained to determine the necessary transformation, implicitly taking into account any potential product deformation due to the training process.

[0061] This shows Fig. 3. A flowchart of a training of a neural network according to an embodiment to perform the necessary position change or transformation to move from the position of the product according to the left-hand illustration in Fig. 2 into the right representation of the Fig. 2 to move on to determine.

[0062] The procedure for training a suitable neural network according to the Fig. Step 3 begins with a first step, step 301, in which a training dataset is provided. This training dataset comprises a large number of indicative data, which can be derived, for example, from images of packaging and the products inside in different relative positions. This can involve several thousand, tens of thousands, or hundreds of thousands of images. Alternatively, the training dataset can be obtained in a known manner from a smaller number of indicative data, such as images, by modifying the corresponding images, for example, using well-known image processing programs, by rotating the images or changing their contrast or color. A suitable training dataset can also be specifically acquired by creating a large number of images of packaging with products inside in different positions.

[0063] The training dataset preferably, but not necessarily, comprises statistically uniformly distributed arrangements of products relative to the packaging, where some arrangements are faulty, for example, the product protrudes beyond the intended packaging recess or is otherwise not fully inserted into the packaging, making sealing impossible. Furthermore, the training dataset also includes indicative data that are indicative of a correct positioning of the product relative to the packaging.

[0064] Preferably, the training dataset consists of indicative data for exactly one packaging type and exactly one product type (for example, sliced ​​sausage), such as that found in Fig.2. Packaging tray shown with corresponding rim. Alternatively, if the use of a neural network to determine a necessary position change in connection with packaging, for example in the form of a bag, is planned, a separate training dataset can be used for this purpose.

[0065] First, the training data is processed to generate a set of reference data 302 and a validation data set 303. This can be done in the usual way. The validation data set preferably comprises a subset of the training data set, such as 100 images, or 1% or 0.1% of the total number of images or indicative data contained in the training data set, and the necessary positional changes of the product relative to the packaging associated with them, and is not used during the normal training process.

[0066] Reference dataset 302 includes a mapping of the remaining training data with the respective necessary position changes. Additionally, a target relative position of the product in the packaging can be specified for both the validation dataset and the reference dataset. However, this is not mandatory. One advantage of using the neural network is that the necessary position change can also be determined without specifying a target relative position, since the neural network is trained to calculate a necessary position change for the indicative data it receives.

[0067] In the next step, 304, the training data (without the previously extracted validation data) is processed by the neural network. All available training data, or a subset or true subset of the training data, can be processed in step 304. The neural network may have been initialized and its parameters set to specific initial values, which, based on experience, will ensure that the neural network makes at least partially accurate or nearly accurate predictions regarding the necessary position change.

[0068] By processing the training data, the neural network generates predictions regarding the necessary positional changes of the respective products based on the indicative data of the training dataset. Using images as indicative data, this can, for example, involve the neural network predicting a first necessary positional change for a first image and a second necessary positional change for a second image. These changes depend on the product's arrangement within the packaging and do not necessarily have to be identical. The training dataset may also contain indicative information indicating that the product is correctly positioned within the packaging. In such a case, the predicted necessary positional change would be zero.

[0069] After the neural network has processed the training data in steps 304 and 305 and predicted the necessary position changes, a comparison with the reference data is performed in step 306. In this step, the necessary position changes predicted by the neural network for the respective indicative data are compared with the actual necessary position changes from the reference data, and, as is standard practice in neural network training, differences or other quantities indicative of a deviation from the reference data are determined.

[0070] Subsequently, in step 307, the parameters of the neural network are updated, with the aim of modifying these parameters for the next training cycle to increase prediction accuracy. This can be done using known methods, such as gradient descent algorithms or similar techniques, and is generally familiar to those skilled in the art.

[0071] If the neural network or its parameters were updated in step 307, a validation step 308 can follow. In this step, the neural network, using its current parameters, determines predictions of the necessary position changes for the validation dataset, which was not used during training. If the comparison with the validation dataset shows that the neural network predicts the necessary position changes with sufficient accuracy, the training can be terminated, and in step 310, the neural network is a trained neural network that can then be used in the packaging machine.

[0072] If the comparison with the validation data set shows that the neural network is not yet sufficiently trained, a further training cycle can be carried out in step 309 according to steps 304 to 307 and optionally 308.

[0073] Alternatively or additionally, it can be stipulated that validation does not occur after each training cycle, i.e., after each iteration of a training dataset through the neural network. For example, it can be stipulated that if step 307 detects that the parameter values, or at least the value of one parameter, change from the first to a subsequent training cycle by a value greater than a predefined threshold, no validation step is performed, and the next training cycle begins without a validation step. This process is repeated until the parameters of the neural network no longer change, or change so little that their change falls below a threshold.Once this is achieved, the validation dataset can be used to verify the quality of the training and to rule out the possibility that the neural network was not generally optimized to predict necessary position changes, but rather specifically trained on the indicative data of the training dataset. If the latter is the case, known methods can be used to reverse this specialization of the neural network.

[0074] Once the neural network is trained, it can be used in the control unit, as described in the preceding embodiments, to determine the necessary position change based on indicative data. This eliminates the need to specify a target relative position and / or a specific physical model of the product, thus reducing the information required for production. In one embodiment, the neural network can also be further trained during normal operation based on the determined necessary position changes. This can be achieved by monitoring, particularly automatically, the sealing result, at least for those packages for which the neural network has determined a non-zero necessary position change. If the sealing result is good, or...If a result meets predefined quality criteria, this can be used to adjust the parameter values ​​of the neural network, just as negative seal results that do not meet the requirements can be used to adjust the parameters of the neural network.

Claims

[1] Packaging machine (100) for packaging products (131) in packagings (130), the packaging machine (100) comprising a transfer device (107) for transferring a product (131) to a packaging (130), a detection device (101) for detecting indicative data for a relative position of the inserted product (131) and the packaging (130), a correction device (102) for changing a relative position of an inserted product (131) and the packaging (130), and a control unit (180), wherein the control unit (180) is configured to control the correction device (102) based on the indicative data to change the relative position of the inserted product (131) and the packaging (130);wherein the packaging machine (100) comprises a transport device (105) for transporting packages (130) with product (131) inserted along a transport direction (T) and wherein the transport device (105) extends at least partially along the correction device (102), wherein the packaging machine (100) comprises a traversing device (106) for moving the correction device (102) relative to the transport device (105) along at least one dimension, wherein the traversing device (106) can move the correction device (102) along a direction parallel and / or transverse to the transport direction (T). [2] Packaging machine (100) according to claim 1, wherein the control unit (180) is configured to control the correction device (102) based on the indicative data and a target relative position (241) of the inserted product (131) and packaging (130). [3] Packaging machine (100) according to claim 1 or 2, wherein the control unit (180) is configured to determine a necessary change in position (Δr) of an inserted product (131) and to control the correction device (102) based on the necessary change in position (Δr). [4] Packaging machine (100) according to claim 3, wherein the control unit (180) comprises a neural network (181) trained to determine the necessary change in position (Δr). [5] Packaging machine (100) according to one of claims 1 to 4, wherein the detection device (101) comprises a camera or a laser scanner. [6] Packaging machine (100) according to one of claims 1 to 5, wherein the correction device (102) comprises a multi-axis robot. [7] Method for changing the relative position of a product (131) placed in a package (130) and the package (130) in a packaging machine (100), the packaging machine (100) comprising a transfer device (107) for transferring a product (131) to a package (130), a detection device (101) for detecting indicative data for a relative position of the product (131) placed in the package (130) and the package (130), a correction device (102) for changing the relative position of the product (131) placed in the package (130) and the package (130), and a control unit (180), wherein the control unit (180) controls the correction device (102) based on the indicative data to change the relative position of the product (131) placed in the package (130),wherein the packages (130) with the product (131) inside are transported along a transport direction (T) by a transport device (105) of the packaging machine (100) and wherein the transport device (105) runs at least partially along the correction device (102), wherein the packaging machine (100) comprises a traversing device (106) for traversing the correction device (102) relative to the transport device (105) along at least one dimension and wherein the method comprises a traversing of the correction device (102) depending on a position of the package (130) in the transport device (105), wherein the traversing device (106) can move the correction device (102) along a direction parallel and / or transverse to the transport direction (T). [8] Method according to claim 7, wherein the control unit (180) controls the correction device (102) based on the indicative data and a target relative position (241) of the inserted product (131) and the packaging (130). [9] Method according to claim 7 or 8, wherein the control unit (180) determines a necessary change in position (Δr) of the inserted product (131) and controls the correction device (102) based on the necessary change in position (Δr). [10] Method according to claim 9, wherein the control unit (180) comprises a neural network (181) trained to determine the necessary position change and wherein the neural network (181) determines the necessary position change (Δr). [11] Method according to any one of claims 7 to 10, wherein the detection device (101) comprises a camera or a laser scanner and / or wherein the correction device (102) comprises a multi-axis robot.

Images